2021 Ingenium: Journal of Undergraduate Research

Page 1

Undergraduate Research at the Swanson School of Engineering

University of Pittsburgh Swanson School of Engineering Undergraduate Research Benedum Hall, 3700 O’Hara Street, Pittsburgh, PA 15261 USA Spring 2021

The image on the cover shows thermal gradient G vectors at the boundary of the melt pool. (See page 70 by Tyler Paplham, Department of Mechanical Engineering and Materials Science.) Please note that neither Ingenium nor the Swanson School of Engineering retains any copyright of the original work produced in this issue. However, the Swanson School does retain the right to nonexclusive use in print and electronic formats of all papers as published in Ingenium.

Ingenium 2021

Table of Contents A Message from the Associate Dean for Research............................................................. 4 Message from the Coeditors in Chief................................................................................. 5 Graduate Student Review Board – Ingenium 2021............................................................. 6 u Representations of population activity during sensorimotor transformation for visually guided eye movements* Eve C. Ayar, Michelle R. Heusser, Neeraj J. Gandhi Department of Neuroscience, Department of Bioengineering, Center for Neuroscience (CNUP), and Center for Neural Basis of Cognition (CNBC)..................................................................... 7 u Simulating the effect of different structures and materials on OLED extraction efficiency Benjamin Bailey, Paul Leu Department of Industrial Engineering.................................................................................... 13

 Spike decontamination in local field potential signals from the primate

superior colliculus Megan Black, Clara Bourrelly, Neeraj Gandhi, Ahmed Dallal Department of Electrical and Computer Engineering, Department of Bioengineering................ 16

 Hardware acceleration of k-means clustering for satellite image compression Francis J. Fattori, Alan D. George Department of Electrical and Computer Engineering.............................................................. 20 u Calphad modeling and prediction of magnetic phase diagram for the Fe-Cr-Ni system Peiyun Feng, Liangyan Hao, Wei Xiong Physical Metallurgy & Materials Design Laboratory, Department of Mechanical Engineering & Material Science................................................................................................................. 25

 Computer vision methods for tool guidance in a finger-mounted device for the blind

Yuxuan Hu, Rene R. Canady, Roberta Klatzky, George Stetten Visualization and Image Analysis (VIA) Laboratory, Department of Bioengineering.................... 30  Immunomodulators and macrophages in endometriosis pathogenesis and progression Hilda Jafarah, Isabelle Chickanosky, Alexis Nolfi, Mangesh Kulkarni, Clint Skillen, Bryan Brown Department of Bioengineering, Carnegie Mellon University, McGowan Institute of Regenerative Medicine........................................................................................................ 35  Testing the compressive stiffness of endovascular devices Lucy Jones, Jonathan Vande Geest Soft Tissue Biomechanics Laboratory, Department of Bioengineering..................................... 40  Design and manufacture of two experimental test benches for the in vitro assessment of a method to perform in situ fenestration of stent grafts Nicholas P. Lagerman, Timothy K. Chung, Mohammad H. Eslami, David A. Vorp Department of Bioengineering, Division of Vascular Surgery, University of Pittsburgh Medical Center, Department of Surgery, Department of Cardiothoracic Surgery, and Department of Chemical and Petroleum Engineering, McGowan Institute for Regenerative Medicine, and Center for Vascular Remodeling and Regeneration.......................................................... 44 u Fluid flow simulation of microphysiological knee joint-on-a-chip Eileen N. Li, Zhong Li, Ryan M. Ronczka, Hang Lin Department of Bioengineering, Department of Orthopaedic Surgery, University of Pittsburgh School of Medicine.............................................................................................................. 49

 Association between functional dedifferentiation and amyloid in preclinical Alzheimer’s Disease Elizabeth J. Mountz, Jinghang Li, Akiko Mizuno, Ashti M Shah, Andrea Weinstein, Ann D. Cohen, William E. Klunk, Beth E. Snitz, Howard J. Aizenstein, Helmet T. Karim Department of Bioengineering, Department of Psychiatry, Department of Neurology, Physician Scientist Training Program, University of Pittsburgh School of Medicine................... 55

u Levator Ani muscle dimension changes with gestational and maternal age Oreoluwa Odeniyi, Megan Routzong, Steven Abramowitch Department of Bioengineering.............................................................................................. 64 u Development of a post-processing method for estimating solidification parameters from finite-element modeling of laser remelting in directed energy deposition of Ni-Mn-Ga magnetic shape-memory alloy single crystals Tyler Paplham, Jakub Toman, Markus Chmielus Department of Mechanical Engineering and Materials Science............................................... 68 u Demonstrating the antibiofouling property of the Clanger cicada wing with ANSYS Fluent simulations Brady C. Pilsbury, Paul W. Leu Laboratory for Advanced Materials at Pittsburgh, Department of Industrial Engineering............ 72 u Identifying potential mammalian heme/CO sensors through a bioinformatic analysis of the Per-Arnt-Sim (PAS) protein domain JonPaul Plesniak, Matthew Dent, Jason Rose, Jesus Tejero Department of Chemical Engineering, Heart, Lung, Blood, and Vascular Medicine Institute, Division of Pulmonary, Allergy and Critical Care Medicine...................................................... 76 u Topological descriptor selection for a quantitative structure-activity relationship (QSAR) model to assess PAH mutagenicity Caitlin Sexton, Trevor Sleight, Carla Ng, Leanne Gilbertson Gilbertson Lab Group, Ng Lab Group, Department of Civil and Environmental Engineering........ 81  Innovations for reimagined educational experiences in the face of a pandemic† Sophie Shapiro, Stuart McCutchen, Zeve Cohen, Elizabeth Gilman, William Clark Department of Bioengineering, Department of Electrical Engineering, Department of Mechanical Engineering, Innovation, Product Design, and Entrepreneurship Program.............. 85

 User-specific factors associated with ladder handrail use Jami Stuckey, Michelle Tsizhovkin, Kurt Beschorner, Erika Pliner, Daina Sturnieks, Stephen Lord Human Movement and Balance Laboratory, Department of Bioengineering; Falls, Balance, and Injury Research Centre, Neuroscience Research Australia......................... 90

 Relationship between center of pressure and contact region during gait: implications for shoe wear and slipping risk Joseph Sukinik, Rosh Bharthi, Sarah Hemler, Kurt Beschorner Human Movement and Balance Laboratory, Department of Bioengineering............................. 94

 Analysis of stride segmentation methods to identify heel strike

Victoria Turchick, Yulia Yatsenko, and Mark Redfern Human Movement and Balance Laboratory, Department of Bioengineering............................. 98  Assessment of luminal fillers in nerve guidance conduits for promoting regeneration following peripheral nerve injury Carol Vinski, Tyler Meder, Bryan Brown Department of Bioengineering, McGowan Institute for Regenerative Medicine, Renerva, LLC ................................................................................................................... 102 u Neural Network-based approximation of model predictive control applied to a flexible shaft servomechanism Yuanwu He, Yuliang Xiao, Nikhil Bajaj Department of Mechanical Engineering and Materials Science............................................. 107 u Finite element analysis of stents under radial compression boundary conditions with different material properties Yaoyao Xu, Jonathan P. Vande Gees Department of Mechanical Engineering and Material Science, Department of Bioengineering............................................................................................ 111 Index............................................................................................................................... 114

u Bioinformatic analysis of fibroblast-mediated therapy resistance in HER2+ breast cancer Jacob C. McDonald, Ioannis K. Zervantonakis Tumor Microenvironment Engineering Laboratory, Department of Bioengineering, UPMC Hillman Cancer Center.............................................................................................. 60 * Reviewers’ Choice †

Editors’ Choice

Category Definitions

u Computational Research—using computational techniques to address a scientific question  Device Design—focusing on the development of a product or device  Experimental Research—using laboratory methods to achieve a novel overarching experimental aim  Methods—developing new techniques and tools for research and design  Review—summarizes the current state of knowledge on a particular topic


A Message from the Associate Dean for Research Marcus Tullius Cicero was one of the first theorists to discuss “ingenium” as a cognitive process whereby we discover our world. In his dialogue De Oratore, written in 55 BCE, he notes that “Oratory is an art [like architecture] in the loose sense that its successes can be codified and taught; but the chief virtue of the orator is inborn, ingenium, from which sharpness of mind arise sharpness in invention, richness in exposition and ornament, firm and long-lasting memory.” Here at the University of Pittsburgh Swanson School of Engineering, we strive to support our students in their successes, in how they are taught and mentored, in an effort to encourage and echo that precise message. And this is how our own Ingenium was born. On behalf of the Swanson School and U.S. Steel Dean of Engineering James R. Martin II, I proudly present the seventh edition of Ingenium: Undergraduate Research at the Swanson School of Engineering, a compilation of articles representing the achievements of our exceptional undergraduate students and their 2020 summer David A. Vorp, Ph.D. research projects. Despite a year full of turbulence and uncertainty, our students once again showed that engineering is at the forefront of leading the way to change, adaptability, and flexibility. Living through a once-in-a-lifetime pandemic, they thrived and persevered. They showed resilience and rose to the challenging times that we’ve endured. Some of them even embraced the pandemic and used it as a way to further their academic and research goals. As with each year and with each edition of Ingenium, this year was no different, despite the challenges brought forth by the pandemic. We’ve continued to see notable and impressive academic and professional growth and development in our undergraduate students when they are given opportunities to engage in scientific research. We witnessed students taking the knowledge, skills, and information that they learn in their course work and applying it in a meaningful and intentional manner outside the classroom. These students are the future of both our highly accredited institution and our world. They will go on to become engineers, scientists, physicians, or whatever else they set their mind to, and they will, undoubtedly, make significant impacts in the fields of technology, medicine, travel, space, communication, and so much more. The student authors of the articles in this issue of Ingenium studied mostly under the guidance of faculty mentors in the Swanson School. In most cases, the research was largely conducted remotely due to the pandemic. At the conclusion of the summer research program, students were asked to submit an abstract summarizing the results of their research, and the abstracts were reviewed by the Graduate Student Review Board (GSRB). The authors of the highest-ranked abstracts were invited to submit full manuscripts for peer review by the GSRB for inclusion in Ingenium. Therefore, Ingenium serves as more than a record of our undergraduate student excellence in research; it also is a practical experience for our undergraduate students in scientific writing and in the author’s perspective of the peer-review process and additionally provides graduate students with an opportunity to experience the editorial review process and the reviewer’s perspective of the peer-review process. I would like to acknowledge the hard work and dedication of the coeditors in chief of this issue of Ingenium, Camille Johnson and Jennifer Mak, as well as the production assistance of Reiko Becker, the team in the Office of University Communications and Marketing, and Jaime Turek. This issue would not have been possible without the hard work of the graduate student volunteers who constitute the GSRB and who are listed by name in this issue. It also is altogether fitting to thank the faculty mentors and other coauthors of the reports included in this issue. I hope that you enjoy reading this edition of Ingenium and that the many talents of our students inspire the engineers of the future. Hail to Pitt!

David A. Vorp, Ph.D. Associate Dean for Research Swanson School of Engineering University of Pittsburgh

4 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Message from the Coeditors in Chief

Jennifer Mak

Greetings! We are pleased to present the seventh edition of Ingenium: Undergraduate Research at the Swanson School of Engineering, featuring 24 articles from undergraduate students at the University of Pittsburgh Swanson School of Engineering. These students faced unprecedented challenges this year because of the COVID-19 pandemic and yet displayed incredible patience, persistence, and creativity throughout the manuscript writing and review process. This year, we have chosen to showcase two featured articles: the Reviewers’ Choice article selected by members of the Graduate Student Review Board (GSRB) and the Editor’s Choice article selected by us. We felt that these articles demonstrated particularly high quality and unique research as well as excellent writing, and we hope you enjoy them as much as we did! Ingenium provides Swanson School undergraduate students with an introduction to the peer-review process and allows them to experience communicating their research in abstract and manuscript form. All manuscripts submitted to Ingenium were reviewed by Swanson School graduate students who graciously volunteered their time to provide valuable feedback and suggestions to participants to help them improve their writing. We hope that the articles presented in this year’s edition of Ingenium showcase the diversity present in engineering labs at the University of Pittsburgh. We would like to thank everyone involved in the production of this year’s Ingenium, without whom none of this would have been possible. First, thank you to Associate Dean for Research David Vorp for his vision and continued commitment to this publication. We also are extremely grateful for the support of Jaime Turek, who guided and supported us in this process throughout the year. Thank you to all of the graduate students who served on GSRB and gave their time and expertise to review the submissions of the student authors. Finally, we thank Reiko Becker, Marygrace Reder, and the entire team at the Office of University Communications and Marketing who assisted us in the production and design of Ingenium. We are honored to have served as coeditors in chief for this year’s edition of Ingenium and are excited to share this publication with the University and the surrounding Pittsburgh community. As always, Swanson School students have demonstrated their ability to think critically, work collaboratively, and effectively communicate their ideas to a broad audience. We hope that as you read through this year’s Ingenium, you are as impressed as we are with these students’ contributions to their respective fields and their evident passion for scientific research.

Camille Johnson

Jennifer Mak Coeditor in Chief

Camille Johnson Coeditor in Chief


Graduate Student Review Board – Ingenium 2021 *Co-Editors-in-Chief Name........................................................................................................................Department Allen, Abigail..............................................................................................................Bioengineering Azar, Reem................................................................................................................Bioengineering Baker, Andrew...........................................................................................................Mechanical Engineering and Materials Science Bentley, Elizabeth......................................................................................................Bioengineering Blake, Ryan...............................................................................................................Mechanical Engineering and Materials Science Bohlke, Kayla.............................................................................................................Bioengineering Bose, Rohit................................................................................................................Bioengineering Bouzid, Zeineb...........................................................................................................Electrical and Computer Engineering Bowen, Shaniel..........................................................................................................Bioengineering Buettner, Nathanial....................................................................................................Civil and Environmental Engineering Calabrese, Tia............................................................................................................Bioengineering Cardoza, Alvaro..........................................................................................................Electrical and Computer Engineering Day, Brian..................................................................................................................Chemical and Petroleum Engineering Deng, Yifan................................................................................................................Chemical and Petroleum Engineering Ding, Ruikang............................................................................................................Mechanical Engineering and Materials Science Dobrota, Nemanja......................................................................................................Civil and Environmental Engineering El-Hajj, Bashear.........................................................................................................Civil and Environmental Engineering Ghadge, Shrinath.......................................................................................................Chemical and Petroleum Engineering Gueldner, Pete...........................................................................................................Bioengineering Henao Garcia, Sebastian............................................................................................Mechanical Engineering and Materials Science Jantz, Maria...............................................................................................................Bioengineering Jawinska, Dorota.......................................................................................................Bioengineering Johnson, Camille*......................................................................................................Bioengineering Karpe, Sanjana..........................................................................................................Chemical and Petroleum Engineering Kieffer, John..............................................................................................................Electrical and Computer Engineering Lee, Vincent...............................................................................................................Bioengineering Li, Haoran..................................................................................................................Bioengineering Mak, Jennifer*...........................................................................................................Bioengineering Maldonado, Alexander................................................................................................Chemical and Petroleum Engineering Marini, Ande..............................................................................................................Bioengineering Mckeon, Shane..........................................................................................................Bioengineering Omecinski, Katelin.....................................................................................................Bioengineering Patil, Rituja................................................................................................................Chemical and Petroleum Engineering Pawar, Ritesh.............................................................................................................Chemical and Petroleum Engineering Rekant, Julie..............................................................................................................Bioengineering Rodriguez, Brittany....................................................................................................Bioengineering Rodriguez De Vecchis, Rafael.....................................................................................Mechanical Engineering and Materials Science Routzong, Megan.......................................................................................................Bioengineering Shapiro, Monica.........................................................................................................Chemical and Petroleum Engineering Sleight, Trevor............................................................................................................Civil and Environmental Engineering So, Seth.....................................................................................................................Electrical and Computer Engineering Spencer-Williams, Isaiah............................................................................................Civil and Environmental Engineering Suri, Anisha...............................................................................................................Electrical and Computer Engineering Ting, Jordyn...............................................................................................................Bioengineering Vulimiri, Praveen........................................................................................................Mechanical Engineering and Materials Science

6 Undergraduate Research at the Swanson School of Engineering

Representations of population activity during sensorimotor transformation for visually guided eye movements* Eve C. Ayara,c,d, Michelle R. Heusserb,d, Neeraj J. Gandhia,b,c,d Department of Neuroscience, bDepartment of Bioengineering, c Center for Neuroscience (CNUP), d Center for Neural Basis of Cognition (CNBC)


Eve Ayar is a junior Neuroscience student. She is interested in investigating patterns of population activity during sensorimotor integration. After graduation, she plans to pursue a Ph.D. in Neural Computation. Eve Ayar

Michelle Heusser is a Ph.D. candidate in Bioengineering at the University of Pittsburgh. Her research in the Gandhi lab focuses on the time course of neural population activity in the superior colliculus and its relationship to motor behavior. Michelle R. Heusser

Neeraj (Raj) Gandhi, Ph.D.

Neeraj (Raj) Gandhi, Ph.D. is Professor and Graduate Program Director in the Department of Bioengineering. He also developed the Professional Masters program focused on neural engineering. His research focuses on understanding neural communication during sensation, action, and cognition.

Significance Statement

Sensorimotor transformation is a process that humans perform over 100,000 times a day—for example, when we look at or reach for objects of interest. Many areas of the brain register a sensory stimulus and convert the stimulus-related information into an appropriate motor output. Deficits in sensory and/or motor processes are implicated in a number of neurological disorders such as Parkinson’s Disease [1]. We are interested in how the context of visual behavioral tasks impacts patterns of population activity during sensorimotor transformation.

Category: Computational Research

Keywords: saccade, dimensionality reduction, eye movement, sensorimotor Abbreviations: superior colliculus (SC), peristimulus time histogram (PSTH) *

Reviewers’ Choice

Ingenium 2021


For visually guided eye movements known as saccades, neurons in the superior colliculus in the midbrain emit a volley of action potentials to register a visual (sensory) stimulus, and they also discharge another high frequency burst of spikes to move the line of sight. We investigated the representations of sensory- and motor-related population activity during two paradigms, the delayed saccade and gap tasks. The two tasks differ in instructing the time to initiate the eye movement, permitting different temporal evolutions of the sensorimotor transformation. Dimensionality reduction methods were used to visualize the pattern of activity of recorded neurons and determine if the population responses in both tasks exhibit similar visual and motor patterns despite differences in event timing and cognitive context. Preliminary analyses suggest that the visual patterns explored during the delay and gap tasks largely overlap, while patterns of motor activity present differences, leading us to believe that downstream structures may differentiate signal processing between behavioral tasks.

1. Introduction

Consider a monkey in a tree searching for a banana. If he sees something yellow appear in his visual field, he will look at it. Simply put, this is the process of sensorimotor transformation: the brain registers a sensory input (e.g., banana) and then converts it to a motor output (e.g., the act of looking at or reaching for the banana). The superior colliculus (SC) is a structure in the brain that is integral in this process because it contains both visual and motor related signals that rapidly redirect the visual axis. We want to understand how sensorimotor transformation occurs in different tasks. If the context of a behavioral task matters, we expect to see differences in SC neural activity patterns because the signals involved in sensorimotor transformation will be processed differently. To characterize different representations of population activity, neural activity patterns were analyzed across two conditions: the delayed saccade task and gap task (Figure 1). The delayed saccade task requires the animal to volitionally withhold movement generation until a later point in time. This condition separates the visual and motor bursts in time. The gap saccade task, in contrast, requires the animal to react immediately. Thus, the movement can happen as soon as visual signals are processed. In this condition, the visual and motor neural bursts can temporally overlap and may even merge into a single burst. For both tasks, we care about activity at two key time points in a trial, 1) estimated visual burst time and 2) saccade onset, indicated by the lines in Figure 1. These key points occur in both tasks but have different timelines because one task has an imposed delay, and the other does not. In this project, we focus on the similarities and differences between visual and motor bursts under the two conditions in a low-dimensional state space, with the goal of determining whether there is a different pattern of population activity relayed to downstream areas. 7

We hypothesize that the gap task’s population response will likely have a different pattern of motor activity than the delayed saccade task. If this hypothesis is supported, this may mean that the context of a behavior matters – the pattern of SC neural activity underlying a behavior will change due to the differing conditions of the task. However, it is possible that context does not matter, and patterns of activity will be similar. In this case, downstream structures may not differentiate how neural signals are processed depending on the context of the behavior.

Figure 1. Task timeline schematics and key for a typical the delayed saccade (left) and gap task (right) trial.

8 Undergraduate Research at the Swanson School of Engineering

2. Methods

All procedures were approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh and followed the guidelines of the Public Health Service Policy on Humane Care and Use of Laboratory Animals. A rhesus monkey (Macaca mulatta) underwent surgery to implant a recording chamber and allow access to the SC. On each recording session, a 16-channel linear microelectrode array (AlphaOmega Inc.) was inserted orthogonal to the SC surface of the monkey to record a small population of neurons. Every trial is initiated by presentation of a fixation point usually at the center of the screen, and the animal is required to direct its line of sight on this stimulus. In the delayed saccade task (Figure 1A), a single target, as described earlier, is presented while the monkey continues to fixate on the central stimulus. Then the fixation point is turned off, acting as the cue to make a movement to the target. This time in between the target appearance and the go cue is defined as the “delay” period which can be anywhere between 600-920 ms for each trial. The length was varied in order to keep the animal guessing when to make an eye movement, which served to minimize anticipatory responses. In the gap task (Figure 1B), the fixation point is turned off and a gap in time occurs before a single target is presented on the screen, acting as the cue to make a movement to the target. In both tasks, 16 channels of activity were recorded. We’ll refer to this as our neural population, as the channels were inspected for task-related activity consistent with the characteristic activity profile of neurons located in the SC. The target was presented at a location for which the majority of SC neurons being recorded emit many spikes.

Ingenium 2021

In both tasks, the key points in each trial are visual burst time and saccade onset. For the delayed saccade task, the visual burst peak time (estimated to be 160 ms across all trials, determined visually using PSTHs) is represented with blue and saccade onset with red (Figure 2A). For the gap task, the visual burst time (estimated to be 140 ms across all trials) is represented with cyan and saccade onset (determined using each trial’s measured reaction time) with pink (Figure 2B). These colors are used throughout the rest of our analyses comparing the neural activity underlying these key external events.

Each trial was stored for analysis in MATLAB. Two data sessions were used in these analyses (02/08/17, 03/18/17). Both sessions showed task-related neural activity in all 16 channels. Dimensionality reduction was used to summarize the population activity of many neurons. This allows us to better visualize neural activity and compare across tasks. Specifically, Gaussian Process Factor Analysis (GPFA) was used on 16-channel spike trains aligned to target onset with a 20 ms smoothing factor [2]. Operating on a high-dimensional dataset, GPFA extracts a reduced number of factors that account for a large proportion of the variance and, additionally, it smooths these factors in time. Temporal smoothing while simultaneously performing dimensionality reduction provides a more intuitive visual representation of how activity evolves over time and why activity for closer key points of a given trial are more similar that activity for those further apart in a given trial. We applied the GPFA algorithm to delayed saccade task trials only; gap task trials were then projected into the same low-dimensional space to allow for a fair comparison of neural activity by ensuring that the weights attributed to each neuron for both tasks are the same [3]. Once the activity was summarized in a low-dimensional space, the Euclidian distances between the means of similar activity for each task were calculated. The raw data was then normalized by the Euclidean distance between the means of the visual and motor activity for the delayed saccade task. 95% confidence ellipsoids were plotted along with these means in a low-dimensional space. Histograms were created for each latent factor to analyze the separation of activity at the key events across tasks for a given factor.

3. Results

Figure 2. High-dimensional neural activity during the delayed saccade and gap tasks. (A) Peristimulus time histogram (PSTH) of high dimensional activity during the delayed saccade divided into two plots: time in trial aligned to target onset and time aligned to saccade onset. The visual burst time is represented by the blue line (approximated to be 160 ms after target onset) and the motor burst is represented by the red line (at saccade onset, 0 ms). (B) PSTH of high dimensional activity during the gap task. The visual burst is represented by the cyan line (approximated to be 140 ms after target onset) and the motor burst is represented by the pink rectangle. (A, B) There are 16 channels, and all appear to have task-related activity.

So far, analyses have been performed on two data sessions (panel A/C and panel B/D, respectively, in Figure 3). As expected, based on preliminary findings in our lab, the visual subspace is separable from the motor subspace within each task condition, indicative of a unique neural population activity pattern during the respective sensory and motor behavioral epochs. Interestingly, there is greater overlap for the visual activity for the gap and delayed saccade tasks (cyan asterisks and blue dots, respectively, in Figure 3A) while the motor activity has less overlap across the tasks (pink asterisks and red dots, respectively, in Figure 3A). In session 2, the visual activity has some overlap whereas comparatively, the motor activity is essentially non-overlapping (Figure 3B). In both sessions, the visual activity overlap is greater than the motor.


To allow for better visualization, 95% confidence ellipsoids were plotted for the visual and motor activity in both the gap and delayed saccade tasks along with their respective means (Figures 3C and 3D). Again, there is greater overlap in both sessions across the two tasks for visual activity (blue ellipsoids) than for motor activity (red ellipsoids, Figures 3C and 3D). To further analyze the separation of visual and motor activity across the two tasks, the Euclidean distances between the means of each cluster (visual vs. visual and motor vs. motor) were calculated. These raw distances were then normalized using the Euclidean distance between the mean of visual and motor activity in the delayed saccade task (i.e., the task condition with the greatest separation). Table 1 displays these distances. The values agree with our results in Figure 3, in which the distance between the motor clusters (0.6453) is greater than the distance between visual clusters (0.5561). Table 1: Euclidean Distances Between Means of Activity for Delayed Saccade and Gap Tasks



Visual Burst



Saccade Onset



Table 1. Euclidean distances measured between the means of visual and motor points to compare separation across tasks. The raw measurements are normalized using the Euclidean distance between the means of visual and motor activity during the delayed saccade task (4.8560).

Diving deeper into the data, we analyzed the separation of activity across both tasks in each individual latent dimension. In plotting histograms for both tasks for either visual (Figure 4) or motor (Figure 5) activity we find via visual inspection that there are roughly normal distributions of latent magnitude values along each factor. Given that these data are preliminary, we did not perform a statistical test to determine normality. There was separation for both visual and motor epochs in the first factor, with a large amount of separation for motor, as expected. In session 1 (Figure 5A), there was also greater separation for motor activity in the second factor in comparison to visual activity for that session. This is true, although less so for the second session, and a similar trend can even be seen in the third factor for both sessions.

Figure 3. Comparison of the population activity at key events for the gap and delayed saccade tasks. After dimensionality reduction, three latent dimensions were retained. (A, B) Dots and asterisks represent delayed saccade and gap latent activity, respectively, and each point is 20 ms summary of population activity. (C, D) 95% confidence ellipsoids for visual and motor latent activity along with their respective means. Blue/cyan represent visual and red/pink represent motor, which are the key events that are being compared.

10 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Figure 4. Comparison of the latent activity magnitude (a.u.) along each latent dimension retained after GPFA for the visual activity during the gap and delayed saccade tasks. The vertical lines represent the medians of visual activity. Same color convention as Figure 3.

Figure 5. Comparison of the latent activity magnitude (a.u.) along each latent dimension retained after GPFA for the motor activity during the gap and delayed saccade tasks. The vertical lines represent the medians of motor activity. Same color convention as Figure 3.


4. Discussion

In the delayed saccade task, the target appears before the go cue is presented so sensorimotor transformation has already started. The delay period essentially increases the amount of time neurons have to evolve their activity. In the gap task, in contrast, there is no delay period; thus, the motor command can initiate as soon as visual signals are processed. Given the inherent complexities of the neural dynamics, the short time period of sensorimotor transformation may limit a thorough evolution of population activity in this task. We observed less separation of visual activity between the two tasks, an expected result because the timing of the visual burst is similar in each. In the gap task, the motor burst often overlapped or merged with the visual burst. The merged bursts reflect why the low-dimensional representation of data show similar patterns across the visual and motor activity for the gap task and less so for the delayed saccade task. Each individual neuron in the gap task does not have much time to evolve its activity. Thus, when looking across the whole population of SC neurons, we can infer that the population as a whole does not evolve its activity in such a short period of time. Other studies have looked at the specific patterns across individual latent dimensions and have deduced that each factor may represent a different neural or behavioral property [4]. In our study, we find that there is greatest separation in the first factor; however, the second and third factors also show separation. With GPFA, there is not a specific mapping between the factors and what properly they represent. In future work, we plan to use other algorithms such as demixed principal component analysis (dPCA) to explore what each of these dimensions may represent, which may provide insight to the differing patterns of neural activity that underly different behaviors [4].

5. Conclusions

Our current results imply the exact distribution of population activity during sensorimotor transformation differs across task conditions. This supports the idea that downstream structures may differentiate how signals are processed in response to different behavioral conditions. For instance, these signals ascend through the thalamus to the frontal cortex, which could potentially discern the different cognitive demands of the two tasks. Thus, the context of a behavior matters when considering how sensorimotor transformation occurs. The same signals also descend to the saccade generator region in the brain stem. The observation that the motor subspaces associated with the two tasks are not identical, albeit partially overlapping, indicates that the actual motor subspace may be larger than that characterized from an individual paradigm. These notions are ripe for future investigations.

12 Undergraduate Research at the Swanson School of Engineering

6. Acknowledgements

This project was supported by NIH Grants EY024831 and EY022854, and NIH MH 5R90DA023426-15 Interdisciplinary Training in Computational Neuroscience. I would like to acknowledge the members of Dr. Gandhi’s Cognition and Sensorimotor Integration Lab.

7. References

[1] Pretegiani, E., & Optican, L. M. (2017). “Eye Movements in Parkinson’s Disease and Inherited Parkinsonian Syndromes,” in Frontiers in neurology, 8, 592, doi. org/10.3389/fneur.2017.00592 [2] Yu, B. M., Cunningham, J. P., Santhanam, G., Ryu, S. I., Shenoy, K. V., & Sahani, M. (2009). “Gaussian-process factor analysis for low-dimensional single-trial analysis of neural population activity,” in Journal of neurophysiology, 102(1), 614–635, doi.org/10.1152/jn.90941.2008 [3] Cowley, B. R., Kaufman, M. T., Butler, Z. S., Churchland, M. M., Ryu, S. I., Shenoy, K. V., & Yu, B. M. (2013). “DataHigh: graphical user interface for visualizing and interacting with high-dimensional neural activity,” Journal of neural engineering, 10(6), 066012, doi. org/10.1088/1741-2560/10/6/066012 [4] Kaufman, M. T., Churchland, M. M., Ryu, S. I., & Shenoy, K. V. (2014). “Cortical activity in the null space: permitting preparation without movement,” in Nature neuroscience, 17(3), 440–448, doi.org/10.1038/nn.3643

Ingenium 2021

Simulating the effect of different structures and materials on OLED extraction efficiency Benjamin Bailey, Paul Leu Department of Industrial Engineering Ben Bailey was born and raised in Mount Pleasant, Pennsylvania. His research interests are nanostructures and using them to increase efficiency in OLEDs.

Ben Bailey

Dr. Leu is an Associate Professor in the Department of Industrial Engineering and the Department of Mechanical Engineering and Materials Science. He received his BS in Mechanical Engineering at Rice University in 2002, his MS from Stanford University in 2004, and his PhD from Stanford University Paul Leu, Ph.D. in 2008. Dr. Leu’s lab research focuses on designing and understanding advanced materials by computational modeling and experimental research.

Significance Statement

Organic light-emitting diodes (OLEDs) have lower efficiency than other types of lights. With a lower efficiency, more power is required to use OLEDs. Using microstructures on the surface of the OLED increases efficiency by decreasing the reflectivity of the light generated within the structure.

Category: Computational Research

Keywords: OLED, EE, Nanocone, FDTD


In order to increase the extraction efficiency (EE), the transmission of light generated in the structure must be maximized. With a maximized EE, there is less power required to produce the same amount of light. This goal is achieved by using surface structures on top of the glass layer of an organic light-emitting diode (OLED) device. Among all the available structures, we focus on the nanocones in this research, because the nanocone structures in previous studies displayed more extraction enhancement than other simple structures. Also, this research is a basis to begin a larger project, with more complicated structures, that time limitations prevented us from going deeper into. Using finite-difference time-domain (FDTD) analysis, the size of the nanocones can be optimized to maximize EE. It was found that minimizing the surface area not covered by nanocones, as the distance between nanocones was the largest factor in the simulations.

1. Introduction

Organic light-emitting diodes (OLEDs) are devices that emit light when electricity is transferred from a cathode, through an organic emissive layer, to an anode. Light is emitted from the emissive layer and extracted from the device. The efficiency of the OLED describes how much power is output by light compared to the electrical input. High efficiency OLEDs are desirable as less electricity, or energy, required to light anything from a smart card with an OLED display to lighting devices for homes and businesses. The purpose behind this project is to maximize the efficiency of OLED devices by optimizing nanocone structures in a square lattice symmetry. Lim et al. studied the efficiency of different nanostructures on the surface of OLEDs; they found that the external quantum efficiency (EQE) of nanocones in a square lattice symmetry is fourteen to thirty-five percent [1]. Kim et al. corroborated the results of the nanocone structures, reaching 31.8 percent extraction efficiency (EE) [2]. This study will begin the study of EE using various nanostructures to continue increasing the efficiency of OLEDs. Since light is reflected back into the device when entering a flat layer, it could be hypothesized that the angle of the sides of the nanocones will be the most important factor. This angle is from changing the top radius, the bottom radius, and the height of the cone. The larger the difference between the two radii along with a low height would give a smaller angle, while decreasing the difference between the two radii and increasing the height would increase the angle. This study differs from other studies by comparing the different sizes of nanocones used to increase EE. Also, the majority of past research on this topic was done in a lab, while the work shown in this paper is simulated using FDTD analysis.


2. Methods

The study consisted of using the computer software Lumerical to run the finite difference time domain simulations [3]. Within Lumerical, a script was written and used to optimize a set of variables: pitch, cone height, radius at the top of the cone, radius at the bottom of the cone. The script is stopped when the radius at the top of the cone reaches the radius of the bottom of the cone, since the size of top cannot exceed the size of the bottom. The script is also stopped when the diameter at the bottom of the cone reaches the size of the pitch. The source type used was a planar source at the point (0, 0, 0.01), the FDTD is an auto non-uniform mesh, with all PML boundary conditions.. Table 1 below shows the positions and dimensions of each layer. Figure 1 below shows the different variables. Layer




x span

y span

z span










Organic Layer







No specification








ITO Dielectric








SiN Dielectric SiO2















dx, dy




x span

y span

z span









The efficiency stated is from the transmission monitor at 550 nanometers; the center of visible light on the electromagnetic spectrum, as this is the middle of the visible range of the electromagnetic spectrum. The EE is calculated by the ratio of the total energy emitted from the source and the total energy transmitted through the surface of the device, as seen in the appendix.

3. Results

The most efficient results were found with a pitch of 100 nanometers, a cone height of 100 nanometers, a top radius of 10 nanometers, and a bottom radius of 50 nanometers. With a pitch and bottom radius equal, the surface area covered is minimized. At a wavelength of 550 nanometers, the maximum efficiency reached in the study was 15.1 percent, compared to 13.0 percent efficiency (a 16.2 percent increase) for flat glass with no nanostructures. This matches perfectly with the findings of Lim et al. when the study has a 13 percent efficiency- the nanocones increase the efficiency to 15 percent [1]. Figure 2 shows the EE across the entire electromagnetic spectrum, while Table 2 shows results comparing different extraction enhancements.

Table 1. Positions and dimensions of different layers, all in micrometers, as well as mesh information.

Figure 2. Efficiency vs. Wavelength. Shows the graph of the efficiency against the wavelength of light.

Figure 1. Variables of the cone. The shown figure is “Trial 1” from Table 2.

Using Lumerical, the internal structure of the OLED is represented by different refractive indexes within the software. Therefore, the specific materials used in the simulation are not important, as the only efficiency loss tested is light lost after the organic layer.

14 Undergraduate Research at the Swanson School of Engineering


Pitch (nm)

Height (nm)

Top Radius (nm)

Bottom Radius (nm)

EE (%)

Enhancement (%)

No cone




























Table 2. Various trials with percent increase

Ingenium 2021

The light that comes out of the OLED device is at its highest intensity at the center of the nanocone, while it is at its lowest at the uncovered glass surface. Figure 3 shows the energy field that is emitted from the device.

6. Acknowledgements

Funding for this project was provided by the Swanson School of Engineering, the Office of the Provost, and Dr. Paul Leu.

7. References

Figure 3. Energy field. This shows the intensity of energy that is coming out of the surface of the OLED. The red is where the intensity is the highest (in the cone), while the blue is where the intensity is the smallest (uncovered glass surface).

4. Discussion

Throughout the simulations, data suggests that efficiency is lost where there is more space not covered by the nanocone. When the pitch is much larger than the diameter- a cone with an equal diameter and pitch would touch at the base- there is a large amount of space uncovered by the nanocone. Also, the best results were all from the same height- 100 nanometers. This suggests that the angle of the incline combined with the amount of uncovered surface area are the greatest factors on efficiency. As the wavelength increases, the EE decreases. This is due to stress relaxation, piezoelectricity-induced quantum confined Stark effect, and stress relaxation [4]. Although, the first half of the visible light spectrum has a much higher EE- a result from a lack of these phenomena. This explains the disparity in results from Kim et al. and this study- all wavelengths tested at once against each wavelength individually across the spectrum.

[1] Lim, Tae-Bin, Kwan Hyun Cho, Yong-Hoon Kim, and Yong-Cheol Jeong, ‘Enhanced Light Extraction Efficiency of OLEDs with Quasiperiodic Diffraction Grating Layer’, Optics Express, 24.16 (2016), 17950–59 <https://doi.org/10/gg686t> [2] Kim, Yangdoo, Kwan Kim, Daehong Huh, and Heon Lee, ‘Improvement of Light Extraction Efficiency of OLED Using Various Optical-Functional Nano-Structures’, Ceramist, 21 (2018), 64–79 <https://doi.org/10.31613/ceramist.2018.21.1.06> [3] ‘Nanophotonic FDTD Simulation Software - Lumerical FDTD’, Lumerical, Last visited 11 November 2020 <https://www.lumerical.com/products/fdtd/> [4] ‘External Quantum Efficiency - an Overview | ScienceDirect Topics,’ Last visited 11 November 2020 <https://www.sciencedirect.com/topics/engineering/external-quantum-efficiency>

8. Appendix

Eq. 1: EE = (Total Energy Transmitted)/(Total Energy Created)

5. Conclusions

In summary, using nanocones to increase the EE of an OLED device is effective. Eliminating the uncovered glass surface with nanocones was the largest factor on EE, because that is where the most light gets reflected back into the device. This was determined because the EE was the greatest when the diameter of the bottom radius of the cone was equal to the length of the pitch, so the area of uncovered glass was minimized.


Spike decontamination in local field potential signals from the primate superior colliculus Megan Blacka, Clara Bourrellyb, Neeraj Gandhib, Ahmed Dallala Department of Electrical and Computer Engineering, bDepartment of Bioengineering a

Megan is a senior from McMurray, PA. She is studying electrical engineering with a minor in mechanical engineering. After graduation, she plans to pursue a PhD program with research interests in signal processing and control. Megan Black

Dr. Dallal primary focus is on education development and innovation. His research interests include biomedical signal processing, biomedical image analysis, and computer vision, as well as machine learning, networked control systems, and human-machine learning. Ahmed Dallal, Ph.D.

Significance Statement

Spiking activity in the superior colliculus of primates causes contamination in the local field potential signals. This work focuses on removing spike bleed from LFP recordings using an adaptive method for spike removal.

Category: Methods

Keywords: spike contamination, signal processing, local field potential Abbreviations: local field potential (LFP), superior colliculus (SC), adaptive spike removal (ASR)

16 Undergraduate Research at the Swanson School of Engineering


Communication inside the superior colliculus (SC) of primates depends largely on neuronal spiking activity and local field potentials (LFP). While the monkey is performing a visuomotor task, spiking activity for neurons align with the LFP in the SC. Measured LFP signals contain transmembrane voltage changes and contamination from spiking activity. By removing spike bleed contamination, LFP signals can provide insight into communication in the SC. This project focuses on an adaptive method to remove contamination in lower frequencies as low frequencies contain valuable information about the LFP response to stimulus during a task. The adaptive spike removal (ASR) method tunes spike bleed removal based on the Fourier components of the LFP signal. Analysis in the time and frequency domains revealed key patterns in the deviation between spike removed LFP and spike contaminated LFP signals. Further investigation based on neuron types is proposed for future research into spike bleed removal to tune contamination removal based on the spike waveform for each neuron type.

1. Introduction

Analyses of neural communication in the superior colliculus (SC) during the sensorimotor transformation are mainly based on the analyses of local field potential (LFP) and spiking activity. During a visuomotor task, various neurons in the SC synchronize spiking activity to LFP. Investigating this relationship between spikes and LFP provides insight into how the SC processes information. Rather than reflecting transmembrane voltage changes alone, LFP signals can be contaminated by spiking activity occurring in neurons in close proximity, within ~200 μm [1]. One solution to spike bleeds might be to move LFP electrode tips further apart, > 200 μm, however, this creates new issues as the LFP is not homogeneous over that distance. These limitations necessitate a post-recording process in order to decontaminate LFP signals and filter out spike bleeds. Current methods [2] focus on removing the average contribution of a neuron during a spike. These approaches rely on estimating the spike, designing a filter to predict the influence of spikes on LFP, and altering data to remove spikes and interpolating. These methods are limited in their ability to remove the contamination of spikes, often leaving contamination in lower frequencies. Unlike existing methods, this project explored an adaptive method that allows removal to be tuned based on the Fourier components in time segments surrounding each spike. This method aims to limit the effects of spike bleeds in lower frequencies, 25 -200 Hz, as these lower frequencies contain important modulations that provide info about communication in the SC. This project focuses on implementing an adaptive method to remove spike contamination in recorded LFP signals from the primate superior colliculus.

Ingenium 2021

2. Methods 2.1 Data acquisition from SC of primates

Laminar probes (16 or 24 channels) were used to record neural activity across different layers of the SC while monkeys performed the well-known delayed saccade task [3], which facilitates analysis of visuomotor transformation. This task consists of several steps that create distinct modulation of the LFP signals. First, the monkey fixates on a central fixation point presented on a screen. Next, a peripheral target appears, while the monkey continues to fixate on the central fixation point. After a variable delay period, the monkey is given a go cue by the disappearance of the central fixation point. This go cue allows the monkey to make a saccade, characterized by a rapid eye movement toward the peripheral target. To allow for continuity between trials, segments of LFP can be aligned based on target onset or saccade onset.

2.2 LFP signal processing and analysis

Processing of the recorded LFP signals in MATLAB includes four stages: (i) boundary spike rejection, (ii) adaptive spike removal [4], (iii) filtration and downsampling, and (iv) data alignment. Figure 1 illustrates this process for removing spike contamination from a schematic LFP signal. The LFP signal is shown in blue for the duration of a trial with individual spikes shown as red tick marks. The major trial landmarks are indicated along the time axis. For comparison, original datasets are processed using only stages iii and iv to reflect the spike contaminated LFP signals.

2.2.1 Boundary spike rejection Each spike occurrence requires 250 ms of LFP signal centered around the spike timestamp to successfully remove contamination. This necessitates that spikes occurring in the first and last 250 ms of each trial be ignored and excluded from further analysis. Ignored spikes are shown in purple tick marks in Figure 1, Stage i. 2.2.2 Adaptive spike removal The adaptive spike removal (ASR) method published previously in the Journal of Neuroscience Methods [4] was employed. Each spike is analyzed in a ±250 ms window centered around the location of the spike. The LFP signal is decomposed into its Fourier components to remove spike bleed contributions. This method tunes removal based on the frequency of highest peak in the normalized power spectral density for frequencies between 2-200 Hz. Signals are processed based on this frequency to filter and remove the effects of spike contamination. Figure 1, Stage ii illustrates the removal of spike contamination by the removal of red tick marks. 2.2.3 Filtration and downsampling The LFP waveforms, originally sampled at 30 kHz, were downsampled to 1 kHz after applying the ASR method. Our analysis focuses on the lower frequency components. Data was then passed through a low pass, fourth order Butterworth filter at cut-off frequency of 250 Hz. Fourth order notch filters were also used to eliminate the power line interference at 60, 120, and 180 Hz. Figure 1, Stage iii shows the filtered signal as a smoother blue line. 2.2.4 Data alignment Each trial is separated into 1-sec segments aligned on target onset and saccade onset, shown in Figure 1, Stage iv. For segments aligned on target onset, the target is presented at 0.4 sec. For segments aligned on saccade onset, the saccade begins at 0.6 sec. LFP data is averaged across trials for each channel on the probe. This allows analysis of how the average trends change before and after spike removal. Average LFP signals before and after spike removal are plotted, along with spiking activity, to compare the spike bleed free and spike contaminated LFP signals.

Figure 1: Process for removing spike bleed


3. Results

Specific channels were isolated for analysis as the LFP response varied depending on channel depth. Figure 2 shows average LFP signals for before (blue trace) and after (red trace) removal of spike bleed, and the spike train showing average spiking activity (black trace, note that for visualization, the spiking activity is flipped and rescaled to overlap the LFP signal). Figure 2a shows signals aligned on target onset (400 ms) and Figure 2b shows signals aligned on saccade onset (600 ms). During periods of increased spiking activity, the spike bleed removed signal (red trace) deviates from the contaminated LFP signal (blue trace). This indicates removal of spikes and contamination. When spiking activity is low (around zero), the spike bleed removed (red trace) and contaminated (blue trace) signals show limited differences. Conversely, when spiking activity increases (most negative values of the black trace), the spike bleed removed signal is very distinguishable from the filtered signal.

Figure 3 shows the spectrogram for average LFP signals before and after spike bleed removal. In Figure 3a, there is a change in the magnitude of power, as shown by intensity of the color, between 500 and 600 ms. This aligns with period of high average spiking activity after target onset (400 ms). Figure 3b shows the spectrogram for average LFP signals aligned on saccade onset (600 ms). At the time of saccade onset, an artifact present in the spike contaminated average LFP signal is eliminated after ASR. These major differences in power for LFP data before and after spike bleed removal occur at the same time points as deviations in the LFP analyzed in the time domain caused by periods of high spiking activity. As observed in Figure 2, Figure 3 shows periods of low differences in power caused by low spiking activity.

Figure 3: Spectrogram of average LFP signal before (1) and after (2) ASR aligned on target onset (a) and saccade onset (b)

Figure 2: Average LFP signal before and after ASR and spike train aligned on target onset (a) and saccade onset (b). Note that for visualization, the spiking activity is flipped and rescaled on the right axis to overlap to LFP signal.

18 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

4. Discussion

The success of ASR method in removing spike artifacts from the LFP signals could be appreciated by examining the before and after frequency spectra. High frequency components, associated with the spike itself, were removed while low-frequency components remained uncompromised. This effect was best appreciated when spikes occurred in close succession, as a burst of action potentials. The repetitive filter aspect of ASR was able to remove the compound contamination, as shown in the distinct differences in LFP signal after ASR. For trials aligned on target onset, the periods of high spiking activity and strong downward deflection in the LFP are nearly simultaneous and reflect sensory processing of the sensory stimulus in the visual world. This modulation causes neurons to spike, creating the spike bleed contamination in the signal. Removal of spike contamination yields a more veridical LFP signal and potentially better insight into neuronal communication inside the brain. Specifically, insight into how the presence of a peripheral target is communicated in the SC can be explored. The analysis performed on data aligned on saccade onset yielded similar findings, which may help in deciphering communication during movement generation. Frequency domain analysis can provide insight into how spike removal affects frequency components in the LFP signal. The low frequencies analyzed in this project contain valuable information to aide in understanding how spiking activity and LFP interact in the SC.

5. Conclusion

6. Acknowledgements

The Swanson School of Engineering and the Office of the Provost for jointly funding this project. Dr. Dallal, Dr. Gandhi, and Dr. Bourrelly for their mentorship.

7. References

[1] Watson, B.O., Ding, M. and Buzsáki, G. (2018), Temporal coupling of field potentials and action potentials in the neocortex. Eur J Neurosci, 48: 24822497. doi:10.1111/ejn.13807 [2] Zanos, T. P., Mineault, P. J., & Pack, C. C. (2011). Removal of spurious correlations between spikes and local field potentials. Journal of neurophysiology, 105(1), 474–486. doi:10.1152/jn.00642.2010 [3] Massot, C., Jagadisan, U. K., & Gandhi, N. J. (2019). Sensorimotor transformation elicits systematic patterns of activity along the dorsoventral extent of the superior colliculus in the macaque monkey. Communications biology, 2, 287. doi:10.1038/s42003-019-0527-y [4] Boroujeni, K. B., Tiesinga, P., & Womelsdorf, T. (2020, January 15). Adaptive spike-artifact removal from local field potentials uncovers prominent beta and gamma band neuronal synchronization. Journal of Neuroscience Methods, 330, 108485. doi:10.1016/j.jneumeth.2019.108485 [5] Quiroga, R. Q., Nadasdy, Z., & Ben-Shaul, Y. (2004). Unsupervised Spike Detection and Sorting with Wavelets and Superparamagnetic Clustering. Neural Computation, 16(8), 1661-1687. doi:10.1162/089976604774201631

Investigation into removing spike bleed in LFP signals can provide insight into the relationship between neuronal spiking activity and LFP. This project focused on applying a process for removing spike contamination from LFP signals and showed deviations from contaminated LFP signals during periods of high spiking activity. This spike bleed removal process was applied to all spikes, however, spikes recorded from a particular channel can come from various cells creating distinct waveforms. Further research could include sorting spikes based on its cell and removing spike bleed on the different spike populations. This analysis would include spike detection and sorting software [5] to confirm the identification of spikes and separate into clusters.


Hardware acceleration of k-means clustering for satellite image compression Francis J. Fattori, Alan D. George Department of Electrical and Computer Engineering

Francis J. Fattori

Francis Fattori is a junior computer engineering student from Oxford, PA with interests in parallel processing methodologies, heterogeneous computing architectures and spacebased computing applications. Upon completing his undergraduate career, Francis hopes to continue conducting research in these areas while pursuing a master’s degree.

Dr. George is Department Chair, R&H Mickle Endowed Chair, and Professor of Electrical and Computer Engineering at the University of Pittsburgh (Pitt), and Fellow of the IEEE. His research interests focus upon high-performance architectures, apps, networks, services, systems, Alan D. George, and missions for reconfigurable, Ph.D. parallel, distributed, and dependable computing, from satellites to supercomputers.

Significance Statement

This article explores the applicability of a hybrid CPU-FPGA system-on-chip design in accelerating a color-quantization application for satellite image compression, highlighting the improvements in both performance and energy efficiency against traditional linear computing methods.

Category: Experimental Research

Key Words: hybrid system-on-chip (SoC) platform, field-programmable gate array (FPGA), parallel computing, space-based computing.

20 Undergraduate Research at the Swanson School of Engineering


Image compression is a vital component of remote camera modules on satellites to facilitate file storage and network transfer. K-means clustering is an effective algorithm for lossy image compression, but its computational complexity can render the algorithm inefficient when implemented with serially functioning processors. Issues of execution latency are magnified for space-based embedded platforms, which contain radiation-hardened processors and memory units of lower overall performance and efficiency. This paper introduces a hybrid systemon-chip (SoC) design involving both a Central Processing Unit (CPU) and a Field-Programmable Gate Array (FPGA) to serve as an accelerator for a k-means clustering image-compression application on board satellites. A PYNQ-Z2 development board housing a Xilinx Zynq-7020 SoC was used for application testing. Multiple program executions with various test images revealed that the hybrid accelerator performed k-means clustering roughly 100 times faster than the software-only baseline while consuming only 1.19 % of the energy. The application functioned at a compression ratio of 4:1 and produced output images with only minor losses in image quality.

1. Introduction 1.1 Onboard Image Compression for EO Satellites

Improvements in space camera units have enabled Earth Observation (EO) satellites to capture high-resolution images. In order for these photos to be retained for future use, image information must be stored in onboard memory units and/or downlinked to a database on Earth. However, radiation-tolerant flash memory systems present in most satellites are limited in storage capacity, and typical extraplanetary telecommunication networks do not have a bandwidth capable of downlinking full-resolution, raw images obtained by satellite cameras. Thus, a data-compression protocol is required to encode digital image information using fewer bits than the original representation. Data-compression algorithm classification as lossy or lossless and the respective roles of these algorithm types in onboard image-compression modules is elucidated in [1]. To circumvent the transfer and retention of valueless high-resolution images, common procedure for satellite photo transmission begins with downlinking lossy compressed images for preliminary analysis. Only after the image has been classified as meaningful for the particular application will image information produced by lossless compression be conveyed to ground stations. The work in this paper addresses the first component of this procedure by developing and analyzing a hybrid CPU-FPGA architecture to accelerate a lossy-compression algorithm known as k-means clustering. In alignment with the recommendations of the Consultive Committee for Space Data Systems, most modern satellite image-compression units utilize complex algorithms involving wavelet transforms and bit-plane encoding [2]. Although k-means clustering is more limited in use, it can serve as the foundation of a simple yet flexible

Ingenium 2021

compression application for satellite subsystems with stringent power and performance constraints. Additionally, the procedures and results of this study provide a general guidepost for the FPGA acceleration of other clustering algorithms in color-quantization applications.

1.2 K-Means Clustering

Clustering is a method by which the observations of a dataset are partitioned into a specific number of groups. K-means clustering is a commonly used unsupervised algorithm that performs this task by classifying data points into k disjoint clusters via the use of centroids. The steps that characterize Lloyd’s algorithm, the traditional k-means clustering process, are described in [3]. The output of this algorithm is a set of k clusters, and a labeling of each data point that specifies the centroid to which it is assigned. With slight modifications, k-means clustering can be applied to digital images to achieve color-quantization by grouping similar colored pixels [4]. The information provided by the algorithm when applied in this manner enables a fundamental reformatting of image data that consumes less storage space.

1.3 Heterogenous System-on-Chip Computing

A system-on-chip is an integrated circuit that consolidates all major computing elements on a single silicon substrate. This unification of what would otherwise appear as a multi-chip system permits a more compact form-factor, reduces energy expenditure, and facilitates rapid data transfer. Given that space-based computing demands both low-latency application execution and minimal power consumption, the SoC is a befitting computing tool for on-orbit embedded systems. A hybrid SoC platform containing both a CPU and FPGA introduces a reconfigurable logic region to the conventional processing system. The FPGA fabric can be programmed to accelerate a particular algorithm, as redundant data processing operations will execute faster on programmable logic hardware than the processing system software. Advancements in satellite-imaging technologies and the resultant increase in image resolution has spawned a research field concerned with the hardware acceleration of on-orbit image processing and compression. The inclusion of a reconfigurable FPGA fabric has become standard practice for algorithm acceleration in this computing domain [1]. Traditional software implementations of the k-means clustering algorithm are time-consuming. The serial processing nature of a CPU is deficient for k-means clustering, which requires multiple passes through highly repetitive distance calculations to achieve convergence. A hybrid SoC platform is better suited for a program of this type. The reconfigurable, parallelizable architecture of the FPGA fabric enables a fully-pipelined design capable of clustering data points at a much higher rate [5]. When paired with the rapid data allocation and control-flow capabilities of the CPU, the resulting accelerator can apply the target algorithm with significant speedup and energy savings. Xilinx, Inc. is a semiconductor manufacturing company that specializes in onboard, programmable SoC production. Xilinx’s Zynq-7000 product line offers cost-optimized

SoCs that integrate both software and hardware, making these chips ideal for space-based embedded systems. Xilinx also delivers the hardware development tools necessary to leverage high-level design techniques for their embedded platforms, namely the Vivado Design Suite and SDSoC.

2. Methods 2.1 K-Means Clustering for Image Compression

The application accelerated in this study employs k-means clustering to reduce the number of distinct colors in 3-channel digital images [3]. The k value is selected in such way as to maximize compressibility while maintaining sufficient image quality, which is explored in section 3. k centroids μ1 … μk are initialized to the RGB values of randomly selected pixels in the image, ensuring that all clusters C1 … Ck contain at least one pixel at the conclusion of the algorithm. Considering an image of N total pixels, each pixel P1 … PN is assigned to the centroid μi of closest RGB value. To simplify computations and conserve FPGA resources, Manhattan distance is utilized in place of Euclidean distance [6]. This process is expressed by the formulas: d(Pn, μi) = |Pn - μi|


Ci = { Pn | d(Pn , μi) ≤ d(Pn , μj) }


where 1 ≤ i , j ≤ k and i ≠ j. All arithmetic operations performed on pixels and clusters involve separate calculations for R, G, and B attributes. Next, each centroid μi must be set to the mean RGB value of all pixels Pn in their particular cluster Ci, as defined in the equation


where the resulting RGB values of μi are truncated to integers to avoid expensive floating-point operations. The processes represented by formulas (2) and (3) are repeated until the average change in centroid RGB value is less than or equal to a threshold distance of 2. The final RGB values of the centroids are the k unique colors that will appear in the color-quantized image. Given that the optimization objective of k-means clustering is to minimize the Manhattan distance between the centroid and member pixels of each cluster (4)

with respect to Ci and μi , the colors provided by the final centroids are adequate in preserving the visual information of the image. In the compressed image, all pixels will assume the RGB value of the centroid to which they are assigned.


2.2 Evaluating Image Quality

The compression ratio of this application can be varied according to the number of colors k to which the image is quantized. Although greater compression is advantageous in minimizing image file size, it is accompanied by reductions in image quality. Peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) index are two common image quality metrics that measure the power of corrupting noise and feature dissimilarities in a compressed image as compared to an uncompressed standard. MATLAB was used to determine PSNR and SSIM values for images compressed with this application. All image quality measurements were conducted on 10 Earth observation photos captured by digital space cameras mounted on the STP-H5-CSP experimental pallet of the NSF Center for Space, High-Performance, and Resilient Computing (SHREC). These photos were deliberately selected from this image set to provide a diverse range of geomorphological and meteorological scenes of various colors and patterns. The resolution of all test images is 489×410 pixels.

2.3 Accelerator Architecture & Design

The FPGA is responsible for assigning each pixel to the closest centroid and updating cluster properties accordingly. Pixel units are stored in registers and enter a data pipeline that determines the cluster to which the pixel belongs. The use of parallel processing methodologies permits concurrent distance calculations and creates a throughput of one pixel assignment per clock cycle, which serves as a significant improvement in data processing efficiency. Upon exiting the pipeline, a pixel unit is evaluated for cluster membership and the corresponding elements of an RGB accumulation array are incremented appropriately. Once the pipeline is depleted and all pixels have been grouped, program control returns to software. The DMA delivers hardware output to the CPU, which then computes new centroid values and evaluates algorithm convergence.

3. Results

Compressed image quality was assessed for various k values in the clustering algorithm, and thus for various compression ratios in the application. 10 test images were compressed at seven different k values, each of which is a power of 2 within the range of 4 to 256 inclusive. As k increases along this set of selected values, the corresponding compression ratio decreases in a linear fashion. k values are restricted to powers of 2 since these numbers mark an upper boundary after which compressed pixel units must assume an additional bit to specify cluster membership. PSNR and SSIM were averaged across all test images after compression at each k value. The results are presented in Figure 2.

Figure 1: System Architecture Block Diagram. The high-level operations of each SoC computing region and the dataflow within and between the SoC and DDR memory is illustrated pictorially.

For this project, the Vivado Design Suite and SDSoC were coupled to enable Xilinx SoC programming using C-based design procedures. All hardware compilations targeted the PYNQ-Z2 development board, containing a Xilinx Zynq-7020 SoC. The final system architecture is illustrated in Figure 1. The CPU begins application execution by parsing the input image into 32-bit pixel elements that report individual RGB intensities values. Using this pixel information, the initial centroid values that seed the k-means clustering algorithm are computed. The CPU places both the pixel and centroid arrays in physically contiguous memory spaces in DDR, and invokes the direct memory access (DMA) controller to transfer these arrays to the FPGA. The instantiation of the DMA offloads data transfer tasks that would otherwise encumber the CPU and create a software bottleneck.

22 Undergraduate Research at the Swanson School of Engineering

Figure 2: MATLAB Image Quality Plot. The blue curve corresponds to the left axis, depicting PSNR against k value. The orange curve corresponds to the right axis, depicting SSIM against k value. The k values analyzed are 4, 8, 16, 32, 64, 128, and 256.

Ingenium 2021

Upon designing the application for a 4:1 compression ratio, Vivado implementation reports delineated the degree of FPGA resource utilization required to realize the hardware accelerator. Availability for each of the four primary computing resources on the Zync-7020 are compared to the resources demanded by the hardware function. The resource usage metrics highlighted in Table 1 apply exclusively to the compression of 489×410 images, as image resolution dictates the size of the pipeline in the accelerator. Resource Type





# Available





# Used





% Utilization





Table 1: Vivado resource utilization report for the FPGA accelerator. Hardware usage is measured over four component types: Lookup Table (LUT), Flip Flop (FF), Block Random Access Memory (BRAM), and Digital Signal Processor (DSP).

The k-means clustering compression program was implemented not only as a software-hardware accelerated application, but also as a software only baseline application. Each implementation compressed the same 10 test images, and the average runtime and energy requirements are depicted below in tabular form. Power Draw

Execution Time

Energy Consumption

Software Only

1.402 W

0.69276 s

0.9712 J


1.672 W

0.00690 s

0.0115 J

Relative Reduction




Table 2: Application performance results for the software only application and hybridized application. Relative reduction metrics are obtained as the ratio of software only results to hybrid results.

The execution times displayed in Table 2 were collected by physically running the application on the Zynz7020 SoC at a clock frequency of 100 MHz. The power metrics were acquired from Vivado power analysis reports and serve as estimates for actual application execution. Energy consumption was computed as the product of instantaneous power draw and the execution time for each application.

4. Discussion

The plots of Figure 2 assume the expected pattern for image quality assessed at varying cluster quantities k. Beginning at high k values and moving left across the x-axis, initial reductions in k are met with slight degradations in PSNR and SSIM. However, more dramatic regressions in image quality appear upon dropping below a particular k value, which in this case is k=64. Since the marginal increase in image quality is negligible beyond k=64, this cluster quantity was selected for application development and testing. Although the exact k-value that constitutes the elbow of such image quality curves depends largely on the color composition of the captured images, the reverse exponential plot structure applies universally to any multispectral image set. The resource utilization metrics of Table 2 reveals that the accelerator requires a relatively small fraction of FPGA computing components. The proposed design consumes very few BRAM units, as the majority of data arrays are partitioned completely to permit uniform data access in the clustering pipeline. This, in turn, depletes a greater number of FFs that are required to store the individual elements of the arrays. Hence, FFs along with LUTs comprise the vast majority of storage, logic, and arithmetic operations in the accelerator. The low proportion of resource usage suggests that this application can be scaled to compress images of higher resolution and/or implement additional clusters for greater compressed image fidelity. The inclusion of the FPGA fabric to the k-means clustering compression application led to significant performance improvements. Although both solutions generate identical image output, the FPGA parallelized approach to pixel clustering operates roughly 100 times faster than the CPU serial implementation. At any given moment during application execution, the amount of power needed by the hybridized design slightly exceeds that of the software only design. These results were anticipated, as the use of the accelerator activates additional resources on the SoC. However, the substantial curtailment of execution time in the hybrid application more than compensated for this increase in power. After execution, the accelerated application consumed 84 times less energy than the software baseline. This savings is particularly significant for lossy compression applications of this type, as they will likely be invoked with every satellite image acquisition.


5. Conclusion

This paper proposed a hybrid CPU-FPGA accelerator for an onboard lossy compression application for satellite images. K-means clustering was employed to quantize image color gamut to 64 and offer a compression ratio of 4:1, producing files that consume less space in flash memory and can be transferred to ground stations at faster speeds with lower energy. The accelerated design proved successful in offering considerable reductions in application runtime and energy consumption. These performance improvements are vital for space-based embedded SoC devices, as they are limited in computing capability and energy access.

6. Acknowledgements

Funding was provided by the Mascaro Center for Sustainable Innovation (MCSI) Undergraduate Research Program coordinated by Dr. David V.P. Sanchez, Gena Kovalcik, and Ellen Cadden. Additional assistance was provided by the Center for Space, High-performance, and Resilient Computing (SHREC) Summer Undergraduate Research Group.

7. References

[1] S. Lopez et. al., The Promise of Reconfigurable Computing for Hyperspectral Imaging Onboard Systems: A Review and Trends, Proceedings of the IEEE, vol. 101, no. 3 (2013), pp. 698-722. [2] C. Thiebaut and R. Camarero, CNES Studies for On-Board Compression of High-Resolution Satellite Images, in: B. Huang (Eds.), Satellite Data Compression, Springer, New York, 2011, pp. 29-46. [3] S. Na, L. Xumin and G. Yong, Research on k-means Clustering Algorithm: An Improved k-means Clustering Algorithm, Third International Symposium on Intelligent Information Technology and Security Informatics (2010), pp. 63-67. [4] T. Saegusa and T. Maruyama, Real-Time Segmentation of Color Images based on the K-means Clustering on FPGA, International Conference on Field-Programmable Technology (2007), pp. 329-332. [5] M. Gokhale et. al., Experience with a Hybrid Processor: K-Means Clustering, The Journal of Supercomputing, vol. 26, no. 2 (2003), pp. 131-148. [6] M. Estlick, M. Leeser, J. Theiler, and J. Szymanski, Algorithmic transformation in the implementation of k-means clustering on reconfigurable hardware, ACM/SIGDA 9th International Symposium on Field Programmable Gate Arrays (2001), pp. 103-110.

24 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Calphad modeling and prediction of magnetic phase diagram for the Fe-Cr-Ni system Peiyun Feng, Liangyan Hao, Wei Xiong Physical Metallurgy & Materials Design Laboratory, Department of Mechanical Engineering & Material Science Peiyun Feng was born and raised in Chengdu, China. She transferred to Pittsburgh for undergraduate study and her passion for computational material science motivates her to become a researcher in material science and mechanical engineering. Peiyun Feng

Dr. Wei Xiong is the director of the Physical Metallurgy and Materials Design Laboratory at the University of Pittsburgh. Using the computational methods with high-throughput experiments, his lab works in materials design and advanced manufacturing, which covers a wide range of inorganic Wei Xiong, Ph.D. materials, and focuses on studying process-structure-property-performance relationships.

Significance Statement

Fe-Cr-Ni alloys are one of the most studied ternary alloy systems. In this work, the lack of experimental data for the Fe-Cr-Ni ternary thermodynamics at low temperatures was solved by theoretical modeling. The compositional dependency of both magnetic transition temperature and magnetic moment was studied, which contributes to superalloy development.

Category: Computational Research

Keywords: Magnetic phase diagram, Fe-Cr-Ni, Fe-Cr, Fe-Ni


The compositional dependency of both magnetic transition temperature and magnetic moment of binary Cr-Fe, Fe-Ni, and Cr-Ni alloys, and ternary Fe-Cr-Ni alloys was investigated using magnetic modeling. Previous work on these systems were mainly based on the Inden–Hillert–Jarl magnetic model and Weiss Factor introduced by Hertzman and Sundman. In this work, a revised model proposed by Xiong was applied to study the compositional dependency of both magnetic transition temperature and magnetic moment, which reduced artificial parts of calculations. The reliability of the revised model was proved by the fact that calculated Curie temperature, Néel temperature, and magnetic moment of three binary systems all agree well with experimental data. The predicted variation trend of ternary magnetic properties was also supported by available experimental and theoretical calculation results. The low-temperature properties prediction can be benefited by the showing the improvement of description of the magnetic phase diagrams. The improved simulation of thermodynamic properties is critical to steel and superalloy development.

1. Introduction

The Fe-Cr-Ni alloy is one of the frequently used and studied materials. Many industrial austenitic, ferritic, and martensitic steels are formed based on this alloy system [1]. Their distinct magnetic and mechanical properties make them the promising candidate for structural and functional materials. For example, 304 stainless steel with a significant content of Fe, Cr, Ni is frequently applied in light water and fast breeder reactors [2]. However, there have been limited experimental data for the Fe-Cr-Ni ternary thermodynamics at low temperatures due to slow kinetics, which makes the equilibria to be hard to reach [1]. To solve this problem, theoretical modeling such as ab initio calculations [3] is usually applied. In this work, the compositional dependency of both magnetic transition temperature and magnetic moment was studied to understand the magnetic ordering contribution to the alloy thermodynamics of the Fe-Cr-Ni ternary alloys, which is critical to steel and superalloy development. The Inden-Hillert-Jarl (IHJ) magnetic model was proposed in the Calphad annual conference in 1978 [4]. This model was designed only for the ferromagnetic (FM) state. To evaluate the antiferromagnetic (AFM) state, Hertzman and Sundman improved the model by introducing the so-called Weiss Factor (-1 for body-centered cubic and -3 for face-centered cubic) [5]. This model applied a single Redlich-Kister [6] polynomial to fit both the Curie temperature and Néel temperature. This has the advantage that both FM and AFM were fitted, and the critical magnetic transition point could be found where this fitted curve passed zero. But large discrepancy between the fitted curve and experimental information occurs when using this model to demonstrate features of both FM and AFM states simultaneously. Artificial parts of fitting curves will result from this model, thus leading to inaccurate 25

prediction. Xiong et al. [7] improved the IHJ model by using two Redlich-Kister polynomials to fit the Curie and Néel temperatures separately and this model was adopted in this work. Besides, the effective magnetic moment that presents the real magnetism distribution to reproduce the heat capacity suggested by Xiong et al. [7] was used. Local magnetic moment of atoms can be used to calculate the effective magnetic moment, otherwise, the fitting of heat capacity due to magnetism can generate a reasonable value of the effective magnetic moment.

3. Results and Discussion

2. Thermodynamic Modeling

Experimental magnetic transition temperature and local magnetic moment available in literature [9-46] were collected for ternary Fe-Cr-Ni and binary Cr-Fe, Fe-Ni, and Cr-Ni alloys. The Redlich-Kister series expansions (Eqs. (1)-(2)) were applied to fit the magnetic properties of binary systems.



where xA and xB are compositions for element A and B, respectively. T(A) (T(B)) and β(A) (β(B)) are magnetic transition temperature and magnetic moment of pure element A (B). a and b are the constant in the Redliche-Kister polynomial term used to fit experimental data. In this work, those parameters are fitted by Thermo-Calc. It is noteworthy that all magnetic properties for pure Fe, Cr and Ni were from values published in the SGTE database [7]. Based on Xiong’s revised model [8], the endpoints of each fitted polynomial were set to be the pure element parameters, which are also called unary parameters. If the element is antiferromagnetic, the Curie temperature is negative, and the Néel temperature is positive, vice versa. Ternary parameters of Fe-Cr-Ni alloys were fitted in the same way by inputting both unary parameters of pure Fe, Cr, and Ni and binary parameters of Cr-Fe, Fe-Ni, and Cr-Ni alloys. All these parameters were used for describing the compositional dependency of magnetic properties. The local magnetic moment of atoms was used to calculate the effective magnetic moment through Eq. (3): (3)

where β* is the effective magnetic moment, xi is the composition of element i and βi is the local magnetic moment of element i.

26 Undergraduate Research at the Swanson School of Engineering

Figure 1: Magnetic moment (μB) at 298 K for the Fe-Cr-Ni alloys (a) of fcc phase and (b) of bcc phase. Symbols denote the experimental data [9-23]. Red solid lines represent the effective magnetic moment calculated in this work using Xiong’s revised model.

Ingenium 2021

there will not be artificial part of fitted Néel temperature curves by using this improved model. Comparing with experimental data points, the fitted curves in both Figs. 1 and 2 demonstrate that the binaries could be described satisfactorily by this improved model. As a result, the prediction of magnetic properties of binary systems requires high accuracy. For some systems Néel temperature parameters were not fitted because some phases in certain systems do not exhibit AFM state. For example, bcc Fe-Ni binary system only shows FM state over its whole composition range. Thermodynamics in low-temperature range is of vital importance to the magnetism of alloys and thus have significant needs in various applications. The reliability of the descriptions of unaries and binaries would directly influence the reliability of ternary prediction. By extrapolation of resulted binary properties and fitting of ternary experimental data, the compositional dependency of the magnetic transition temperature and magnetic moment for ternary Fe-Cr-Ni alloys were shown by heatmaps in the middle triangle of Figs. 1 and 2. Normalized by the color bar, the variation of colors in the heatmaps demonstrates the general trend of Curie temperature, Néel temperature and magnetic moment variation. Figure 1(a) shows that the effective magnetic moment decreases rapidly with increasing Cr content in the fcc phase. This is because Cr does not exhibit large magnetic moment in fcc phase (magnetic moment is 0 for pure Cr in fcc phase). Figure 1(b) shows that similar trend occurs in bcc phase of Fe-Cr-Ni alloys. The magnetic moment reaches maximum at the Fe rich side and decreases as Fe content decreases. For magnetic transition temperature shown in the heatmaps in Fig. 2, a boundary could be seen where Curie temperature drops to below zero (from red to purple). This possibly demonstrates the composition range where FM transforms to ATM. In conclusion, these figures demonstrate reasonable descriptions of magnetic phase diagrams of Fe-Cr-Ni alloys. Figure 2: Magnetic transition temperature for the Fe-Cr-Ni alloys (a) of fcc phase and (b) of bcc phase. Symbols denote the experimental data [24-46]. Red solid and dashed lines (no experimental data for certain systems) represent the Curie temperature calculated in this work using Xiong’s revised model, while blue solid lines represent the Néel temperature calculated in this work.

Figs. 1 and 2 shows the magnetic property curves of binary Cr-Fe, Fe-Ni, and Cr-Ni alloys in this work. The Curie and Néel temperatures were described by two separate Redlich-Kister polynomials. If both fitted Curie and Neel temperatures curves are below zero at specific composition, that indicates that the real magnetic transition temperature is zero and this phase is non-magnetic. As shown in Fig. 2, there is no cross point of Curie and Néel temperature curves above 0 K, which means that FM and ATM states would not co-exist above 0 K. Consequently,

4. Summary

In this work, a revised magnetic model [9] was applied to calculate the magnetic properties of ternary Cr-Fe, FeNi, Cr-Ni, and Fe-Cr-Ni alloys. The calculated Curie temperature, Néel temperature, and magnetic moment of three binary systems all agree well with experimental data, which proves the reliability of the new magnetic model adopted in this work. Meanwhile, the extrapolation from the binary systems to the Fe-Cr-Ni system can satisfactorily predict the same variation trend of magnetic properties suggested by available experimental and theoretical calculation results. This work can hopefully stress the importance of magnetic models for the further development of the CALPHAD method and draw more attention to the improvement of description of magnetic phase diagrams.


5. Acknowledgements

Swanson School of Engineering at University of Pittsburgh. W.X. acknowledges funding from the National Science Foundation under award DMR - 1808082, and from the University of Pittsburgh, through the Central Research Development Fund (CRDF).

6. References

[1]Wróbel J S, Nguyen-Manh D, Lavrentiev M Y, et al. Phase stability of ternary fcc and bcc Fe-Cr-Ni alloys, J. Physical Review B. 91 (2015) 1098-0121. [2]Toyama T, Nozawa Y, Van Renterghem W, et al. Grain boundary segregation in neutron-irradiated 304 stainless steel studied by atom probe tomography, J. Journal of nuclear materials. 425 (2012) 71-75. [3]Kro ̈ mann et al. Phys. Stat. Sol. B 251, 54, 2014. [4]Hillert M, Jarl M. A model for alloying in ferromagnetic metals. Calphad. 2 (1978) 227-238. [5]Hertzman S, Sundman B. A thermodynamic analysis of the Fe-Cr system, J. Calphad. 6 (1982) 67-80. [6]Redlich O, Kister A T. Algebraic representation of thermodynamic properties and the classification of solutions, J. Industrial & Engineering Chemistry. 40 (1948) 345-348. [7]Dinsdale A T. SGTE data for pure elements, J. Calphad. 15 (1991) 317-425. [8]Xiong W, Chen Q, Korzhavyi P A, et al. An improved magnetic model for thermodynamic modeling, J. Calphad. 39 (2012) 11-20. [9]Liangyan Hao, Andrei Ruban, Wei Xiong. CALPHAD Modeling Based on Lattice Stability from Zero Kevin and Improved Magnetic Model: A Case Study on the Cr-Ni System. CALPHAD, to be published. [10] Sadron C. Ferromagnetic moments of elements and the periodic system, J. Ann. Phys. 17 (1932) 371-452. [11] Van Elst H C, Lubach B, Van den Berg G J. The magnetization of some nickel alloys in magnetic fields up to 15 kOe between 0° K and 300° K, J. Physica. 28 (1962) 1297-1317. [12] B. Chiffey and T. Hicks. The spontaneous moment of nickel-chromium alloys. Physics Letters A. 34 (1971) 267–268. [13]Besnus M J, Gottehrer Y, Munschy G. Magnetic properties of Ni Cr alloys, J. Physica Status Solidi. 49 (1972) 597-607. [14]Tange H, Yonei T, Goto M. Forced Volume Magnetostriction of Ni-Cr Alloys, J. Journal of the Physical Society of Japan. 50 (1981) 454-460. [15] Kankainen A, Elomaa V V, Eronen T, et al. Mass measurements in the vicinity of the doubly magic waiting point Ni 56, J. Physical Review C. 82 (2010). [16]Müller J B, Hesse J. A model for magnetic abnormalies of FeNi Invar alloys, J. Zeitschrift für Physik B Condensed Matter. 54 (1983) 35-42. [17] Xiong W, Zhang H, Vitos L, et al. Magnetic phase diagram of the Fe–Ni system, J. Acta materialia. 59 (2011) 521-530.

28 Undergraduate Research at the Swanson School of Engineering

[18]Tsoukalas et al. Zeitschrift für Metallkunde. 78 (1987) 498-501. [19]Inoue J, Yamada H, Shimizu M. Calculation of the Magnetic Properties for fcc Fe–Ni Alloys, J. Journal of the Physical Society of Japan. 46 (1979) 1496-1503. [20] Collins M F, Forsyth J B. The magnetic moment distribution in some transition metal alloys, J. Philosophical Magazine, 8 (1963) 401-410. [21]Fischer S F, Kaul S N, Kronmüller H. Critical magnetic properties of disordered polycrystalline Cr 75 Fe 25 and Cr 70 Fe 30 alloys, J. Physical Review B, 65 (2002). [22]Tarafder K, Ghosh S, Sanyal B, et al. Electronic and magnetic properties of disordered Fe–Cr alloys using different electronic structure methods, J. Journal of Physics: Condensed Matter. 20 (2008). [23] Klaver T P C, Drautz R, Finnis M W. Magnetism and thermodynamics of defect-free Fe-Cr alloys, J. Physical Review B. 74 (2006). [24]Peschard M. Contribution a l’etude des ferronickels, D. Univ. de Strasbourg. (1925) [25]Gossels G. Versuche über die Stabilität der Hysteresis Fe-Ni‐Legierungen, J. Zeitschrift für anorganische und allgemeine Chemie. 182 (1929) 19-27. [26] Morita H, Fujimori H, Hashimoto S, et al. Study on the Fe-Ni phase diagram by multilayer samples, J. Journal of magnetism and magnetic materials. 126 (1993) 173-175. [27]Fukamichi K. Magnetically Insensitive Ternary Invar Alloys of Cr, Fe and 4d Transition Metals, J. Transactions of the Japan Institute of Metals. 18 (1977) 81-86. [28]Yamamoto H. A study on the nature of aging of FeCr alloys by means of the Mössbauer effect, J. Japanese Journal of Applied Physics. 3 (1964) 745. [29]Adcock, J. Iron Steel Inst. 124 (1931) 99. [30]Rajan N S, Waterstrat R M, Beck P A. Temperature‐Dependence of the Electrical Resistivity of bcc Cr–Fe Alloys, J. Journal of Applied Physics. 31 (1960) 731-732. [31]P. Oberhoffer and H. Eisen. Stahl und Risen. 47 (1927) 2021-2031. [32]R. Vilar and G. Cizeron, Mem. Etud. Sci. Rev. Met. 79 (1982) 687. [33]Tsunoda Y, Mori M, Kunitomi N, et al. Strain wave in pure chromium, J. Solid State Communications. 14 (1974) 287-289. [34] Mitchell M A, Goff J F. Electrical Resistivity and the Depression of the Néel Temperature in Cr-Mo and CrFe, J. Physical Review B. 5 (1972) 1163. [35]Swartzendruber L J, Itkin V P, Alcock C B. The Fe-Ni (iron-nickel) system, J. Journal of phase equilibria. 12 (1991) 288-312. [36]A. A. Jansson, Thermodynamic Evaluation of the Cu–Fe–Ni System, Report No. TRITA-MAC 0340, The Royal Institute of Technology. (1987) 1–10. [37]Simpson M A, Smith T F. Thermodynamic studies and magnetic ordering of Ni-Cr alloys close to the critical composition, J. Australian Journal of Physics. 35 (1982) 307-320.

Ingenium 2021

[38]Chaudhary V, Ramanujan R V. Magnetocaloric properties of Fe-Ni-Cr nanoparticles for active cooling, J. Scientific reports. 6 (2016) 1-9. [39]Iorga A, Codescu M M, ŞABAN R, et al. Low Curie Temperature in Fe-Cr-Ni-Mn Alloys, J. UPB Sci. Bull., Series B. 73 (2011) 8. [40]Almasan C, Datta T, Edge R D, et al. Low-temperature phase and magnetic interactions in FCC Fe-Cr-Ni alloys, J. Journal of magnetism and magnetic materials. 80 (1989) 329-338. [41]Rhys-Jones T N, Grabke H J, Kudielka H. The effects of various amounts of alloyed cerium and cerium oxide on the high temperature oxidation of Fe-10Cr and Fe-20Cr alloys, J. Corrosion Science. 27 (1987) 49-73. [42] Lavrentiev M Y, Wróbel J S, Nguyen-Manh D, et al. Magnetic cluster expansion model for random and ordered magnetic face-centered cubic Fe-Ni-Cr alloys, J. Journal of Applied Physics. 120 (2016). [43] Wróbel J S, Nguyen-Manh D, Lavrentiev M Y, et al. Phase stability of ternary fcc and bcc Fe-Cr-Ni alloys, J. Physical Review B. 91 (2015). [44] Ōhara S, Kōmura S, Takeda T. Magnetic Properties of Pseudo-Iron Fe1-x (Cr0. 5Ni0. 5) x Ternary Alloys, J. Journal of the Physical Society of Japan. 34 (1973)14721476. [45]Rode V, Deryabin A, Damashke G. Ferro-antiferromagnetic transition in the alloy system Fe 65 (Ni 1-x Cr x) 35 at low temperatures, J. IEEE Transactions on Magnetics. 12 (1976) 404-406. [46]Warnes L A A, King H W. The low temperature magnetic properties of austenitic Fe-Cr-Ni alloys: 2. The prediction of Néel temperatures and maximum susceptibilities, J. Cryogenics. 16 (1976) 659-667.


Computer vision methods for tool guidance in a finger-mounted device for the blind Yuxuan Hu, Rene R. Canady, B.S., Roberta Klatzky, Ph.D. and George Stetten, M.D., Ph. D. The Visualization and Image Analysis (VIA) Laboratory, Department of Bioengineering Yuxuan Hu is a Bioengineering senior at the University of Pittsburgh. He will be a graduate student next year at ETH in Zurich studying surgical vision and imaging.

Yuxuan Hu

Rene R. Canady, B.S.

Rene R. Canady received a BS in bioengineering in 2020 from the University of Pittsburgh and is currently pursuing a Ph.D. in Sociology at Washington University, St. Louis. Her research interests include engineering ethnography and racial controversies in health. Roberta Klatzky is a professor of Psychology and Human-Computer Interaction at Carnegie Mellon. She enjoys combining basic research with applications.

Roberta Klatzky, Ph.D.

George Stetten, M.D., Ph.D.

George Stetten is Professor of Bioengineering at the University of Pittsburgh. He directs the Visualization and Image Analysis (VIA) Laboratory and the Music Engineering Laboratory (MEL) and is a fellow in the American Institute for Medical and Biological Engineering.

Significance Statement

Tool handling and close-up operation are challenges for the visually impaired that have not been well addressed by assistive devices. We have come up with novel and computationally efficient computer vision methods for real-time tool detection and target motion classification, to improve on “FingerSight,” a finger-mounted haptic device designed to help the visually impaired complete daily tasks.

Category: Methods

Keywords: Assistive Technology, Haptics, Computer Vision, Image Analysis 30 Undergraduate Research at the Swanson School of Engineering


People with visual impairment often find difficulty in performing high precision tasks such as interacting with a target using a tool. We propose a device for the visually impaired that can provide accurate localization of targets via vibratory feedback that makes tool-handling tasks easier. The device is an adaptation of our existing system, “FingerSight,” a finger-mounted device consisting of a camera and four vibrators (tactors) that respond to analysis of images from the camera. We report here on a new design for the hardware and optimization of real-time algorithms for tool detection and motion classification. These include the determination that a duration of tactor vibration of 90 ms yielded a minimum error rate (5%) in tactor identification and that the best parameters for the tool recognition algorithm were threshold = 13, kernel size = 11, yielding an average tracking error of 23 pixels in a 640 x 480 pixel camera frame. Anecdotal results obtained from single healthy blind-folded subject show the device’s functionality as a whole and potential for providing guidance for the visually impaired manipulating tools in real-life scenarios.

1. Introduction

In 2015, approximately 3.22 million people in the United States were visually impaired, while 1.02 million of them were blind. And by 2050, the number of people afflicted by visual impairment is projected to double [1]. Such increasing prevalence of visual impairment has driven scientists and engineers to develop various assistive technologies, many of which utilize computer vision and haptics. Significant progress has been made in the area of user mobility in pedestrian environments. For example, a system using tactors attached to the torso has been developed to localize the user with respect to the surroundings and guide travel while avoiding obstacles [2]. Another study focuses on detecting aerial obstacles with a stereo vision wearable device [3]. However, few solutions have addressed the commonplace problem of finding targets in peripersonal space, i.e., the space within reach where objects can be grasped and manipulated. Our laboratory previously developed a system with a hand mounted binocular cameras and five vibrators to help the user locate nearby targets in 3D space using a depth map generated with computer vision methods [4]. Our present system, “FingerSight,” is a wearable device originally intended to help the visually impaired navigate in the environment and locate targets in peripersonal space. The device, mounted on the finger, contains a camera and a set of vibrators (tactors) that are activated based on computer vision analysis of the real-time camera image. Previous research on this technology by Satpute et al., in our laboratory, demonstrated its effectiveness for guiding blindfolded participants to move their hand to reach an LED target located in front of their body [5]. Based on the working prototype by Satpute et al., our present research focuses on incorporating the experience of using hand-held tools into the basic FingerSight framework. Accurate localization and feedback are needed

Ingenium 2021

particularly when targets are to be optimally contacted using a tool. Two major challenges are real-time recognition of a variety of tools and immediate handling of the contact of tool and target. In this paper, we present a new version of FingerSight with improved functionality and novel computer vision methods to address the two challenges. We developed an experiment to determine the optimal vibratory signal duration in terms of discriminating which of the four tactors was activated. For our initial algorithm to detect hand-held tools used in everyday situations, we chose a plastic fork, since dining is a common scenario in which target may not ideally be manipulated by hand. We conducted experiments to optimize two key parameters in our computer vision algorithm: intensity threshold and kernel size.

2. Methods 2. 1. Hardware Configuration and Error Rate Computation

The hardware design of our device aims to optimize tactile feedback delivery for guidance of the hand in 3D space, while providing portability and comfort over a range of finger sizes without impeding normal use of the hand. It combines a miniature video camera and four tactors embedded into a soft plastic ring attached by wire to a portable Raspberry Pi 4 computer. The ring has four expandable joints within a single 3D printed unit, built on an Ultimaker 3 FDM 3D printer using NinjiaFlex filament, made from formulated thermoplastic polyurethane material. Elongation of the filament by up to 660% in expandable joints provide comfort over a range of finger size, while alowing flexible wires to be securely routed. The polyurethane composition provides adequate vibration reduction, reduceing cross-transmission of vibratory signals between the tactors. The four cylindrical tactors are attached within the ring to contact the user’s index finger, providing guidance in 4 directions in a finger-centric coordinate system. (Fig. 1)

Figure 1. Left: The ring with tactors wired in. Tractors are marked with numbers each of which represents a unique direction in which a movement is expected after vibration (consistent with Table. 2). Right: device worn on hand with camera inserted and guiding hand-held spoon towards plastic strawberry.

We performed an experiment on how well users can discriminate between the four directions when individual tactors are activated on a healthy and blind-folded, 21-yeard-old right-handed male. It was designed not to prove statistical significance, but rather to produce a viable system for further experimentation. Three trials were

conducted with different durations of vibratory signals (60 ms, 90 ms, and 120 ms), delivered by four identical vibrators with an operating voltage of 3.7 V and frequency of 205 Hz. The participant was initially trained by activating the tactors in sequence with the correct answer viewed on a screen. Then, 40 signals (10 in each direction) were given in random order. For each signal, the participant was instructed to press one of the four arrow keys to record the detected location, with no time limit. The results were pooled over trials and for each tactor, with a confusion matrix and error rate computed.

2.2 Computer Vision Algorithm

2.2.1 Tool Detection We assume that the tool is held in a stationary grip without motion relative to the hand, at least over the short term. Thus, the tool will ideally appear to be stationary in the view of the camera, while the background will constantly change as the user moves their hand around. To detect the stationary tool, we calculate the mean pixel intensity at every pixel in the camera’s view over the latest 10 video frames and construct a “mean frame” M(x, y). Assuming the tool stationary is in the view of the camera, the intensity of pixels i(x, y) within the tool for a given frame should fall within some threshold h of the mean intensity for that pixel, while other pixels representing the background could be expected to exceed that threshold. Thus a binary mask b(x, y) can be generated by applying the given threshold to segment the tool from the background. (1)

To eliminate salt-and-pepper noise in the segmentation, we apply mathematical morphology operators of erosion and dilation with kernel size k to reduce the noise, and then we sort contours by area, with the largest contour being the tool. (Fig 2)

Figure 2. Demonstration of the tool detection algorithm. Ten most recent frames (left) are used to calculate the intermediate “mean frame” (middle), which is subtracted from the current frame (middle under), whose absolute value is thresholded to yield a binary image (right) where positive pixels (white) represent the tool.

Since the tip of the tool will always be at the location farthest away from the bottom right location (if the user is right-handed), we could identify the tip location within the largest contour by sorting the pixel coordinates. By comparing locations of the tip of the tool and the location with the target’s center, the algorithm produces a direction in which the hand should move to reach the target and gives instruction accordingly through the tactors. We had previously studied various strategies for activating the tactors 31

to indicate direction. As reported in Satpute et.al.’s study, correction of individual axes leads to better results than simultaneous signals in the case of finger mounted tactor guiding [5], and we adapted this approach in our design, e.g., the bottom tactor will activate to the vector (100, -200), instead of both the right and bottom ones at the same time. Two parameters in the algorithm that directly affect the efficacy of tool detection are the binary mask threshold value h and kernel size k for erosion and dilation, and the direct indicator of tracking accuracy is the distance between the location of the tip of the tool detected and its actual location S. To simulate different holding postures, we set the camera at 2 different distances from the tip of the tool (7 cm and 10.5cm) and recorded 10-second videos of the participant waving the tool around in spontaneous, smooth motion. The videos were recorded against a printed checkerboard to ensure a consistently changing background. We tested 9 settings with 3 empirically chosen values each for h and k: h = 1, 7, 13 & k = 11, 17, 23. Since the tool was held firmly in both videos, we manually located one tip location for each, then calculated S in every frame. By comparing the mean and standard deviation of S in each of the 9 settings, we could determine the optimal algorithm parameters. 2.2.2 Immediate Motion Classification When the tool is about to reach the target, the user may be negligent and keep moving in the last instructed direction at a relatively high speed. If the user is unsuccessful at acquiring the target, the relative motion of target to tool can fall into one of the four categories: (1) the tool passes the target without making contact; (2) the tool bumps the target straight forward; (3) & (4) the target rolls off to one or the other side of the tool. At a detection rate as fast as the camera’s frame rate allows, we decrease the complexity of the algorithm by using the tool tip location computed in the previous section and divide the camera’s field of view into four quadrants around the tool tip. A thresholded color value c(x, y) is calculated for each sample based on its intensity and thresholds applicable to the target’s color. (2) A sum of all c(x, y), Ai, (i = 1, 2, 3, 4) for each of the 4 quadrants Qi (see Fig. 3) is used to generate a location vector with regard to the tool tip for the target. (2)

32 Undergraduate Research at the Swanson School of Engineering

At this stage, we use a threshold on the size of the object in the frame to determine whether the object is close enough to the tool for motion classification to commence. Once the running program determines imminent contact of the tool and object, the above-described operation is performed in every two consecutive frames (thus operating at the camera’s frame rate), and by comparing the 2 location vectors obtained, we can acquire an immediate motion vector for the target. According to the motion vector calculated, the appropriate tactor sequence is delivered to instruct corrective maneuvers according to Table I.

Fiure 3. A demonstration of our immediate motion classification method. Here the target (foam cherry) rolls off to the left after contact with the tool (spoon). Appropriate motion will be twisting hand to the left immediately. Motion Category

Corrective Maneuver

Vibrator Sequence

Miss without Contact

Moving hand backward

Vibrator opposite of the last suggested direction: Vibrate Twice

Bumped Forward

Moving hand forward

Vibrator in the last suggested direction: Vibrate Twice

Roll off to the Right

Twisting hand to the right until feeling contact with the target again

Clockwise sequence

Roll off to the Left

Twisting hand to the left until feeling contact with the target again

Counter-clockwise sequence

Table I. Target motion categorization, corrective maneuver and corresponding tactor sequence.

Ingenium 2021

3. Results 3.1 Discrimination between 4 Tactors

Three runs of 40 tests each were completed on the same trained participant, one for each tactor duration, and confusion matrices were constructed, as shown in Table 2, 3 & 4. A signal duration of 90 ms yielded the minimum overall error rate, 5% (2 erroneous indications over 40 samples).

3.2 Algorithm Parameters for Tool Detection

We completed 18 trials with 9 parameter combinations and 2 holding postures. Error S was calculated as the Pythagorean distance in pixel coordinates between the actual location of the tool tip and its detected location. Means and standard deviations for S over 10-second intervals (5 iterations per second, 50 samples in total) were calculated and the number of frames with successful tool detection N was recorded. Fig. 4 shows that the combination of binary threshold value 13 and kernel size 11 yielded the least average error distance (68.4 and 23.4 pixel), as well as detection of the tool in all 50 samples under both of the posture estimations.

Table 2. Confusion Matrix of Signal Duration 60ms. The overall error rate is 27.5%.

Table 3. Confusion Matrix of Signal Duration 90ms. The overall error rate is 5% (lowest among the three sets).

Table 4. Confusion Matrix of Signal Duration 120ms The overall error rate is 10%

Figure 4. Average Error Distance and its standard deviation over 50 samples are shown for the each of the 2 posture estimations (tool tip to camera distance) and 9 parameter settings. The 6 trials with threshold value of 1 all yielded no valid detection (N = 0), and thus is not displayed in the charts. Threshold value 13 and kernel size 11 yielded the lowest average error while succeeded in all 50 attempts of detection in both posture estimations.


4. Discussion

The experiments presented here demonstrate the system’s promise to provide guidance for the visually impaired of a finger-mounted, computer vision based, assistive device. The tactor localization experiment provides a means of determining the optimal signal duration of the device for individual users to transmit tactile information with satisfactory accuracy, although the hardware’s overall reliability still needs to be proven through experiments with multiple participants. Tool detection experiments yielded an error distance of 68.4 and 23.4 pixels in a 640 x 480 camera frame given optimal setting of algorithm parameters. Such performance is expected to be independent of utensil choice or holding posture, as long as the tip of the tool is present in the camera’s view. This component, when applied alone, showed vulnerability when exposed to a completely homogenous background, since a changing background is a premise of its basic function, although such errors could be reduced by clustering to eliminate momentary errors in tool tip location. We selected plastic utensils for the experiments as the shifting reflection under bright lighting on silverwares can be detected as motion, thus hindering the tool detection process. Although one subject is insufficient to yield statistical significance, our tactor discrimination experiment yielded an error rate of 5% with signal duration of 90ms, whereas Satpute et al. reported 11.2%. Anecdotal results, obtained by attempting to scoop up a plastic cherry with a spoon while blindfolded show success in reaching the target and in responding to corrective instructions immediately after making contact. At its current stage, our system cannot determine the exact location of objects in 3D space, as localization and guidance are computed from 2D information obtained by single camera. Depth information could be obtained by using stereo cameras, as in the “PalmSight” system [4], but this would be difficult to fit onto a single finger. More extensive experiments on the reliability of the motion classification algorithm and the FingerSight device in general are yet to be conducted, as a result of limited participant recruitment due to the Covid-19 pandemic.

5. Conclusion

We have made significant progress towards adding new functions and hardware designs to the previous FingerSight paradigm of guidance of manual interaction and navigation for the visually impaired in real-life scenarios. Experiment results were used determine optimal device configuration or algorithm parameters for tool detection and tactor localization by comparing quantifiable errors. Future research will include designing formal experiments to demonstrate the efficacy of our current prototype and developing more sophisticated algorithms for target detection and localization in 3D space.

34 Undergraduate Research at the Swanson School of Engineering

6. Acknowledgement:

Funding from the Center for Medical Innovation and the Swanson School of Engineering Summer Undergraduate Research Internship (SURI) program at the University of Pittsburgh.

7. References

[1] R. Varma et al., “Visual impairment and blindness in adults in the united states,” JAMA Ophthalmology, vol. 134, no. 7, pp. 802–809, 2016. [2] Young Hoon Lee, Gérard Medioni, “RGB-D camera based wearable navigation system for the visually impaired” , Computer Vision and Image Understanding, Volume 149, 2016, Pages 3-20. [3] Juan Manuel Saez Martinez, Francisco Escolano Ruiz. Stereo-based Aerial Obstacle Detection for the Visually Impaired. Workshop on Computer Vision Applications for the Visually Impaired, James Coughlan and Roberto Manduchi, Oct 2008, Marseille, France. inria-00325455 [4] Z. Yu, S. Horvath, A. Delazio, J. Wang, R. Almasi, R. Klatzky, J. Galeotti, G. Stetten “PalmSight: An assistive technology helping the blind to locate and grasp objects,” Robot. Inst., Carnegie Mellon Univ., Pittsburgh, PA, USA, Tech. Rep., CMU-RI-TR-16-59, Dec. 2016. [5] S. A. Satpute, J. R. Canady, R. L. Klatzky and G. D. Stetten, “FingerSight: A Vibrotactile Wearable Ring for Assistance With Locating and Reaching Objects in Peripersonal Space,” in IEEE Transactions on Haptics, vol. 13, no. 2, pp. 325-333, 1 April-June 2020, doi: 10.1109/ TOH.2019.2945561.

Ingenium 2021

Immunomodulators and macrophages in endometriosis pathogenesis and progression Hilda Jafaraha, Isabelle Chickanoskyb, Alexis Nolfia,c, Mangesh Kulkarnia,c, Clint Skillena,c, Bryan Browna,c a c

Department of Bioengineering, bCarnegie Mellon University, McGowan Institute of Regenerative Medicine

Hilda Jafarah

Hilda Jafarah is a senior bioengineering student who was born and raised in Jeddah, Saudi Arabia. She’s motivated to combine her passion for immunomodulation, tissue engineering, and drug delivery to advance medicine especially in women’s health. After graduation, she aspires to pursue an Md/PhD.

Dr. Bryan Brown is an Associate Professor in the Department of Bioengineering with secondary appointments in the Department of Obstetrics, Gynecology, and Reproductive Sciences and the Clinical and Translational Science Institute at the University of Pittsburgh. He is also Bryan Brown, Ph.D. a core faculty member of the McGowan Institute for Regenerative Medicine where he serves as the Director of Educational Outreach. Dr. Brown is also an Adjunct Assistant Professor of Clinical Sciences at the Cornell University College of Veterinary Medicine and Chief Technology Officer of Renerva, LLC, a Pittsburgh-based start-up company.

Significance Statement

Endometriosis, affecting 200 million women worldwide, causes chronic pelvic pain, infertility, and increases the risk of cancer. This research found that endometriosis development is caused by dysregulated macrophage involvement and subsequent cytokine cascade. This knowledge opens doors to understanding disease identification, potential disease biomarkers, and immunotherapies for diagnosis and treatment.

Category: Review Paper

Keywords: Endometriosis, Cytokines, Macrophages, Pathophysiology, Immune response


A comprehensive literature review was conducted focusing on recent findings on the topic of the pathogenesis of endometriosis which is characterized by the ectopic growth of endometrial cells within the peritoneal cavity. Women with endometriosis display a dysfunctional innate and adaptive immune response where immune cells including natural killer cells (NK cells), dendritic cells (DC cells), cytotoxic T cells (T cells), and, specifically, macrophages behave differently. An overpowering imbalance of immunosuppressive factors play a role in the immunoescape of endometrial cells, and the upregulation in angiogenic and neurogenic factors promote the formation and growth of ectopic lesions. With the lack of efficient models, the pathogenesis of endometriosis remains to be unclear; however, current research emphasizes the role of a dysfunctional immune response to refluxed endometrial cells in women with endometriosis including the differential response of immune cells and the expression of various hormones, immunomodulatory cytokines, growth factors, prostaglandins, and angiogenic and neurogenic factors.

1. Introduction

Endometriosis (EMS) is a gynecological disease characterized by the endometrial lining binding outside of the uterine cavity, forming “lesions”. Typically lesions form on the lining of the pelvic cavity (peritoneum) or the organs of the cavity (e.g., ovaries). It is associated with chronic pelvic pain, infertility, and fatigue. EMS affects about 1015% of women of reproductive age, up to 50% of infertile women, and is prevalent in 71–97% of women with chronic pelvic pain [1]; meanwhile, diagnosis is usually delayed by an average of 10 years from the onset of symptoms. Sampson’s theory of retrograde menstruation, defined as the pathogenic reflux of endometrial cells during menses through the fallopian tubes, is commonly accepted to explain the mechanism through which endometrial tissue travels into the peritoneal cavity; however, it fails to explain the discrepancy between the 76-90% of women who experience retrograde menstruation and the 10-15% of women affected by EMS [1]. This literature review allows for a comprehensive understanding of immune malfunction in macrophage and cytokine recruitment in EMS. The macrophage is a unique subset of monocytes with vital roles in both innate and adaptive immunity as it can differentiate into a wide variety of phenotypes ranging from the two extremes: M1 and M2. Pro-inflammatory, “M1-like,” macrophages perpetuate inflammation, secrete pro-inflammatory signaling molecules, and destroy tissues while anti-inflammatory, “M2-like,” macrophages aid in tissue healing and restorative processes. Some studies propose that M1-like macrophages initiate EMS, while others suggest that a localized tissue-level M2-like macrophage population enhances the growth and development of ectopic lesions [2,3]. The hormonal and immune involvement of M1-like and M2-like macrophages during the pathogenesis of EMS recruit various immunomodulators to the peritoneal cavity and result in a cytokine 35

cascade. An imbalance of immunomodulators and a dysfunctional immune response could aid in the immunoescape, the evasion of phagocytosis or immune triggered apoptosis, of endometrial cells. In this literature review, the immune dysfunction in EMS was outlined to illustrate its involvement in the pathogenesis of the disease. This work opens doors to understanding stage-specific disease identification, potential biomarkers of the disease, and further immunotherapies that could replace laparoscopic surgery as the gold standard of EMS diagnosis and treatment.

2. Methods

Using various search engines and data bases such as PubMed and ScienceDirect, a comprehensive literature review was conducted focusing on 120 peer-reviewed EMS research papers published from 1997-2020. This range of research included the disease pathogenesis, immune cell involvement in EMS, and the role of key biomolecules involved in immunomodulation. Qualitative figures were made illustrating the role of immunomodulators at each stage of pathogenesis based on the results of the 120 papers studied.

3. Endometriosis and the Immune System 3.1 Macrophages in Endometriosis

Women with EMS display a dysfunctional innate and adaptive immune response where immune cells including natural killer cells, dendritic cells, cytotoxic T cells, and macrophages behave differently. Among the dysfunctional immune responses, macrophages are the most critical in EMS. In fact, EMS has been termed a “disease of the macrophage.” Macrophages contribute to the pathogenesis of

EMS by creating an environment that favors implantation, promotion of the formation, growth, and development of lesions, and augmentation of pain sensitivity and inflammation. In EMS patients, macrophages display various dysfunctions including a loss of scavenging ability [4] and an obstruction of phagocytosis through the upregulation of CD200 [5] while continuing the release of inflammatory, oxidative, and nociceptive mediators. This augments the immunoescape and implantation of endometrial cells and inflammation. Furthermore, macrophages in EMS abnormally downregulate IL-24 in the endometriotic milieu causing an increase in the invasiveness and proliferation of endometrial stromal cells through various cytokines [6]. In addition to dysfunctional macrophages, EMS patients present a phenotypic imbalance where the concentration of M2-like to M1-like ratio increases through the Smad2/ Smad3 pathway, and macrophages experience an M1 to M2 polarization through various upregulated cytokines, including IL-17A, which can mobilize granulocytes and stimulate the production of cox-2, in the endometriotic milieu [7, 8, 9]. Therefore, there is a deviation towards M2-like polarization which results in the secretion of growth factors and angiogenic factors, such as VEGF and TGF, aiding in endometriosis grafting, development, and maintenance. An M1 to M2 transition results in the functional shift from an acute inflammatory response to the promotion of tissue repair and regeneration including extracellular matrix synthesis and angiogenesis. Finally, macrophages in endometriosis also promote neurogenesis and sensitization [10], and it was shown that macrophage depletion alleviates abnormal pain in mice models with induced EMS.

Figure 1. In EMS, macrophages (MO) experience various dysfunctions including the loss of scavenging ability (A) due to the lack of adherence to the peritoneal cavity. In MO and endometrial stromal cells co-cultures, MO upregulate IL-10 and TGF-𝜷 which inhibits the cytotoxicity of NK cells permitting the immunoescape of refluxed endometrial cells (B). (C)Finally, ectopic lesions upregulate estrogen which directly (black) affects the MO resulting in an increase in the production of MMP’s and VEGF and indirectly (red) limits the phagocytosis ability of MO’s by inducing an increase of CD200 production in ESC’s. Both promote the growth of lesions.

36 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

3.2 The Role of Immunomodulators in Endometriosis Pathogenesis

An overpowering imbalance of immunomodulators including cytokines, matrix metalloproteinases, and growth factors are involved in the progression of EMS. The development of EMS is multifaceted and can be simplified into seven different factors or stages: the priming of the microenvironment, the initial lesion formation, a dysregulated immune response, the evasion of apoptosis, the promotion of angiogenesis, the continuation of cell proliferation, and the induction of chronic inflammation. The involvement of cytokines in each of these stages shown in Figure 2, as well as the immune malfunction, facilitates the establishment and development of ectopic lesions. During the onset of EMS, an abnormal priming of the microenvironment allows for the implantation and survival of the initial endometriosis lesion. The priming takes place through various hormones and cytokines which continue to be involved in later stages of EMS. The initial endometriotic lesion forms shortly after retrograde menstruation occurs. As endometrial fragments, including endometrial cells characterized by plasticity and cells with reduced differentiation status, begin to leave the uterus through the fallopian tubes and ovaries, specific immunomodulators promote the initial lesion formation. The endometriotic lesion formed has an altered estrogen signal and progesterone resistance. This causes genetic and epigenetic changes, an altered immune response and immunoescape, and aberrantly activated signaling pathways [23]. During the priming of the microenvironment and the formation of the initial lesion, many matrix metalloproteinases (MMPs) were shown to be elevated including MMP-2, which can promote the breakdown of the basement membrane in eutopic endometrium as well as aid in the adhesion and angiogenesis in the ectopic endometrium [11]. In addition to MMPs, interleukins including IL-15, IL-8, and IL-6 are also involved in this stage [14-16]. Most importantly, IL-8 acts as an autocrine growth factor in the endometriotic milieu [15].

Retrograde menstruation occurs in 90% of menstruators, though only 10-15% of these patients develop EMS. The discrepancy between the population of patients who experience retrograde menstruation and those who develop EMs occurs because of the dysfunctional immune response to the initial lesion formation. The dysfunctional immune response results in lesion growth, neurogenesis, inflammation, and angiogenesis. This stage is carried out by a plethora of immunomodulators including MMPs, EOTAXIN, growth factors, interferons, and interleukins. This dysfunctional immune response initiates the evasion of apoptosis, vascularization, increased proliferation, and the chronic inflammation. Apoptosis is a programmed cell death mechanism that acts as a safety measure and is induced by local immunomodulators. In EMS, the dysfunctional immune response results in the evasion of apoptosis. This step is crucial to the development of EMS, and many hypothesize that it could be the cause of the statistical discrepancy mentioned before. Some important immunomodulators inhibiting the apoptosis of endometrial cells include IP-10, EGF, and IFN-g [19,20,22]. This reflects the parallelism between ectopic endometrial cells and tumorigenic cells. Vascularization is crucial for the sustainability of the invasive lesions. Without blood flow and oxygenation, endometriotic cells cannot survive in the peritoneal environment. Current research has shown a significant elevation in angiogenic factors most importantly VEGF [25]. IL-15 also induces angiogenesis through the elevation of VEGF secretion [14]. Through the immune dysfunction, endometriotic cells continue to proliferate and create more lesions aided by angiogenesis and the evasion of apoptosis. During this step of pathogenesis, the lesions spread creating more severe stages of disease development such as Stages III and IV or Deep Infiltrating EMS (DIE) as categorized by the American Society of Reproductive Medicine. Many of the immunomodulators involved in this step of pathogenesis are remnants from the previous steps including IL-15 [14] and EGF [20]. One unique immunomodulator is IL-1b which has been shown to increase the proliferation and migration of endometriotic cells as well as the induction of VEGF through the COX-2 cascade [24,25].


Figure 2. Endometriosis pathogenesis begins with a primed peritoneal microenvironment which supports the implantation of ectopic endometrial stromal cells during initial lesion formation. The lesion causes a dysregulated immune response, resulting in the evasion of apoptosis and increased angiogenesis. This begins the cycle of survival for the endometriotic tissue as the malfunctioning of the immune system prevents the removal of the harmful tissue. With access to vasculature, the lesion promotes cell proliferation and spreads throughout the peritoneal cavity. This spread leads to chronic inflammation which continues for the duration of disease presence.

Chronic inflammation in the peritoneum is a key indicator of EMS. Even though inflammation exists throughout the development of EMS, the chronic inflammation which occurs illustrates the systemic role that inflammation plays in the continuation of endometriotic growth and the preparation of the surrounding microenvironment for other lesion development.

4. Clinical Significance

Despite the prevalence of EMS, there is little known about the involvement of the immune system and the general pathogenesis of the disease. Understanding the role of macrophages and immunomodulators in the progression of EMS can aid in the development of early detection and possibly less invasive treatments. An effective pre-clinical study proved that trichostatin A had anti-proliferative activity on endometrial stromal cells by reducing the expression of COX-2, which is elevated due to high expression of IL-1b and other inflammatory cytokines, and its administration in mice significantly decreased the size of endometriotic implants [26]. Additionally, human recombinant TNFa antagonists, TNFRSF1A and c5N, demonstrated inhibitory activity on endometriotic lesions in baboons without affecting their menstrual cycle [27, 28]. Finally, this knowledge can help in the development of in vitro model systems of EMS for further investigation of the disease pathogenesis as well as for drug testing.

38 Undergraduate Research at the Swanson School of Engineering

5. Conclusion

The pathogenesis of EMS remains to be unclear due to the lack of models; however, current research emphasizes the role of a dysfunctional immune response to retrograde menstruation in EMS patients. Moreover, the aberrant expression of hormones, immune cells, immunomodulatory cytokines, growth factors, prostaglandins, and angiogenic and neurogenic factors contribute to the development of EMS. Understanding the immunomodulators and their trends in the stages of EMS development could allow for better disease detection and assessment as well as immunotherapeutic treatments. However, the question remains whether the differential immune response results from the refluxed endometrial cells and formed lesions or is the differential immune response innate to EMS patients.

6. Acknowledgments

This work was funded by the Swanson School of Engineering, the Office of the Provost, the Department of Bioengineering and was conducted under the mentoring of Dr. Brown and Alexis Nolfi with the help of Dr. Kulkarni and Clint Skillen.

Ingenium 2021

7. References

[1] C. Hogg, A. Horne, and E. Greaves, Endometriosis-Associated Macrophages: Origin, Phenotype, and Function, Front. Endocrinol. 11 (2020). [2] P. Murray and T. Wynn, Protective and pathogenic functions of macrophage subsets, Nat. Rev. Immunol. 11 (2011) 723-737. [3] A. Mantovani, A. Sica, S. Sozzani, P. Allavena, A. Vecchi, M. Locati, The chemokine system in diverse forms of macrophage activation and polarization, Trends Immunol. 25 (2004) 677-86. [4] S. Brunty, N. Santanam, Current assessment of the (dys)function of macrophages in endometriosis and its associated pain, Ann. Transl. Med. 7 (2019) S381-S381. [5] L. Weng, S. Hou, S. Lei, H. Peng, M. Li, D. Zhao, Estrogen-regulated CD200 inhibits macrophage phagocytosis in endometriosis, J. Reprod. Immunol. 138 (2020). [6] H. Yang, W. Zhou, K. Chang, J. Mei, L. Huang, M. Wang, The crosstalk between endometrial stromal cells and macrophages impairs cytotoxicity of NK cells in endometriosis by secreting IL-10 and TGF-beta, Reproduction 154 (2017) 815-25. [7] M. Nie, Q. Xie, Y. Wu, Serum and Ectopic Endometrium from Women with Endometriosis Modulate Macrophage M1/M2 Polarization via the Smad2/Smad3 Pathway, J. Immunol. Res. (2018) 1-14. [8] M.Johan, W. Ingman, S. Robertson, M. Hull, Macrophages infiltrating endometriosis-like lesions exhibit progressive phenotype changes in a heterologous mouse model, J. Repro. Immunol. 132(2019). [9] J. Miller, S. Ahn, R. Marks, IL-17A Modulates Peritoneal Macrophage Recruitment and M2 Polarization in Endometriosis, Fron. Immunol. 11 (2020). [10] E. Greaves, J. Temp, A. Esnal-Zufiurre, S. Mechsner, A. Horne, P. Saunders, Estradiol is a critical mediator of macrophage-nerve cross talk in peritoneal endometriosis, Am. J. Pathol.185 (2015). [11] X. Sui, Y. Li, Y. Sun, C. Li, G. Zhang, Expression and significance of autophagy genes LC3, Beclin1 and MMP-2 in endometriosis, Exp. Therap. Med. 16(3) (2018) 1958-62. [12] H. Liu, J. Wang, H. Wang, N. Tang, Y. Li Y. Zhang, T. Hao, Correlation between matrix metalloproteinase-9 and endometriosis, Int. J. Clin. Exp. Pathol. 8(10) (2015) 13399-404. [13] A. Luddi, C. Morocco, L. Governini, B. Semplici, V. Pavone, S. Luisi, F. Petraglia, P. Piomboni, Expression of Matrix Metalloproteinases and Their Inhibitors in Endometrium: High Levels in Endometriotic Lesions, Int. J. Mol. Sci. 21(8) (2020) 2840. [14] A. Arici, I. Matalliotakis, A. Goumenou, G. Koumantakis, S. Vassiliadis, B. Selam, N. Mahutte, Increased levels of interleukin-15 in the peritoneal fluid of women with endometriosis: inverse correlation with stage and depth of invasion. Hum Reprod. 18(2) (2003) 429-32. [15] R. Rozati, Identification of disease specific proteins associated with endometriosis in Indian women, Int. J. Sci. Res. 8(9) (2019).

[16] S. Li, X. Fu, T Wu, L. Yang, C. Hu, R. Wu, Role of Interleukin-6 and Its Receptor in Endometriosis, Med. Sci. Monit. 23 (2017) 3801-3807. [17] J. Miller, S. Monsanto, S. Ahn, K. Khalaj, A. Fazleabas, S. Young, B. Lessey, M. Koti, and C. Tayade, Interleukin-33 modulates inflammation in endometriosis. Sci. Rep. 7 (2017) 17903. [18] S. Monsanto, A. Edwards, J. Zhou, P. Nagarkatti, M. Nagarkatti, S. Young, B. Lessey, C. Tayade, Surgical removal of endometriotic lesions alters local and systemic proinflammatory cytokines in endometriosis patients, Fertil. Steril. 105(4) (2016): 968-977.e5. [19] E. Othman, D. Hornung, H. Salem, E. Khalifa, T. El-Metwally, A. Al-Hendy, Serum cytokines as biomarkers for nonsurgical prediction of endometriosis, Eur. J. Obstet. Gynecol. Reprod. Biol. 137(2) (2008) 240–246. [20] Y. Fan, H. Chen, W. Chen, Y. Liy, Y. Fu, and L. Wang, Expression of inflammatory cytokines in serum and peritoneal fluid from patients with different stages of endometriosis, Gynecol. Endocrinol. 34(6) (2018) 507-512. [21] M. Inagaki, S. Yoshida, S. Kennedy, N. Takemura, M. Deguchi, N. Ohara, T. Maruo, Association study between epidermal growth factor receptor and epidermal growth factor polymorphisms and endometriosis in a Japanese population, Gynecol. Endocrinol. 23(8)(2007). [22] A. Delbandi, M. Mahmoudi, A. Shervin, S. Heidari, R. Kolahdouz-Mohammadi, A. Zarnani, Evaluation of apoptosis and angiogenesis in ectopic and eutopic stromal cells of patients with endometriosis compared to non-endometriotic controls, BMC Women’s Health 20 (2020). [23] P. Klemmt and A. Starzinski-Powitz, Molecular and Cellular Pathogenesis of Endometriosis, Current Women s Health Reviews. 14(2)(2018) [24] F. Huang, J. Cao, Q. Liu,Y. Zou, H. Li, T. Yin, MAPK/ERK signal pathway involved expression of COX-2 and VEGF by IL-1β induced in human endometriosis stromal cells in vitro, Int J Clin Exp Pathol, 6(10)(2013) [25] S. Matsuzaki, J. Pouly, M. Canis, Dose-dependent pro- or anti-fibrotic responses of endometriotic stromal cells to interleukin-1β and tumor necrosis factor α, Sci Rep. 10(1)(2020) [26] S. Ferrero, G. Evangelisti and F. Barra, Current and emerging treatment options for endometriosis, Expert Opinion on Pharmacotherap.19 (2018) [27] T. D’Hooghe, N. Nugent, S. Cuneo, et al. Recombinant human TNFRSF1A (r-hTBP1) inhibits the development of endometriosis in baboons: a prospective, randomized, placebo- and drug-controlled study, Biol Reprod. 74(1)(2006) [28] H. Falconer, J. Mwenda, D. Chai, et al. Treatment with anti-TNF monoclonal antibody (c5N) reduces the extent of induced endometriosis in the baboon, Hum Reprod. 21(7)(2006)


Testing the compressive stiffness of endovascular devices Lucy Jonesa, Dr. Jonathan Vande Geestb a

Soft Tissue Biomechanics Laboratory, bDepartment of Bioengineering

Lucy Jones

Lucy Jones is a senior bioengineering student planning on graduating in the spring. Her interest in biomechanics motivated her to research with Dr. Jonathan Vande Geest at the Soft Tissue Biomechanics Laboratory for nearly two years. She has plans to work for Epic Systems following her undergraduate career.

Dr. Jonathan Vande Geest is a Professor in the Department of Bioengineering, Department of Mechanical Engineering and Material Science, the Department of Ophthalmology, the McGowan Institute for Regenerative Medicine, the Louis J. Fox Center for Vision Restoration, and the Vascular Medicine Institute at the Jonathan Vande University of Pittsburgh. He received his Geest, Ph.D. BS in Biomedical Engineering from the University of Iowa in 2000 and his PhD in Bioengineering from the University of Pittsburgh in 2005. Dr. Vande Geest began his career at the University of Arizona in the Department of Aerospace and Mechanical Engineering and joined the U of A’s Department of Biomedical Engineering in 2009. Dr. Vande Geest returned to the University of Pittsburgh in January of 2016.

Significance Statement

Peripheral artery disease (PAD) is a condition that causes the stenosis of arteries in the limbs. In this work, we used a fabricated 3D printed compression apparatus to assess the compressive resistance of several commercially available stents. Eventually, we will test the compressive resistance of our prototype endovascular device.

Category: Device Design

Keywords: endovascular device, compressive strength, peripheral artery disease, stent

40 Undergraduate Research at the Swanson School of Engineering


Peripheral Artery Disease (PAD) is a condition characterized by the stenosis of arteries in the limbs. It can lead to negative consequences ranging from numbness, to amputation, or even death. Treatment methods for PAD include surgical intervention, or angioplasty, and stent implantation. However, these treatment methods are known to fail at high rates and could cause infection, vascular injury, or the restenosis of the artery. As a result, our team has created a new endovascular device that can exert force on the artery walls in order to keep it open while also maintaining the flexibility necessary to contend with complex mechanical regions such as the knee. In preparation for the assessment of our novel device, we have fabricated a simple 3D printed apparatus to evaluate the compressive resistance of our device and commercially available stents. In this work, we have assessed the compressive resistance of several commercially available stent devices. Our findings show that stents can generally withstand compressive loads up to 400g before they experience severe buckling. In the future, we intend to evaluate the compressive resistance of our endovascular device and compare it to that of stents.

1. Introduction

Peripheral Artery Disease (PAD) is a disease affecting approximately 202 million people worldwide [1]. This condition causes the stenosis of arteries and reduces blood flow to distal tissues. If left untreated, PAD will lead to stroke, amputation, or death. There are several available methods of treatment for this condition which include angioplasty, medication, and endovascular stent or stent-graft implantation. Stent grafts are endovascular devices made of a metal mesh and covered in fabric that following deployment expand the artery to keep it patent. People who experience PAD in their femoropopliteal artery need a solution that will be able to endure the complex mechanical challenges associated with that region of the leg (located behind the knee). Stent graft implementation has become a common treatment for PAD in extremities that experience loading from joint movement. Yet, stent graft failure is incredibly common, especially in the femoropopliteal artery, with an approximate 50% loss of patency two years following implantation [2]. Given the adverse effects of PAD, stent failure can be detrimental to the patients’ health and potentially fatal. As a result of the frequent failure of current stents, there has been an effort to make stents more adaptable to their mechanical environment. For instance, there has been success with the Misago, Supera, and Absolute stents, but the flaw of current stents still seems to be the rigidity of the design that is required to create a supportive scaffold inside stenosed arteries. Since stents are still not wholly effective in the long-term treatment of PAD in complex mechanical regions, our goal is to design a new device that will create a more flexible solution and allow for blood flow through the artery with a reduced risk of restenosis. For this phase in the development of our device, we

Ingenium 2021

are evaluating the compressive resistance of commercially available stents so that we can compare the results to those of our new endovascular device. In order to test the compressive strength of existing endovascular stents, we 3D printed a device, shown in Figure 1, that would allow for a known mass to compress the stents. We determined the masses used in our study by extrapolating values from the minimum “hoop stress” or the “compressive force acting on the stent” in an artery [3]. Our hypothesis is that if a mass above 400g is added to a stent using the compression apparatus that we have developed; the stent will collapse and no longer have functionality.

2. Methods

We started by obtaining 4 Taxus Liberté monorail stents which are commercially available and widely used in the treatment of PAD as well as coronary artery disease. Three of these stents had an initial diameter of 2.5mm and one had an initial diameter of 2.75mm. We then deployed our stents using a device called an indeflator. The indeflator is essentially a large syringe that fills balloon catheters with water, causing them to expand, and conforms the stent that is cinched around the balloon catheter to its cylindrical shape. Once the balloon is deflated again, the stent can be removed and used for compression testing. A total of 6 masses were added to each of the stents: 50g, 100g, 200g, 400g, 600g, and 800g respectively using the device shown in Figure 1. As each mass was added, a 12-megapixel wide-angle camera was used to capture a digital image showing the compression of the device. We then utilized a Matlab image processing program to quantify the diameter of the stents under each given load. The diameters of the stents were then compiled into a graph so that the reduction in diameter, and therefore the compression of the stents under various loads, could be quantified and evaluated.

Figure 1. 3D Printed Compression Device. This device consists of 2 parts: a square top plate with a hole in each corner and a square base plate with 4 cylindrical prongs extending vertically out of the corners. The stent is placed on the base plate while the top plate is lowered onto it by lining the holes up with each of the prongs. Then, the desired weight is positioned on the top plate and the change in stent diameter (distance between top plate and base plate) is measured.

3. Results

We were able to determine the diameter of the stents under various loads ranging up to 800g. Figure 2 shows that all of the stents could withstand loading up to 400g, but at 600g, most of the stents collapsed. Meaning, when a mass of 600g was added to the top plate of our compression device, the stent in between the top and bottom plates was completely flattened. Only one stent was able to stay open under 600g of weight. It’s also worth noting that none of the stents exhibited any drastic (more than 0.5mm) changes in diameter until weights exceeding 200g were added. Meaning, the changes in diameter under weights ranging from 50g to 200g were very slight for all of the stents. However, once that weight exceeded 200g, there is a noticeable reduction in the diameter of all of the stents, ranging from 0.45mm to 1.14mm.

Figure 2. Compression of Stents Under Loading. This graph displays the results of 4 different endovascular stents subjected to varying amounts of compressive loading. It shows that all of them could withstand compressive loading up to 400g. At 600g, all but one stent is fully compressed.


Figure 3a shows Stent 1 prior to compression, while Figure 3b displays that same stent after 600g was added to it using the compression device. These figures demonstrate that this stent has completely lost its cylindrical shape under 600g of loading.

Figure 3a. Stent 1 Initial Diameter. This shows the diameter and initial shape of Stent 1 prior to any loads being added.

Figure 3b. Stent 1 After 600g Loading. This displays Stent 1, fully compressed, after 600g of weight was added.

4. Discussion

This experiment yielded results that were mostly to be expected. We were expecting the stents to slowly deform as more weight was added, and then to collapse under a mass exceeding 400g. This did happen in 3 out of the 4 stents and therefore allowed us to tentatively confirm our hypothesis. However, there were some idiosyncrasies that we did not expect. For example, the drastic reduction in diameter of the stents that occurred when the load exceeded 200g was not something we anticipated. Considering the fact that stents are being placed in arteries that can sometimes be bent at extreme angles, we expected them to display a resistance to intense changes so that in physiological environments, they would not fail after one harsh bend. Since this was not the case, we believe that stents are only designed to be unyielding under certain small increments of weight. Our conjecture is that stents fail at higher rates in more complex mechanical regions because once they begin to buckle, they lose their integrity entirely. This is shown in the contrast between Figures 3a and 3b. Stent 1 had become completely compromised after a 600g weight was added to it. In an in vivo environment, this would mean that the stent could no longer keep the artery walls open and restenosis would occur. This would align with our knowledge of stent failure in the femoropopliteal artery since sudden drastic movement is to be expected in that region.

42 Undergraduate Research at the Swanson School of Engineering

This experiment also helped us test the effectiveness of our 3D printed device. Seeing as we were able to conduct the experiment with believable results, we have determined it to be a functional tool in preliminary compressive testing. Our next task will be to quantify the compressive resistance of our novel endovascular device. Eventually, our goal is to compare the compressive resistance of our novel device to existing stents to see whether there are any significant differences. There are many limitations to our study and how it can be applied in a clinical setting. Firstly, our compression device is a rudimentary tool used to give preliminary results. Stents are usually tested using axial, tensile, and radial compression, as well torsion and bending tests [4]. Our compression device is most similar to the radial compression system, but it is still limited in what it can tell us about the overall mechanical advantages of a stent. For instance, we did not incorporate any torque, internal pressure, or bending to our stents which would all be experienced by a stent in the femoropopliteal artery. Therefore, when we move to testing our endovascular prototype, we will be able to compare its compressive resistance to that of stents, but we will not know how it withstands other forces. As a result, we will not get a comprehensive picture of how effective our prototype is in comparison to stents. This means that after compressive testing, we will need to perform other experiments, preferably in vivo, to determine the efficacy of our novel endovascular device.

5. Conclusions

From our results, we can see that our simple 3D printed device can be utilized to quantify the compression of endovascular devices. We will be using this device to find the compressive resistance of our new prototype endovascular device and comparing the results to those that we have generated from the stents. Our study has clinical implications because once we have the in vitro data to support the claim that our device can withstand more loading than current stents, we will be able to move forward with the in vivo testing of our endovascular device. In vivo testing will yield more realistic results in terms of the actual physiological constraints that endovascular devices face and will help us to modify our design criteria to meet the complex mechanical needs of those arteries. However, given that we are in the preliminary stages of our research, these results are promising and give us a baseline compressive strength that we hope to obtain with our prototype. In conclusion, we can confirm our hypothesis, as our testing has shown that all of the stents can withstand compressive loads up to 400g, beyond which they are prone to collapse.

Ingenium 2021

6. Acknowledgements

This work was completed with the generous funding from the Swanson School of Engineering, the Office of the Provost, the Department of Bioengineering, and the University of Pittsburgh’s Center for Medical Innovation grant number F_274.

7. References

[1] Song P, Rudan D, Zhu Y, Fowkes FJI, Rahimi K, Fowkes FGR, Rudan I. Global, regional, and national prevalence and risk factors for peripheral artery disease in 2015: an updated systematic review and analysis. Lancet Glob Health. 2019 Aug;7(8):e1020-e1030. [2]Strecker EP, Boos IB, Göttmann D. Femoropopliteal artery stent placement: evaluation of long-term success. Radiology. 1997 Nov;205(2):375-83. doi: 10.1148/radiology.205.2.9356617. PMID: 9356617. [3] Cabrera, M. S., Oomens, C. W., & Baaijens, F. P. (2017). Understanding the requirements of self-expandable stents for heart valve replacement: Radial force, hoop force and equilibrium. Journal of the Mechanical Behavior of Biomedical Materials, 68, 252-264. doi:10.1016/j. jmbbm.2017.02.006 [4] Maleckis, Kaspars et al. “Comparison of femoropopliteal artery stents under axial and radial compression, axial tension, bending, and torsion deformations.” Journal of the mechanical behavior of biomedical materials vol. 75 (2017): 160-168. doi:10.1016/j.jmbbm.2017.07.017


Design and manufacture of two experimental test benches for the in vitro assessment of a method to perform in situ fenestration of stent grafts Nicholas P. Lagermana, Timothy K. Chung, Ph.D.a, Mohammad H. Eslami, M.D.b,c, David A. Vorp, Ph.D.a,b,c,d,e,f,g Department of Bioengineering, bDivision of Vascular Surgery, University of Pittsburgh Medical Center, cDepartment of Surgery, d Department of Cardiothoracic Surgery, eDepartment of Chemical and Petroleum Engineering, fMcGowan Institute for Regenerative Medicine, g Center for Vascular Remodeling and Regeneration a

Nicholas Lagerman is a bioengineering student originally from Reading, Pennsylvania. He is continuing in the field of medical device design because of his love for biology and design.

Nicholas P. Lagerman

David A. Vorp, Ph.D.

David A. Vorp, PhD, is the Associate Dean for Research at the Swanson School of Engineering. He is a professor of Bioengineering, with secondary appointments in the Department of Cardiothoracic Surgery, the Department of Chemical and Petroleum Engineering, and the Clinical and Translational Science Institute at the University of Pittsburgh.

Significance Statement

There are two primary methods of abdominal aortic aneurysm (AAA) intervention that minimize risk of rupture: open surgery and endovascular aortic repair (EVAR). Fenestrated endovascular aortic repair (FEVAR) is a specialized EVAR procedure that utilizes a patient-specific stent graft to allow blood flow into visceral arteries. Nearly 80% of AAA repairs consist of EVAR or FEVAR [1] as their respective mortality rates are significantly lower than open surgery. FEVAR costs almost 3 times more than standard EVAR and requires substantial time to intervention because a fenestrated stent graft is custom-made for each patient [2]. The ability to create fenestrations in stent grafts during procedures would likely decrease cost and time to intervention. Our work develops two test benches that aid in the development of a novel device for detecting orifices in AAA geometries and creating in situ fenestrations in stent grafts.

Category: Device Design

Keywords: abdominal aortic aneurysm, test bench, modular perfusion phantom, 3D printing, fiber optic wire, infrared, fenestration. Abbreviations: Abdominal aortic aneurysm (AAA), endovascular aortic repair (EVAR), fenestrated endovascular aortic repair (FEVAR), computer-aided design (CAD) 44 Undergraduate Research at the Swanson School of Engineering


This project focuses on the production of two experimental test benches that will be used in parallel towards the development of a novel in situ fenestration device: 1) An in vitro abdominal aortic aneurysm (AAA) phantom to deploy the sensing technology developed in parallel, and 2) A testing block to determine the feasibility of bending fiber optics for the sensing of orifices in the AAA geometry. This was accomplished by designing and generating AAA models, a perfusion chamber, and a fiber optic test block. Three complex geometries (juxtarenal, pararenal, and suprarenal) were modeled using computer-aided design (CAD) software and 3D printed for the perfusion phantom using an elastic resin to simulate the aortic wall’s elasticity. A perfusion chamber to house the models and the connected flow system was designed to allow for the models to be modular. A testing block consisting of fiber optic wire channels with five bending angles and six radii was designed and 3D-printed for the fiber optic test bench. An infrared emitter and receiver circuit with an analog voltage output were placed at opposite ends of the test block to measure the amount of infrared light being transmitted through a fiber optic wire. Both the bend angle (p = 0.135) and bend radius (p = 0.523) tests demonstrated that there was no significant difference in the voltage output between the variations in angles and radii. There is no evidence that the bend radius or angle would influence signaling in the detection of orifices.

1. Introduction

An abdominal aortic aneurysm (AAA) is an enlargement of the abdominal portion of the aorta. Rupture of the aorta may occur if the untreated aneurysm continues to grow in size, and 60% of people will die before reaching the emergency room [3]. To prevent rupture, clinicians usually intervene with one of two: open surgery and endovascular aortic repair (EVAR). There are close to 30,000 AAA repairs that occur annually, and approximately 80% of AAA interventions are EVAR [1] since it is both less invasive and has a lower mortality rate (1.5% compared to 12%) than open surgery [4]. EVAR is performed by inserting a catheter sheath into the femoral artery through a small incision located near the groin [4]. Then a catheter that houses a radially compressed stent-graft is inserted through the sheath. For standard sized grafts, an aneurysm with a minimum neck size of 10 mm (between the aneurysm and renal arteries) is required to properly deploy and anchor the stent-graft to reinforce the vessel wall, whereby preventing rupture [5]. However, there are AAA with complex geometries (juxtarenal, pararenal, suprarenal) which require a specialized fenestrated stent-graft that is deployed and anchored superior to the renal arteries and requires the re-perfusion of blood into the visceral arteries (renal, superior mesenteric artery and celiac) through the use of branching stent-grafts [2]. Therefore, fenestrated endovascular aortic repair (FEVAR) uses a highly customized patient-specific stent-graft to restore blood flow and provide the anchoring needed to prevent migration. There

Ingenium 2021

is only one FDA approved fenestrated stent-graft on the market, the Cook Zenith graft, which requires a minimum of six weeks to create a patient-specific fenestrated stent graft. A potential catastrophic rupture event [5] could occur in this timeframe that requires emergency intervention. Additionally, FEVAR costs up to 3 more times more than EVAR due to customizations of the stent-graft device and technical procedure [2]. To minimize both intervention time and cost, the development of a novel device to identify visceral arteries and fabricate an in situ fenestrated stent graft is essential. A single AAA perfusion phantom was developed by Kung, et al. for the purpose of replicating flow in a AAA model, however it was not designed to accommodate in vitro surgeries [6]. This project focuses on the design of two separate test benches: 1) An in vitro AAA phantom to deploy the sensing technology to be used in the novel device, and 2) A testing block to determine the feasibility and limitations of bending fiber optics for sensing orifices in AAA geometries.

MA) using Elastic Resin (Formlabs, Somerville, MA, Cat. # RS-F2-ELCL-01), totaling 24 hours per model. The juxtarenal model is shown in Figure 1. The resin allowed for simulation of the elastic properties of the aortic wall. A perfusion chamber to hold the models and connected flow system was designed in parallel. The components of the box were individually designed and assembled by laser cutting 0.25-inch-thick acrylic sheets. The finalized CAD design for the perfusion chamber is shown in Figure 2. A simple qualitative fluid loss test was performed by connecting the superior aorta opening to a steady-flow pump that passes fluid at a flow rate of 3.15 liters per minute (LPM). Unwanted fluid loss was characterized by fluid exiting the model from areas other than the visceral arteries. A pulse-duplicating flow loop was sourced for the final AAA phantom that has physiologic blood pressure (120/80 mmHg), a flow rate of 3.0 (LPM), and is able heat fluid to body temperature (37oC).

2. Methods 2.1. Modular AAA Perfusion Phantom Artery


superior mesenteric artery (SMA)

9 mm [9]

common iliac artery

12 mm (normalized) [10]

external iliac artery

8 mm [10]

internal iliac artery

5.5 mm [10]

renal artery

6 mm [11]

suprarenal aorta

21 mm [13]

infrarenal aorta (affected)

50 mm (juxtarenal, infrarenal) 70 mm (suprarenal)

Table 1: Measurements of diameter in arteries. Measurements found in various journal articles that were used in the modeling of abdominal aortic aneurysms. The infrarenal aorta diameters for each morphology were provided from our clinical expert. Anatomical Landmarks

Distance Between

renal arteries to iliac bifurcation

130 mm [8]

SMA to renal arteries

12.5 mm [9]

iliac bifurcation to internal iliac arteries

45 mm [12]

Figure 1: 3D Printed Juxtarenal Model. (A) The juxtarenal AAA model that was the first test print of the design. (B) Sketch of a juxtarenal AAA for comparison to the 3D printed model.

Table 2: Distances between anatomical landmarks. Average measurements from one anatomical landmark to another that were used in the modeling of abdominal aortic aneurysms of complex geometries.

Measurements derived from various research papers on the dimensions of AAA geometries, shown in Table 1 and Table 2, were used in generating three complex geometry AAA models with a wall thickness of 2.5 mm [7,8,9,10,11,12,13]. The three AAA complex geometries were sketched with the measurements from the included tables and modeled using the CAD software SOLIDWORKS (2019, Dassault Systèmes, Vélizy-Villacoublay, France). The openings of each artery were designed to accommodate commercial fittings to allow for easy connection to tubing. The three models were printed in a Formlabs Form 3 SLA 3D Printer (Formlabs, Somerville,

Figure 2: Modular AAA Perfusion Phantom. The final CAD assembly of the perfusion chamber with AAA model to demonstrate positioning.


2.2. Fiber Optic Wire Bend Testing

A testing block, shown in Figure 3, consisting of 1.2 mm diameter wire channels with five bend angles and six bend radii was modeled and printed using non-elastic Formlabs Grey V4 resin. The test block was redesigned using a previous iteration that was developed by a senior design team from the University of Pittsburgh and the pre-existing bend angles and radii were preserved. The following angles and radii were chosen as design constraints because they represent a range of values that can be accommodated in the aorta during deployment of a stent graft. The angles chosen for the testing block were 45°, 60°, 70°, 80°, and 90°, respectively, with a fixed radius of 2 cm. The radii chosen for the testing block were 0.8 cm, 1.2 cm, 1.6 cm, 2.0 cm, 2.4 cm, and 2.8 cm, respectively, with a fixed angle of 90°.

Infrared light was chosen as the wavelength used for sensing; therefore, two separate circuits were developed using an Arduino MEGA (Somerville, MA) microcontroller to send a signal via an infrared emitter and record the signal through a fiber optic wire with a complementary infrared receiver. The circuits were placed at opposite ends of a custom 3D-printed LED holder, shown in Figure 4, consisting of 15 cm of 0.50 Numerical Aperture 400 µm Diameter Core Multimode fiber optic wire (Thorlabs, Newton, NJ) run through the test block channels. A phototransistor (receiver) was attached to the microcontroller to read a continuous stream of analog values that represented voltage changes resulting from intensity changes of the infrared signal received throughout testing. Voltage was the dependent variable because the voltage output of the phototransistor circuit is proportional to the amount of infrared light it receives. A script was coded in MATLAB (2018, MathWorks, Natick, MA) to record the voltage of the receiver circuit. The recorded voltage from signaling periods of 10 seconds was averaged for each bend angle and radius.

3. Results 3.1. Construction of the Modular AAA Perfusion Phantom

Figure 3: Fiber Optic Bend Test Block. (A) The bend angle side of the test block with included angles of 45, 60, 70, 80 and 90 degrees. (B) The bend radius side of the test block with included radii of 0.8, 1.2, 1.6, 2.0 and 2.4 cm.

Figure 4: Fiber Optic Bend Testing Setup. Two circuits placed at opposite ends of a fiber optic wire held in place by the test block and an LED holder.

46 Undergraduate Research at the Swanson School of Engineering

An in vitro model of an abdominal aorta affected by a complex geometry aneurysm was successfully designed and manufactured. A qualitative fluid loss test demonstrated that the model allowed fluid to pass through continuously without any fluid loss. Additionally, the perfusion chamber that houses the models and flow system was successfully designed and constructed.

Ingenium 2021

3.2. Fiber Optic Wire Bend Tests

The data acquired during experimentation for each fiber optic wire bend angle and radius is a result of the average of three ten-second continuous voltage readings. Figure 5 displays the voltage data recorded for the various bend angles, and Figure 6 shows the voltage data recorded for the various bend radii. Fiber optic bundles anchored to catheters will be used to detect a point source in the visceral arteries to determine the location of the orifice requiring a fenestration. Two separate one-way analysis of variance (ANOVA) tests were used to determine if there was any significant difference between the means since there were more than two independent groups (five angle groups and six radius groups). One ANOVA was applied to the bend radius dataset and showed no significant difference (p = 0.523). The second ANOVA was applied to the bend angle dataset and showed no significant difference (p = 0.135).

Figure 5: Fiber Optic Bend Angle Data. The voltage outputs of each bend angle plotted with error and a linear trendline to show that there was no change over time. A p-value of 0.523 was calculated for this data.

4. Discussion

The AAA model demonstrated the ability to allow fluid to pass through it without losing any fluid through its walls. This is critical in maintaining continuous flow in the perfusion phantom since that is the point where loss is most likely to occur. This project will progress with attaching tubing to the openings of the model with fittings that allow for modularity and producing flow through the system with a continuous flow pump. Additional flow tests will be conducted with the final perfusion system to determine if any fluid loss occurs with the addition of more components. The results from the fiber optic wire bend tests (Figure 5 and Figure 6) show that there is no significant difference in the voltage measured while varying the bend angle or radius. We further demonstrated that there is no evidence the bending radius or angle would influence the signal received to detect an orifice. These results suggest that the use of infrared in two fiber optic cables, one for emission and the other for reception, to detect orifices is feasible for use in the novel fenestration device. There is currently no commercially available device that can be used to perform in situ fenestrations in the abdominal aorta [14]. A paper by Le Houérou, et al. provides a method using lasers guided by fluoroscopic imaging to generate in situ fenestrations during pararenal AAA repair procedures [14]. Our work improves upon the work of Le Houérou, et al. because we aimed to develop a method for visceral artery detection without the need for radioactive contrast agents used in fluoroscopic imaging. Additionally, Le Houérou, et al tests the effectiveness of the device on patients, while the AAA phantom we generated allows for use of the device with zero risk to patients. This research is limited because necessary equipment for test bench construction and data collection was not readily available due to a pandemic, resulting in the inability to gather additional numerical characterizations, such as full dimensioning of and flow rates through the phantom. The process of designing and testing the novel fenestration device can further progress past the initial fiber optic tests with functional test benches.

5. Conclusions

Figure 6: Fiber Optic Bend Radius Data. The voltage outputs of each bend radius plotted with error and a linear trendline to show that there was no change over time. A p-value of 0.135 was calculated for this data.

The results from the fiber optic bend angle and bend radius test show that there is no significant difference in the voltage outputs of either bend test. Therefore, infrared signaling through fiber optic wire could be used in the orifice detection component of the in situ fenestration device. Future development will incorporate fiber optic bundles to detect a point source in the visceral arteries to ensure proper placement of fenestrations. The future work that our laboratory intends to pursue includes the completion of AAA perfusion phantom that is connected to a pulse-duplicating flow loop, assessment of the effect of blood on bend testing, and development of a novel in situ fenestration device. After a device is fully developed, the time and cost of intervention on complex geometry AAAs will significantly decrease as in situ fabrication of a fenestrated stent graft will be achieved during a single procedure.


6. Acknowledgements

Thank you to the Swanson School of Engineering, the Office of the Provost, and the Department of Bioengineering for funding this research.

7. References

[1] A. Melillo, J. Trani, J. Gaughan, J. Carpenter, J. Lombardi, Assessing trends, morbidity, and mortality in ruptured abdominal aortic aneurysm repair with 9 years of data from the National Surgical Quality Improvement Program, Journal of Vascular Surgery. 71(2) (2020) 423–431. [2] H. Graves, B. Jackson, The Current State of Fenestrated and Branched Devices for Abdominal Aortic Aneurysm Repair, Seminars in Intervent Radiology. 32(03) (2015) 304–310. [3] L. Robertson, S. Nandhra, Laparoscopic surgery for abdominal aortic aneurysm, Cochrane Database of Systematic Reviews. 7 (2016). [4] R. Greenhalgh, Comparison of endovascular aneurysm repair with open repair in patients with abdominal aortic aneurysm (EVAR trial 1), 30-day operative mortality results: randomised controlled trial, The Lancet. 364(9437) (2004) 843–848. [5] N. J. Swerdlow, W. W. Wu, M. L. Schermerhorn, Open and Endovascular Management of Aortic Aneurysms, Circulation Research. 124(4) (2019) 647–661. [6] E. O. Kung, A. S. Les, F. Medina, R. B. Wicker, M. V. McConnell, C. A. Taylor, In Vitro Validation of Finite-Element Model of AAA Hemodynamics Incorporating Realistic Outlet Boundary Conditions, Journal of Biomechanical Engineering. 133(4) (2011).

48 Undergraduate Research at the Swanson School of Engineering

[7] K. van den Hengel, Variation in wall thickness in healthy AA and AAA’s, in: Abdominal aortic wall thickness and compliance. (2008) 6–7. [8] M. Kapil dev, V. Vimala, Morphometric Analysis of Distance of Origin of Renal Artery from Bifurcation of Abdominal Aorta for Interventional Procedures, Journal of Dental and Medical Sciences. 18(4) (2019) 66–68. [9] Part I: Anatomy, Acta Radiologica: Diagnosis. 11(s306) (1965) 9–71. [10] İ. Selçuk, B. Uzuner, E. Boduç, Y. Baykuş, B. Akar, T. Güngör, Step-by-step ligation of the internal iliac artery, Journal of the Turkish-German Gynecol Association. 20(2) (2019) 123–128. [11] M. Mohiuddin, A. Mansoor, M. Ali, N. Hassan, Analysis of renal artery morphometery in adults: A study conducted by using Multidetector computed Tomography Angiography, Pakistan Journal of Medical Sciences. 33(4) (2017). [12] R. Lorbeer, A. Grotz, M. Dörr, H. Völzke, W. Lieb, J.-P. Kühn, B. Mensel, Reference values of vessel diameters, stenosis prevalence, and arterial variations of the lower limb arteries in a male population sample using contrast-enhanced MR angiography, PLOS One. 13(6) (2018). [[13] J.H. Joh, H.-J. Ahn, H.-C. Park, Reference Diameters of the Abdominal Aorta and Iliac Arteries in the Korean Population, Yonsei Medical Journal. 54(1) (2013) 48. [14] T. Le Houérou, D. Fabre, C.G. Alonso, P. Brenot, R. Bourkaib, C. Angel, M. Amsallem, S. Haulon, S. (2018). In Situ Antegrade Laser Fenestrations During Endovascular Aortic Repair. European Journal of Vascular and Endovascular Surgery. 56(3) (2018) 356–362.

Ingenium 2021

Fluid flow simulation of microphysiological knee joint-on-a-chip Eileen N. Lia,b, Zhong Lib, Ryan M. Ronczkab, Hang Lina,b Department of Bioengineering, bDepartment of Orthopaedic Surgery, University of Pittsburgh School of Medicine


Eileen Li is a senior bioengineering student from northern Virginia. She has a passion in tissue engineering geared towards bone regeneration. After graduation she wishes to pursue a PhD and a career in academia. Eileen N. Li

Dr. Lin is an Assistant Professor in the Department of Orthopaedic Surgery and Department of Bioengineering. His research is focused on investigating the association between aging and OA, establishing an in vitro microphysiological OA model for OA pathogenesis study and drug development, and Hang Lin, Ph.D. testing stem cell-based therapy for the repair of cartilage injury.

Significance Statement

The progression of disease-modifying drugs to combat osteoarthritis (OA) has been limited due to the absence of an appropriate OA model that can effectively simulate the pathologies in humans. Our lab has constructed a 3-dimensional, human cell-derived, multi-tissue microphysiological system (microJoint) and high-throughput microJoint (HTP-microJoint) to mimic the native joint and model OA pathogenesis. Through the utilization of finite element analysis, both microJoints were investigated for characteristics that promote joint tissue development and maintenance.

Category: Computational Research

Keywords: microphysiological system, finite element analysis, shear stress, joint disease, osteoarthritis, model Abbreviations: osteoarthritis (OA), high-throughput microJoint (HTP-microJoint), finite element analysis (FEA)


Osteoarthritis (OA) is a degenerative disorder that effects 240 million people globally, leading to inflammation, chronic pain and restricted movement of joints [1]. Current osteoarthritis (OA) drugs are merely palliative. The development of drugs for OA treatment has been limited by the unavailability of a suitable OA model. To address this issue, we have developed a 3-dimensional (3D), human cell-derived, multi-tissue microphysiological system (microJoint); and further, an additional high-throughput microJoint (HTP-microJoint) chip was designed for high-throughput drug testing and convenient real-time imaging analysis. The 3-dimensionality and multi-tissue nature of the microJoint chips establish many similarities to the native joint and can therefore enhance our investigation of OA in comparison to models that are 2-dimensional or do not have multiple joint tissues. However, the fluid flow characteristics of the chips and their effects on the tissue maturation and maintenance are not fully understood, therefore finite element analysis (FEA) was conducted for both chips to quantify the fluid flow-induced shear stress and flow trajectory parameters within the tissue chambers. Volumetric flow rates utilized in the lab for tissue culture within the microJoint chips were applied to the geometry for the analysis. Laminar flow and the continuum hypothesis assumptions were validated through calculations. Fluid velocity analysis indicated no stagnant media areas within the microJoint chips. Quantified fluid-induced shear stress on the tissues within the microJoint chips have been previously reported to enhance osteogenic and chondrogenic tissue differentiation. The analysis suggests that both microJoint chips provide environments that enhance joint tissue development. Therefore, the microJoint chips can closely mimic native joint tissues and will establish a physiologically relevant model of OA progression.

1. Introduction

Osteoarthritis (OA) is a painful and debilitating disease that affects multiple joint components, including articular cartilage, subchondral bone, synovium, and infrapatellar fat pad. The limited progress in the development of disease-modifying OA drugs is mainly due to the absence of an effective OA model that mimics complex and active tissue crosstalk within the joint.


1.1 Existing models

Animal models represented by mice have a few advantages such as the ability to simulate natural OA progression and the ability to efficiently establish chemically- or trauma-induced OA [2]. However, animal models have limitations as these models are expensive, mimic only part of the features of human OA, and do not accurately predict clinical responses to OA drugs [2]. Along with in vivo models, in vitro models have also been used to investigate OA. Monolayer culture of chondrocytes does not replicate the in vivo 3D tissue niche and often leads to dedifferentiation. Pellet culture of chondrocytes can generate 3D cartilage tissues, but does not replicate tissue crosstalk. Another in vitro model, tissue explants, allows for the investigation cells in their native environment and the study of cartilage as a whole tissue. However, this model has a strong dependence on the donor and tissue harvest site and commonly seen cell death on the tissue edges [2]. To address these shortcomings, we have developed a 3-dimensional (3D), human cell-derived, multi-tissue microphysiological system (microJoint) that can be used to mimic the in vivo joint environment, tissue-specific ECM, interactive physiological biochemical and physical signals, along with modeling OA [3]. In addition, a high-throughput microJoint chip (HTP-microJoint), which significantly reduces cell and material consumption, was also developed to facilitate potential drug testing applications. Previous studies in the lab have shown the successful maintenance of the tissue phenotypes in the microJoint tissue up to 4 weeks as well as the generation of OA models by inducing synovium inflammation [4]. As we move forward, it is essential to analyze and evaluate the effects of flow trajectory and shear stress induced by medium perfusion on tissue health.

1.2 Finite Element Analysis

To analyze and evaluate flow trajectory and shear stress parameters, finite element analysis (FEA) was used for modeling and simulation. FEA has the benefit of applying and analyzing fluid dynamic parameters on a defined geometry. Through FEA, a fluid can be applied within the microJoint geometry and the software will report computationally simulated data regarding shear stress gradient and fluid flow velocity. An analysis of this data can assist us to understand how these parameters have affected the generation and maintenance of the microtissues. In this research, we hypothesize that the medium perfusion parameters used for microJoint and HTP-microJoint culture lead to suitable flow rate and shear stress for the maturation and phenotype maintenance of the tissues.

50 Undergraduate Research at the Swanson School of Engineering

2. Methods 2.1 MicroJoint and HTP-microJoint chip geometry

The microJoint model was developed using Solidworks. The microJoint model is a rectangular prism with the dimensions of 56 mm by 36 mm by 14.75 mm. It consists of 12 cylindrical chambers, which each are 14.75 mm in height and have a diameter of 8 mm, to hold microJoint tissue scaffolds that are 4.75 mm in height and 3.00 mm in diameter, and allow for media flow on the top and bottom of the scaffold (Figure 1A). The HTP-microJoint model was developed utilizing Solidworks and consists of a rectangular prism geometry that has dimensions of 27 mm by 5.5 mm by 1.5 mm, containing 3 combined rectangular prism chambers, with a tissue scaffold placed in the middle of each of these chambers with dimensions of 5 mm by 2 mm by 1.5 mm, allowing for media flow on the top and bottom of the scaffolds (Figure 1B).

Figure 1. Geometry of MicroJoint Chips. Diagram identifying the size of the microJoint chips and microtissue. (A) MicroJoint chip geometry and (B) HTP-microJoint chip geometry and section view to show inside chambers with tissue scaffold area labeled in blue and measurements in mm. Inlets identified with arrows pointing in and outlets indicated with arrows pointing out.

2.2 FEA parameters

FEA was conducted to simulate fluid flow in the microJoint model using Solidworks. The two volumetric flow rates utilized for microJoint culture, 12 mL/s and 0.0167 mL/s, were used in the simulations as these were medium perfusion parameters shown to have successful tissue phenotype maintenance in our previous lab studies [4]. DMEM with 10% fetal bovine serum (FBS) was simulated to flow in the top inlets, while DMEM without FBS was simulated to flow in the bottom inlets. DMEM with and without 10% FBS dynamic viscosity and density values were obtained from previous literature and used to develop user-defined fluids for the simulation [5]. Outlet conditions were set to be environmental pressure. The temperature was set at 37 °C, the incubator temperature, and force of gravity was applied. Global points for the simulation were set as average shear stress and average velocity. An FEA was also conducted on the HTP-microJoint model to simulate fluid flow. Fast and slow flow simulations had inlet volumetric flows of 3.16 mL/s and 0.004 mL/s, respectively. The user-defined fluids for the microJoint simulation were

Ingenium 2021

used for this simulation as well, along with identical outlet conditions and temperature to the microJoint simulation. Global points for the HTP-microJoint were also set as average shear stress and average velocity. Solidworks was used on both chip geometries to automatically develop a suitable mesh for the geometry.

2.3 Fluid mechanics assumptions

Laminar flow was selected to represent the fluid flow within both microJoints as the flow channel and fluid velocity for both chips in all simulations were relatively small. Thus, giving Reynolds numbers less than 2300 (Table 1), allowing for the assumption of laminar flow. The average fluid velocity v was calculated with the following equation 1:

v = Q/A

(Equation 1)

With Q representing volumetric flow rate, A representing the area the fluid is passing through. And the Reynolds number, Re, was calculated with the following equation 2:

Re = ρuD/μ

(Equation 2)

With ρ representing the density of the fluid, u representing the flow velocity, D representing the diameter of the inlet, and m representing dynamic viscosity of the fluid. MicroJoint Chip

Volumetric Flow Rate (μL/s)

Fluid Velocity (mm/s)


Reynolds Number








10.9 0.01



Top Bottom














boundary conditions. Furthermore, although the HTP-microJoint had smaller inlets in comparison to the microJoint, the kn value for the HTP-microJoint is still less than 0.1.

3. Results

The fast media flow for the microJoint led to an average shear stress of 190 mPa for the top scaffold surface and an average shear stress of 115 mPa for the bottom surface (Figure 2A), with a maximum shear stress of 446 mPa and 275 mPa for the top and bottom surfaces, respectively (Figure 3A). The slow media flow for the microJoint led to an average shear stress of 0.1264 mPa, a maximum of 0.619 mPa for the top scaffold, and an average of 0.161 mPa and a maximum of 0.389 mPa for the bottom of the scaffold (Figure 2A, 3B). The HTP-microJoint had a maximum shear stress of 0.05 Pa and an average shear stress of 0.02 Pa during fast flow (Figure 4A, Figure 2B) and a maximum of 55.1 and an average of 25.2 mPa during slow flow (Figure 4B, Figure 2B). Fluid velocity in both microJoint and HTP-microJoint had a minimum fluid velocity of larger than 0 under all flow conditions, indicating no stagnant areas in the chips (Figure 2C).

Table 1. Calculated Fluid Velocity and Reynolds Number for MicroJoint Chips

The continuum hypothesis is also assumed true for this simulation as the calculated Knudsen number is less than 0.1 [6]. The Knudsen number, kn, was calculated with equation 3: kn = λ/L (Equation 3) Where kn represents the Knudsen number, l represents the mean free path of the molecule and L is the characteristic length, which was determined to be the diameter of the inlet. The mean free path of water was used (l = 0.31 nm) [6] as cell culture medium is primarily made of water. Thus, a kn value of 0.0000016 was obtained for the microJoint, allowing for the assumption of the continuum hypothesis along with no-slip boundary conditions [6]. Although DMEM has differing properties from water, with higher dynamic viscosity and density, it is assumed that due to the higher dynamic viscosity, DMEM would have a smaller mean free path than water. Therefore, the real kn value would be less than 0.0000016, which still allows the assumption of the continuum hypothesis and no-slip

Figure 2. MicroJoint chips characteristics. Values of average shear stresses and flow rates in both microJoint chips have been reported to enhance tissue development (A) Average shear stress in the microJoint chip at 12 μL/s and 0.0167 μL/s. (B) Average shear stress in the HTP-microJoint chip at 3.16 μL/s and 0.004 μL/s. (C) Average fluid velocity in both chips in fast flow mode (12 and 3.16 μL/s) and slow flow mode (0.0167 and 0.004 μL/s


Figure 3. Heat map of microJoint. Fluid induced shear stress undergone by the microJoint microtissue are not large enough to induce cell death and instead assist in tissue maturation. Heat map summarizing shear stress in the microJoint chip at 12 μL/s (A) and 0.0167 μL/s (B) flow.

52 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Figure 4. Heat map of HTP-microJoint. Fluid induced shear stress undergone by the HTP-microJoint microtissue are not large enough to induce cell death and instead assist in tissue maturation. Heat map summarizing shear stress in the HTP-microJoint chip at 3.16 μL/s (A) and 0.004 μL/s (B) flow.

4. Discussion

It has been reported that fluid induced shear stresses of over 1 Pa could induce cell apoptosis [6]. Both microJoint chips applied a fluid induced shear stress of less than 1 Pa onto the tissues, indicating that the chips would not cause cell death [7]. Instead, it may enhance osteogenic and chondrogenic properties. For example, the average shear stress for the top scaffold surface during 0.012 mL/s flow for the microJoint has been reported to be within the range that was shown to increase the expression of osteogenic marker genes, such as alkaline phosphatase (ALP), osteocalcin (OCN), collagen I (COL1), and runt-related transcription factor 2 (RUNX2) [7]. Osteogenic proteins ALP, fibronectin (FN), fibronectin receptor (FNR) and transforming growth factor beta-1 (TGF-β 1) are also seen to be upregulated for human osteoblastic cells that undergo fluid induced shear stresses of 1 to 63 mPa [8], a range that includes the average shear stress in the HTP-microJoint during 0.004 mL/s flow. Furthermore, the maximum shear stress of 0.05 Pa in the HTP-microJoint during 3.16 mL/s flow has been shown to increase cytosolic calcium ion concentration in osteoblasts [8]. Finally, a fluid induced shear stress of 0.02 Pa, which is the average shear stress of the HTP-microJoint during 3.16 mL/s flow, is within the range of fluid induced shear stress that has been reported to increase the expression of collagen, glycosaminoglycan (GAG), and proteoglycan, all of which are essential components in the extracellular matrix (ECM) of cartilage [9].

5. Conclusions

Both microJoints were shown to have shear stresses on the scaffold surface that would increase or assist the maturation of tissues in microJoint. The average fluid velocities in both microJoints show no areas of stagnant media in the microJoint chamber, allowing for sufficient growth factor and nutrient diffusion while enabling tissue crosstalk via the shared bottom flow. Lastly, the flow trajectory within the microJoint enables tissue crosstalk, mimicking tissue interactions seen in the native joint. There are some limitations of this study. First, this study was based on the simulation and results will have to be verified in the actual microJoint in future research. Secondly, we have not considered how the loading that joint must bear will affect the results. In the future, we will continue to develop this platform to enhance its clinical relevance. First, we will introduce a new component into the current model to study the influence of mechanical loading on tissue health and will simulate the flow with the presence of mechanics. Second, we will test some known drugs in microJoint to investigate if this novel system is able to precisely predict toxicity and efficacy in humans. Finally, we will use patients-derived cells to fabricate to microJoint, addressing the need of precision medicine in treating OA.


6. Acknowledgements

Summer Research Internship (E.L.) funded by the Swanson School of Engineering and the Office of the Provost. This research is in part supported by the National Institutes of Health (UH3TR002136)

7. References

[1] Nelson A. E. (2018). Osteoarthritis year in review 2017: clinical. Osteoarthritis and cartilage, 26(3), 319–325. [2] He, Y., Li, Z., Alexander, P. G., Ocasio-Nieves, B. D., Yocum, L., Lin, H., & Tuan, R. S. (2020). Pathogenesis of Osteoarthritis: Risk Factors, Regulatory Pathways in Chondrocytes, and Experimental Models. Biology, 9(8), 194. [3] Makarczyk M, Gao Q, He Y, Li Z, Gold M, Hochberg M, Bunnell B, Tuan R, Goodman S, Lin H. Current Models for the Development of Disease-Modifying Osteoarthritis Drugs. Tissue Engineering. [4] Li Z, Lin Z, Lopez MR, O’Donnell B, Li X, Ian J. Moran IJ, Alexander PG, Goodman SB, Bunnell BA, Lin H, Tuan RS. Organ-on-a-chip System for the Modeling of Synovial Joint Pathologies. ORS 2019 Annual Meeting Paper No. 0148 [5] Poon C. (2005) Measuring the density and viscosity of culture media for optimized computational fluid dynamic analysis of in vitro devices. [6] Rapp B, (2017) Chapter 9 – Fluids. In Micro and Nano Technologies, Microfluidics: Modeling, Mechanics and Mathematics. (pp 243-263). Elsevier. [7] P. Wang, P. P. Guan, C. Guo, F. Zhu, K. Konstantopoulos, Z. Y. Wang, Fluid shear stress-induced osteoarthritis: roles of cyclooxygenase-2 and its metabolic products in inducing the expression of proinflammatory cytokines and matrix metalloproteinases, FASEB J. 27(2013) 4664-4677. [8] C. Wittkowske, G.C. Reilly, D. Lacroix, C.M. Perrault, In Vitro Bone Cell Models: Impact of Fluid Shear Stress on Bone Formation, Front. Bioeng. Biotechnol. 4(2016) 87. [9] N. Sharifi, A.M. Gharravi, Shear bioreactors stimulating chondrocyte regeneration, a systematic review. Inflamm. Regener. 16 (2019) 39.

54 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Association between functional dedifferentiation and amyloid in preclinical Alzheimer’s Disease *Elizabeth J. Mountza, *Jinghang Lia, Akiko Mizuno, PhDb, Ashti M Shah, BSc, Andrea Weinstein, PhDb, Ann D. Cohen, PhDb, William E. Klunk, PhDb, d, Beth E. Snitz, PhDb, d, Howard J. Aizenstein, MD/PhDa, b, Helmet T. Karim, PhDb Department of Bioengineering, bDepartment of Psychiatry, Physician Scientist Training Program, University of Pittsburgh School of Medicine, dDepartment of Neurology * These authors contributed equally to this work. a c

Jinghang Li

Jinghang is a senior Biomedical Engineering student from Wenzhou, China. He has been working in the Geriatric psychiatry neuroimaging lab since May 2020. After graduation he plans on pursuing his PhD in Biomedical Engineering applying deep learning on imaging diagnostics. Elizabeth Mountz is a Junior studying Bioengineering at the University of Pittsburgh Swanson School of Engineering. She hopes to continue her research efforts in Graduate School.

Elizabeth J. Mountz

Howard J. Aizenstein, M.D., Ph.D.

Dr. Aizenstein is an expert on the cognitive and affective neuroscience of aging and geriatric brain disorders. He is trained as a geriatric psychiatrist and also a computer scientist. His research program is recognized for expertise in MRI analyses methods, as well as their use for clinical research in aging.

Significance Statement

Preclinical Alzheimer’s disease (AD) is a stage of AD defined by the presence of brain amyloid without overt cognitive impairment, occurring up to two decades prior to diagnosis. Amyloid may deposit asymmetrically, which has been shown to affect neural functional asymmetry in AD but not during the preclinical period. We show that altered functional asymmetry in the hippocampus may appear as early as the preclinical stages of AD and is associated with amyloid deposition.

CATEGORY: Experimental Research

Keywords: preclinical Alzheimer’s Disease, dedifferentiation, amyloid, memory encoding


Preclinical Alzheimer’s disease (AD) is characterized by significant brain amyloid-β (Aβ) pathology without overt cognitive impairment. Aβ has been shown to deposit asymmetrically and has been shown to be associated with asymmetric brain glucose metabolism. In clinical AD, Aβ burden may exceed the compensatory reserve threshold, leading to greater neural functional asymmetry in AD individuals compared to elderly controls, where asymmetry is associated with cognitive function. To better understand AD progression, we investigated the association between markers of asymmetry, AD pathology, and cognitive function in cognitively normal older adults. Using fMRI during a memory encoding task, we calculated functional asymmetry and spread of activation in the hippocampus and dorsolateral prefrontal cortex, which are part of the core and extended memory network. Using positron emission tomography (PET), we measured brain Aβ and global glucose metabolism. We also collected data on APOE allele status and cognitive function. We conducted multivariate linear regression with measures of dedifferentiation (i.e., asymmetry index and spread of activation) as the outcome and markers of AD as independent variables. We additionally investigated their associations with domains of cognitive function. We found that greater global Aβ deposition was associated with greater hippocampal functional asymmetry and lower left hippocampal activation spread during a memory encoding task. Functional asymmetry and spread were not associated with cognitive function. Similar to studies in AD, we found that functional asymmetry was associated with greater Aβ in cognitively normal older individuals. This may hint at the early neurodegeneration in preclinical AD.

1. Introduction

Alzheimer’s disease (AD) is a process of progressive neurodegeneration which leads to severe cognitive dysfunction, including impairments in memory loss and executive control. AD is associated with a buildup of Amyloid-β (Aß) and neurofibrillary tangles in the brain. These cytotoxic proteins especially affect areas associated with memory, such as the hippocampus and dorsolateral prefrontal cortex (DLPFC) [1]. Greater Aß deposition, as measured by positron emission tomography (PET), is associated with functional changes in the brain including hippocampal atrophy, low cerebral glucose metabolism [2] and functional changes in neural activation as measured by functional magnetic resonance imaging (fMRI) [3]. Aß deposition is gradual with a preclinical stage where significant Aß deposition can be detected but no cognitive dysfunction is observed [4]. Aß deposition is often asymmetric, burdening one hemisphere more than the other [5]. One hypothesis for this lack of cognitive dysfunction with significant Aß pathology during the preclinical stage, is that compensatory neural recruitment can help delay cognitive dysfunction. In healthy individuals, neural activation is often lateralized and localized during tasks such as memory encoding due to the highly specialized roles of 55

neural networks within the hemispheres [6]. Compared to healthy individuals, greater bilateral neural recruitment as well as greater spatial extent of activation are thought to be two measures of compensatory activation which are termed dedifferentiation, or the general loss of neural specificity (i.e., regions become more domain general in terms of function). Dedifferentiation may allow the pathology-burdened hemisphere to recruit greater neural resources from the other hemisphere in order to maintain cognitive function [7]. Dedifferentiation in cognitively normal individuals is thus measured through both greater symmetry in (bilateral) fMRI activation (i.e., left dominant DLPFC activation spilled over into right DLPFC), as well as greater spread of local activation (i.e., more widespread DLPFC activation within the hemisphere) (Figure 1) [7]. The Hemispheric Asymmetry Reduction in Old Adults (HAROLD) general aging model posits that, dedifferentiation in the frontal regions of older adults may help maintain cognitive function despite the stresses of aging by recruiting greater neural resources, both bilaterally and locally [7]. These compensatory dedifferentiation mechanisms may still be intact in preclinical AD and may help maintain cognitive function during early Aß deposition.

Figure 1. A schematic representation of two forms of dedifferentiation. Greater dedifferentiation may be indicated by greater spread (upper image) and greater symmetry (lower image) of activation due to greater recruitment of neural resource locally and laterally during a given task.

56 Undergraduate Research at the Swanson School of Engineering

In later stages of AD, increased Aß burden may exceed compensatory capacity, which may result in cognitive impairment [2, 8]. In prodromal [i.e., mild cognitive impairment (MCI)] and mild-to-moderate AD, asymmetric Aß deposition is associated with asymmetric cerebral hypometabolism [5]. Asymmetric (i.e., lateralized to left hemisphere) glucose hypometabolism is associated with Aβ positive (i.e., greater Aß burden than a cut-off) MCI patients [9]. Thus, later stages of AD are associated with functional asymmetry. In this study, we sought to characterize the relationship between functional asymmetry and Aß deposition in preclinical AD in order to better understand methods of early intervention and diagnosis, as Aß deposition may be undetectable in the earliest stages of AD trajectory. While large bodies of research exist for structural asymmetry in AD progression [10], there are little to no studies on functional asymmetry in preclinical AD. Based on the HAROLD general aging model, we hypothesized that in cognitively normal individuals, greater AD pathology (e.g., Aß deposition and cerebral hypometabolism) would be associated with greater dedifferentiation, measured with greater symmetry and spread of activation. Because our participants were cognitively normal participants, we did not expect to observe an association between cognitive function and activation symmetry or spread. We used the face-name encoding fMRI task because it reliably activates the hippocampus and DLPFC [10]. The hippocampus is critical for memory and learning and shows neuronal degeneration early in the onset of AD [1], while the DLPFC is part of an extended memory encoding network by coordinating the relational information [7].

Ingenium 2021

Table 1. Demographics and Summary of Cognitive Domain Scores Note: *p < 0.05. We did not conduct a statistical test on Global PiB (SUVR) since this is how these groups were defined.

2. Methods

Participants and Assessments: We analyzed cross-sectional data from 87 cognitively normal older adults (>65 years). All participants underwent a neuropsychological test battery used by the University of Pittsburgh Alzheimer Disease Research Center (ADRC) to assess cognitive function. Test scores were combined into domain composite z-scores including memory (Logical Memory, Modified Rey-Osterreith Figure, ADRC Word List); visuospatial abilities (Block Design, Modified Rey-Osterreith Figure Copy); language (Animal and Letter Fluency, the 60-Item Boston Naming Test); and executive attention (Trail Making Test A and B, Clock Drawing, Maximum Digit Span Forward and Backward, Stroop Interference Score, and Digit Symbol Substitution) [11] (Table 1). The memory domain was split into learning and retrieval to isolate the role of the DLPFC in delayed memory retrieval. We also collected demographic data and genetic AOPE status. Neuroimaging Data Acquisition and Preprocessing: PET scans were used to measure participant cerebral Aβ deposition with Pittsburgh Compound B (PiB) and glucose metabolism with fluorodeoxyglucose (FDG) tracers. Participants were classified as PiB positive or negative based on our previous study [12]. We collected whole-brain (excluding cerebellum) fMRI (a 3T Siemens Trio scanner) during a face-name memory encoding task. During the task, participants were shown a face-name pair and were asked to subjectively determine whether each face fit the name. This task consisted of two blocks: unfamiliar facename pairs (novel condition) and familiar (i.e., previously seen) face-name pairs (control condition). This task was administrated over three runs and contained 50 facename pairs. We computed the contrast between novel and control. For our analysis, we extracted functional activation in two regions of interest (ROIs): hippocampus and DLPFC of each hemisphere based on the automated anatomical

labeling atlas. After performing the task in the scanner, participants took a post-scan test in which they saw the same faces with two name options. They were asked to select the name they saw in the scanner as an estimate of recognition accuracy. Computation of Dedifferentiation Measures: We computed the asymmetry index (AI) as the hemispheric laterality of the mean activation in each region. We calculated the asymmetry index using the equation:

where L (left) and R (right) are the mean activation values of a given ROI in the corresponding hemisphere [13]. The equation yields a value between -1 and 1, where 0 represents symmetric activation between the hemispheres, and a negative or positive AI indicates right or left hemispheric dominant activation, respectively. We also computed absolute AI (abs_AI) as a measure of non-directional laterality. Smaller abs_AI indicates greater dedifferentiation (i.e., greater symmetric activation). We also measured dedifferentiation with the spread of activation within each ROI [14]. For each participant, we identified the peak activation voxels, then generated a series of spheres with radii ranging from 1 voxel to 25 voxels centered around peak activation, where we calculated the average activation within each sphere. We then plotted the mean activation with respect to different radii. We used a linear interpolation method to estimate the neural activity decline and took the radius where the activation value was half of the observed peak activation as the spread of the activation [e.g., full width half maximum (FWHM)]. Greater FWHM indicates greater spread (or dedifferentiation). Statistical Analyses: We conducted eight multivariate linear regressions in R to investigate the associations between measures of dedifferentiation (AI, abs_AI, left 57

hemispheric FWHM, and right hemispheric FWHM, in both ROIs) and AD related factors (PiB status, global cerebral glucose metabolism, and APOE E4 status). We conducted four linear regressions to investigate the association between left/right hemispheric spread of activation and abs_AI for both ROIs. We also conducted ten multivariate linear regressions to investigate the association between five cognitive domains and both measures of dedifferentiation (abs_AI and FWHM). Other AD related factors were included as predictor variables. All models controlled for demographic factors (age, education, sex, and race), and post-scan recognition accuracy was included as a covariate to predict cognitive function.

3. Results

Among our multivariate linear regressions, two models showed marginal significance: absolute hippocampus AI (F(7,53) = 1.95, R 2 = 0.205, p = 0.08) and spread of activation in the left hippocampus (F(7,53) = 2.07, R 2 = 0.215, p = 0.063). PiB positive individuals (11/66) showed greater asymmetric activation in the hippocampus (ß = 0.33, p = 0.024, Figure 2a). PiB positive individuals showed lower spread of activation in the left hippocampus (ß = -4.18, p = 0.019, Figure 2b). DLPFC dedifferentiation measures were not associated with any AD related factors.

Figure 2. PiB positive group showed significantly greater asymmetric hippocampal activation (a) and less spread of activation in the left hippocampus (indexed by FWHM) (b) during a memory encoding task compared to the PiB negative group.

58 Undergraduate Research at the Swanson School of Engineering

Individuals with lower activation spread in the left hippocampus had greater absolute asymmetry of hippocampal activation (F(1,82) = 12.14, R 2 = 0.129, p = 0.001; ß = -4.33, p = 0.001). Only executive attention domain score was significantly associated with PiB positive status in the two separate models with all AD factors: abs_AI (F(10, 50) = 2.68, R 2 = 0.349, p = 0.01; ß = -0.53, p = 0.016) and spread (F(12, 48) = 3.39, R 2 = 0.459, p = 0.001; ß = -0.61, p = 0.004). No cognitive domain scores were associated with markers of dedifferentiation.

4. Discussion

Contrary to our hypothesis, we found that PiB positive individuals had greater hippocampal asymmetry and lower spread of activation in the left hippocampus. Unlike the prefrontal dedifferentiation pattern posited in the HAROLD model, there were no associations with DLPFC asymmetry or spread of activation. Greater asymmetry in preclinical AD may be a mechanism of either greater recruitment of activation in the right hippocampus or reduced activation of the left hippocampus. We found that lower left hippocampal spread of activation was associated with greater hippocampal activation asymmetry. PiB positive individuals who are cognitively normal may also have greater cognitive reserve, which enables them to compensate for greater pathology to maintain cognitive function. Therefore, the association between PiB positive individuals and greater asymmetry and less spread may be indicative of the individual differences of greater cognitive reserve, which may help to maintain normal cognitive function despite pathology. In the clinical stages of AD, functional changes such as asymmetric cerebral hypometabolism are associated with asymmetric Aß deposition [5]. Further, Aβ positive status in AD is associated with leftward lateralization of glucose metabolism decline [9]. Our results indicate that early pathological Aß deposition in the brain is associated with greater asymmetry in preclinical stages of AD, which is in agreement with previous studies that have associated asymmetry with AD, but hints toward an earlier neurodegenerative effect of Aß deposition. We found no associations between dedifferentiation measurements and cognitive function, as these compensatory mechanisms may help to maintain cognitive function. On the other hand, greater Aβ deposition was associated with lower executive attention. These may suggest that Aß deposition is a predictor of the higher order cognitive function regardless of dedifferentiation patterns among cognitively normal individuals or that a functional MRI task that engages executive function may be a better potential functional marker (i.e., face-name task is primarily memory encoding).

Ingenium 2021

There are several limitations to this study – primarily, the associations between dedifferentiation measures and Aß deposition should be validated in other samples and in studies that utilize longitudinal designs to observe changes in activation spread and symmetry in relation to changes in Aß deposition. These results show correlations and do not imply causation of Aß deposition on functional brain activity. This study did utilize a fairly large cross-sectional sample with both functional imaging and intricate PET imaging, which is a strength.

5. Conclusion

We used two measurements (e.g., functional asymmetry and spread of activation) to quantify the extent of neuronal dedifferentiation as a tool to assess the association with Alzheimer’s disease pathology in preclinical AD. We identified significant differences in markers of neural activity associated with the presence of AD risk factors in the hippocampus. We did not find associations between DLPFC and any AD risk factors. We speculate that AD risk factors such as significant Aß deposition may limit the spread of activation in the left hippocampus which affects functional asymmetry. This observation in preclinical AD suggests an early neurodegenerative effect of Aß, before any onset of cognitive symptoms. Future longitudinal studies should identify whether these changes track the progression of Aß or are an indirect consequence of Aß accumulation.

6. Acknowledgements

This study was supported by the SSOE Summer Research Internship and P50 AG005133, P01 AG025204, R37 AG025516 from the National Institute of Health. We would like to acknowledge the GPN Lab for providing the research infrastructure and continuous mentorship.

7. References

[1] Svenningsson, A.L., et al., beta-amyloid pathology and hippocampal atrophy are independently associated with memory function in cognitively healthy elderly. Sci Rep, 2019. 9(1): p. 11180. [2] Cohen, A.D., et al., Basal cerebral metabolism may modulate the cognitive effects of Abeta in mild cognitive impairment: an example of brain reserve. J Neurosci, 2009. 29(47): p. 14770-8. [3] Edelman, K., et al., Amyloid-Beta Deposition is Associated with Increased Medial Temporal Lobe Activation during Memory Encoding in the Cognitively Normal Elderly. Am J Geriatr Psychiatry, 2017. 25(5): p. 551-560.

[4] Dubois, B., et al., Preclinical Alzheimer’s disease: Definition, natural history, and diagnostic criteria. Alzheimers Dement, 2016. 12(3): p. 292-323. [5] Frings, L., et al., Asymmetries of amyloid-beta burden and neuronal dysfunction are positively correlated in Alzheimer’s disease. Brain, 2015. 138(Pt 10): p. 3089-99. [6] Dolcos, F., H.J. Rice, and R. Cabeza, Hemispheric asymmetry and aging: right hemisphere decline or asymmetry reduction. Neuroscience & Biobehavioral Reviews, 2002. 26(7): p. 819-825. [7] Cabeza, R., Hemispheric asymmetry reduction in older adults: the HAROLD model. Psychol Aging, 2002. 17(1): p. 85-100. [8] Jack, C.R., Jr., et al., Tracking pathophysiological processes in Alzheimer’s disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol, 2013. 12(2): p. 207-16. [9] Weise, C.M., et al., Left lateralized cerebral glucose metabolism declines in amyloid-beta positive persons with mild cognitive impairment. Neuroimage Clin, 2018. 20: p. 286-296. [10] Sperling, R.A., et al., Encoding novel face-name associations: a functional MRI study. Hum Brain Mapp, 2001. 14(3): p. 129-39. [11] Snitz, B.E., et al., Risk of progression from subjective cognitive decline to mild cognitive impairment: The role of study setting. Alzheimer’s & Dementia, 2018. 14(6): p. 734-742. [12] Price, J.C., et al., Kinetic modeling of amyloid binding in humans using PET imaging and Pittsburgh Compound-B. Journal of Cerebral Blood Flow & Metabolism, 2005. 25(11): p. 1528-1547. [13] Seghier, M.L., Laterality index in functional MRI: methodological issues. Magn Reson Imaging, 2008. 26(5): p. 594-601. [14] Gordon, B.A., et al., Spread of activation and deactivation in the brain: does age matter? Front Aging Neurosci, 2014. 6: p. 288.


Bioinformatic analysis of fibroblast-mediated therapy resistance in HER2+ breast cancer Jacob C. McDonalda and Ioannis K. Zervantonakisa, b Tumor Microenvironment Engineering Laboratory, Department of Bioengineering, bUPMC Hillman Cancer Center


Jacob McDonald is a junior bioengineering student at the University of Pittsburgh with a focus in cellular engineering. After graduating, he plans to further his education and bioengineering research by attending graduate school. Jacob C. McDonald

Ioannis Zervantonakis is an Assistant Professor at the Department of Bioengineering, University of Pittsburgh and at the Hillman Cancer Center UPMC. He serves as an Early-Career Associate Scientific Advisor to Science Translational Medicine, has received the 2020 Hillman Early-Career Fellow Ioannis K. for Innovative Cancer Research Award Zervantonakis and his laboratory is funded by an Elsa Pardee Foundation Award.

Significance Statement

Interactions between tumor cells and stromal fibroblasts in the tumor microenvironment have been shown to mediate drug resistance in some HER2 overexpressing breast cancers. This study identifies signaling pathways within tumor cells that are critical to fibroblast-mediated drug resistance and proposes targeting specific proteins to restore treatment sensitivity.

Category: Computational Research

Keywords: breast cancer, tumor microenvironment, fibroblasts, drug resistance

60 Undergraduate Research at the Swanson School of Engineering


Multiple targeted therapies have been developed for the treatment of HER2 overexpressing (HER2+) breast cancers, such as the dual HER2/EGFR kinase inhibitor lapatinib, but many patients eventually develop resistance to these treatments. One proposed cause of HER2 therapy resistance is the interaction between tumor cells and stromal fibroblasts in the tumor microenvironment, which has been reported to activate signaling pathways in HER2+ tumor cells that lead to a decrease in drug sensitivity. Here, we identify and compare both protein-expression and gene-expression signatures of breast cancer cells that exhibit fibroblast-mediated lapatinib resistance. We then integrate proteomic data with cell treatment response data to identify optimal protein targets for combination therapy that may restore lapatinib sensitivity in fibroblast-protected cancer cells. Our signatures for fibroblast protection suggest that fibroblasts reduce lapatinib sensitivity in HER2+ breast cancer cells through reactivation of the PI3K/Akt and mTOR signaling pathways, which results in increased cell cycle signaling and changes in lipid metabolism. Specifically, we found that lapatinib resulted in greater inhibition of proteins in the Akt/mTOR, cell cycle, and lipid synthesis pathways in cell lines that exhibited high sensitivity to lapatinib (i.e. low cell viability). Together, our findings suggest that inhibition of these pathways may be sufficient to restore lapatinib sensitivity in fibroblast-protected breast cancer cells.

1. Introduction

HER2 overexpressing (HER2+) breast cancer accounts for ~20% of all breast cancer cases [1]. The HER2 receptor is a receptor tyrosine kinase in the epidermal growth factor receptor family (ErbB), and heterodimerization of HER2 with other members of the ErbB family leads to signal transduction through the PI3K/Akt and MAPK survival pathways [2]. As a result, overexpression of HER2 can lead to increased cancer cell survival, proliferation, and invasion. Multiple targeted therapies have been developed to treat HER2+ breast cancer, such as the dual HER2/ EGFR kinase inhibitor lapatinib. Lapatinib acts by binding to the kinase domain of the HER2 receptor, preventing its autophosphorylation and therefore its ability to transduce pro-survival signaling [3]. However, even though lapatinib treatment may initially be successful in treating HER2+ breast cancers, many patients will eventually develop resistance to these treatments. Several mechanisms of resistance to lapatinib treatment have been proposed, such as the activation of compensatory signaling pathways through other kinases, mutation of the HER2 kinase domain, or gene amplification [3]. Another proposed cause of HER2 therapy resistance is the interaction between tumor cells and stromal fibroblasts in the tumor microenvironment, which can occur through direct cell-cell contact or through the secretion of soluble factors. These tumor-fibroblast interactions have been reported to activate signaling pathways in some HER2+ tumor cell lines, such as the mTOR

Ingenium 2021

and JAK2/STAT3 pathways, which leads to a decrease in lapatinib sensitivity [4, 5]. By performing an unbiased bioinformatic analysis of the tumor cell proteome and transcriptome, this study aimed to identify the signaling pathways within tumor cells that are critical to fibroblast-mediated lapatinib resistance. After identifying the signaling pathways most affected by fibroblast coculture, we then assessed the relationship between tumor cell viability and the level of protein expression in these pathways. We hypothesized that protein expression in the signaling pathways upregulated by fibroblast coculture would directly correlate with tumor cell viability. Specifically, we proposed to identify proteomic predictors of drug sensitivity (proteins whose lapatinib-induced decrease in protein expression correlates with a reduction in cell growth) and drug resistance (proteins whose lapatinib-induced decrease in protein expression correlates with an increase in cell growth).

2. Methods 2.1 Cell Lines and Culture

A panel of six HER2+ breast cancer cell lines was used in this study. The cell lines EFM192, BT474, and HCC202 have been shown to have decreased sensitivity to lapatinib treatment when cocultured with AR22 fibroblasts, and have therefore been classified as fibroblast-protected. The cell lines HCC1954, HCC1419, and SUM225 do not show an increase in viability in fibroblast coculture and are classified as fibroblast-insensitive [4]. Cells were grown in 96 well plates and treated with either 0.1µM lapatinib or with a dimethyl sulfoxide (DMSO) control for 96 hours. For the tumor-fibroblast coculture, fibroblasts were physically separated from the tumor cells using transwell filters, allowing for the exchange of secreted factors between the fibroblasts and tumor cells without cell-cell interactions due to physical contact. Data for these experiments was previously collected and further details of the experiments performed are provided in the supplementary methods in [4]. The focus of this project was a bioinformatic analysis of this data.

2.2 Protein and Gene Signatures

After 48 hours of treatment, protein expression in the tumor cells was quantified using reverse phase protein arrays (RPPA) and gene expression was quantified using RNA sequencing. RPPA data for 409 proteins was obtained as previously described [6], and RNA was extracted using the Qiagen RNA extraction kit and submitted for sequencing to the Harvard Bauer Core. The fibroblast-protected cell line EFM192 was chosen to derive the fibroblast-protection signatures due to its large increase in viability in AR22 fibroblast coculture [4]. Proteins in lapatinib-treated EFM192 that had a greater than 1.4-fold change in expression when cocultured with AR22 fibroblasts compared to monoculture were selected for the protein expression signature using R. This threshold was chosen to be higher than variation in protein expression between biological replicates, yet low enough to identify multiple proteins for follow-up functional analysis. A total of 52/409 proteins were found to be differentially expressed using these criteria. The gene expression signature was identified using the R package ballgown and consisted of genes with a greater than 2-fold change in expression when cocultured with fibroblasts (q < .05). To identify the cellular functions most significantly affected by fibroblast coculture, Gene Ontology enrichment analysis was performed using R packages clusterProfiler and enrichplot.

2.3 Identification of Protein Targets

Tumor cell response to lapatinib treatment was tracked every 4 hours for 96 hours using IncuCyte live cell analysis. The relationship between lapatinib-induced protein inhibition and tumor cell viability was then assessed by performing a linear regression of the log2 fold change in cell number vs the log2 fold change in protein expression after 48 hours of lapatinib treatment. Proteins from the fibroblast-protection signature whose change in expression had a non-zero correlation with tumor cell viability (p < .05) were identified as potential targets to restore lapatinib sensitivity in fibroblast-protected tumor cells.


Figure 1. Protein Signature - Lapatinib induced protein expression changes (log2 transformed ratios of lapatinib-treated samples normalized by untreated samples) show restoration of pro-survival signaling only in EFM192 and HCC202 fibroblast-protected cell lines that are cocultured with fibroblasts. Blue corresponds to negative log2 ratios indicating protein downregulation while red corresponds to positive log2 ratios indicating protein upregulation. White values indicate no change in expression due to lapatinib treatment.

3. Results

The fibroblast-protection signature consisted of 52 proteins that were identified as being differentially expressed between lapatinib-treated EFM192 monoculture and fibroblast coculture. Compared to monoculture, coculture of lapatinib-treated EFM192 with fibroblasts resulted in ineffective inhibition of proteins in the Akt, mTOR, cell cycle, and lipid metabolism pathways, while the expression of pro-apoptotic proteins was reduced (Fig. 1). This fibroblast-induced effect on protein expression was also observed in the fibroblast-protected cell line HCC202, but it was not as evident in fibroblast-protected BT474 or the fibroblast-insensitive cell line HCC1954. Like the protein signature, Gene Ontology enrichment analysis also revealed the pro-survival effects of fibroblast coculture, with cell cycle regulation (specifically mitosis) and lipid metabolism regulation being among the most significantly enriched groups of genes (Fig. 2).

The extent of protein inhibition due to lapatinib treatment correlated with the viability of tumor cells for 44 of 52 proteins in the fibroblast-protection signature, again including many proteins involved in Akt, mTOR, cell cycle, and lipid metabolism pathways (p < .05, Fig. 3).

Figure 3: Identifying Protein Targets – Lapatinib-induced inhibition of proteins in the protein signature such as phospho-S6 (r2 = .620, p < .001) correlates with the viability of HER2+ cell lines. As the log2 fold change in phospho-S6 expression becomes more negative (greater inhibition), the post-treatment live cell number is reduced. Each dot represents a different biological replicate.

4. Discussion

Figure 2. Gene Ontology Enrichment Analysis - Top 15 most significantly enriched Gene Ontology terms based on p-value (p < 10-9). Plotted is the distribution of log2(coculture/monoculture) fold changes of the core genes in each Gene Ontology term. A distribution to the right of the vertical line indicates that the cellular function is upregulated as a result of fibroblast coculture.

62 Undergraduate Research at the Swanson School of Engineering

The protein and gene expression signatures derived for fibroblast-protected cell lines suggest that tumor-fibroblast interactions reduce lapatinib sensitivity through reactivation of the PI3K/Akt and mTOR signaling pathways, which results in increased cell cycle signaling and changes in lipid metabolism. These findings are consistent with previous studies on drug resistance in HER2+ breast cancer,

Ingenium 2021

which have shown that the PI3K/Akt and mTOR pathways can mediate resistance to HER2 targeted drugs such as lapatinib or trastuzumab [7, 8]. Other studies also reveal changes in cellular metabolism as a possible mechanism for drug resistance, such as Ruprecht et al. who found that some lapatinib resistant HER2+ cancers mediate resistance through a reprogramming of glycolytic activity [9]. There does appear to be some heterogeneity in protein response to fibroblast coculture between the fibroblast-protected cell lines. For example, while lapatinib-treated EFM192 and HCC202 show a recovery of proteins in the Akt /mTOR and cell cycle pathways when cocultured with fibroblasts, these proteins still seem to be inhibited in the fibroblast-protected cell line BT474 (Fig. 1). It is possible that fibroblast-protection in BT474 is mediated by mechanisms other than the Akt/mTOR pathways. Also, protein recovery in the lipid synthesis pathway is seen in the cell line EFM192, but much less so in HCC202 or BT474. This suggests that EFM192 may have a higher dependency on lipid synthesis for survival and proliferation than the other fibroblast-protected cell lines. The variation in proteome response to fibroblast coculture between fibroblast-protected cell lines highlights the complexity of fibroblast-mediated lapatinib resistance. The extent of lapatinib-induced inhibition for proteins in the Akt/mTOR, cell cycle, and lipid synthesis pathways correlated with tumor cell number, supporting our hypothesis that fibroblasts upregulate proteins in tumor cells that directly correlate with tumor cell viability. Our results suggest that inhibition of these pathways may reduce tumor cell viability and restore lapatinib sensitivity in fibroblast-protected cells. It has previously been shown that inhibition of mTOR in combination with lapatinib can overcome fibroblast-mediated lapatinib resistance [4], and based on the results of the linear regression, targeting proteins in the cell cycle or lipid synthesis pathways could also potentially have anti-tumor effects and be used in lapatinib combination therapy. In the future, we plan to perform mechanistic studies to evaluate the causal role of these proteins on drug sensitivity using pharmacological and genetic approaches.

5. Conclusion

6. Acknowledgments

Funding was provided by the Department of Bioengineering at the University of Pittsburgh. Fellow lab members provided helpful discussion and advice during the completion of this project. We thank Dr. Gordon Mills and Dr. Yiling Lu for the RPPA studies.

7. References

[1] M.A. Owens, B.C. Horten, M.A. Da Silva, HER2 amplification ratios by fluorescence in situ hybridization and correlation with immunohistochemistry in a cohort of 6556 breast cancer tissues, Clin Breast Cancer. 5 (2004) 63-69. [2] N. Iqbal, N. Iqbal, Human Epidermal Growth Factor Receptor 2 (HER2) in Cancers: Overexpression and Therapeutic Implications, Mol Biol Int. 2014 (2014) 852748. [3] V. D’Amato et al, Mechanisms of lapatinib resistance in HER2-driven breast cancer, Cancer Treat Rev. 41 (2015) 877-883. [4] I.K. Zervantonakis et al, Fibroblast–tumor cell signaling limits HER2 kinase therapy response via activation of MTOR and antiapoptotic pathways, Proceedings of the National Academy of Sciences. 117 (2020) 16500-16508. [5] A. Marusyk et al, Spatial proximity to fibroblasts impacts molecular features and therapeutic sensitivity of breast cancer cells influencing clinical outcomes, Cancer Res. 76 (2016) 6495–6506. [6] R. Tibes et al, Reverse phase protein array: validation of a novel proteomic technology and utility for analysis of primary leukemia specimens and hematopoietic stem cells, Mol Cancer Ther. 5 (2006) 2512-2521. [7] M.S.N. Mohd Sharial, J. Crown, B.T. Hennessy, Overcoming resistance and restoring sensitivity to HER2-targeted therapies in breast cancer, Ann Oncol. 23 (2012) 3007-3016. [8] G. Deblois et al, ERRα mediates metabolic adaptations driving lapatinib resistance in breast cancer, Nat Commun. 7 (2016) 12156. [9] B. Ruprecht et al, Lapatinib Resistance in Breast Cancer Cells Is Accompanied by Phosphorylation-Mediated Reprogramming of Glycolysis, Cancer Res. 77 (2017) 1842-1853.

It is evident that interactions between cancer cells and the tumor microenvironment play a significant role in sensitivity to drug treatments. Here, we identified several signaling pathways that are activated by fibroblast coculture and could mediate drug resistance in HER2+ breast cancers. Development of lapatinib combination therapies targeting these pathways may help restore the effects of lapatinib treatment in patients who develop resistance. We also highlighted the variation present in proteome response to the presence of fibroblasts, and how different HER2+ cell lines might utilize different mechanisms of resistance. It is important that we continue to gain a better understanding of tumor-fibroblast interactions to develop more effective, patient-specific HER2+ breast cancer treatments.


Levator Ani muscle dimension changes with gestational and maternal age Oreoluwa Odeniyi, Megan Routzong, B.S., Steven Abramowitch, Ph.D. Department of Bioengineering Oreoluwa Odeniyi was raised in Poughkeepsie, NY. Her passion for women’s health motivates her to become an OB/GYN and embark on a path of advocacy/counseling for sexual assault survivors. Oreoluwa Odeniyi

Dr. Abramowitch received his B.S. (1998) in Applied Mathematics and Ph.D. (2004) in Bioengineering from the University of Pittsburgh. Currently, he is an Associate Professor in the Department of Bioengineering and serves as the Director of the Translational Biomechanics Laboratory. Dr. Abramowitch’s research focuses on Steven Abramowitch, understanding the impact of pregnancy, Ph.D. delivery, and other life events (aging, menopause, etc.) on the structural integrity of the pelvic floor in women.

Significance Statement

The field of women’s health is underserved and research concerning pregnant women is especially important as issues with the mother may also affect the fetus. Approximately 3.7 million births occur per year in the US. Women who have had one or more vaginal births are at an increased risk of developing pelvic floor disorders [1]. The increase in M line, which represents the descent of the levator hiatus, is used to signify pelvic floor laxity and could be indicative of injury or some type of mechanical deficiency that would increase the risk of those women developing a pelvic floor disorder.


The pelvic floor is a complex system of interconnected muscles that support the pelvic organs, such as the vagina, urethra, and rectum. The physical stress of pregnancy and childbirth can result in the weakening or permanent damage of the muscles of the pelvic floor, known as the levator ani muscles [2]. Large changes in the dimensions of the levator ani could be indicative of injury or remodeling and understanding these mechanisms could aid in the prevention or treatment of levator ani injury. Our goal was to quantify changes in the levator ani during pregnancy and with age by comparing MRI scans of women (age 20-49) at various timepoints during pregnancy. The use of four reference lines, M line, H line, Pubococcygeal line and Levator plate angle were measured from the midsagittal slice using Slicer (v. 4.11, slicer.org). These measurements were correlated with maternal and gestational age in order to define any significant relationships. There was a general increase in the M line, one of the levator ani muscle parameters, with respect to gestational age, leading to the conclusion of greater muscle laxity throughout gestation.

1. Introduction 1.1. Anatomical Background

Disruption of pelvic floor function serves as a factor that can affect a woman’s quality of life. The pelvic floor is a complex system of interconnected muscles and connective tissues that support the pelvic organs, such as the vagina, urethra, and rectum. The pelvis is divided into anterior, middle, and posterior compartments. The bladder and urethra compose the anterior compartment; the cervix and vagina compose the middle compartment; and the rectum, anus, and anal sphincters compose the posterior compartment. The levator ani muscles, the primary muscular support for the pelvic organs, attach to the sides of the lesser pelvis and unite posteriorly behind the rectum at the levator plate [2]. Three muscles are responsible for the composition of the levator ani; the iliococcygeus, pubococcygeus, and puborectalis, as shown in Figure 1.

Category: Computational Research

Keywords: Levator Ani, M line, pelvic floor, gestational age Abbreviations: Gestational age – GA, Maternal Age – MA, Pubococcygeal line – PCL, Levator Plate Angle – LPA, IRB – institutional review board

Figure 1. Pelvic Floor muscles and other pelvic anatomy from an inferior view

64 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

The U-shaped puborectalis muscle wraps around the posterior wall of the rectum, defining the anorectal junction [3]. The anorectal junction and levator plate can be quantified by the H line, M line, and levator plate angle (Figure 2a and 2b).

with increased maternal age (MA) the muscles likely begin to weaken. Therefore, we hypothesize that there will be an increase in the M line and levator plate angle with increases in both GA and MA.

2. Methods

This IRB approved, retrospective study consisted of 31 pregnant patients between the ages of 21 and 45 who underwent pelvic MRI at their physician’s request—most frequently due to right lower quadrant pain or suspected appendicitis. Using an open source software called 3D Slicer v4.11, displayed in Figure 3, the midsagittal slice in the sagittal view was identified for each patient and four levator ani measurements were taken using definitions from previous literature, shown in Figure 2a and 2b [3].

Figure 2. a) An example of the PCL(green), H line(blue), M line(red), and b) Levator Plate Angle (LPA) measurements

Changes to the H line and M line are indicative of the degree of laxity or changes in material properties of the levator ani muscles [4]. With an increase in the degree of laxity, the levator ani muscles become elongated and when stretched may not provide resistance. This makes it more difficult for these muscles to perform their functions in holding up and supporting the pelvic organs.

1.2. Hypothesis and Reasoning for Experiment

The physical stress of pregnancy and childbirth can result in the weakening or permanent damage of these pelvic floor tissues, changing their mechanical properties and shape [2]. For example, during childbirth, the puborectalis muscle can be stretched to more than three times its original length, which is twice as much as a normal skeletal muscle can endure without injury [4]. Studies have confirmed that defects in levator ani muscles due to age, menopause, or trauma (i.e. childbirth) are risk factors for developing pelvic organ prolapse—a specific pelvic floor disorder in which compromised mechanical support results in the descent of pelvic organs into the vagina [2]. Large changes in levator ani dimensions could be indicative of injury or remodeling and understanding these mechanisms could aid in the prevention or treatment of levator ani injury. Our goal was to quantify changes in levator ani dimensions during pregnancy and with age by comparing MRI scans of women age 20-49 at different gestational ages. With an increase in gestational age (GA), where the fetal weight and size are increasing, there is likely an increase in tissue remodeling and/or stress being put on the levator ani, and

Figure 3. A screenshot of the segmentation software with representative coronal (bottom left), sagittal (bottom right), and axial (top left) slices shown with the 3D segmentation (top right)

The pubococcygeal line (PCL) runs from the inferior border of the pubic symphysis to the last coccygeal joint. The H line runs from the inferior border of the pubic symphysis to the posterior wall of rectum at anorectal junction. The M line is the perpendicular line from the anorectal junction to the PCL representing the vertical descent of the levator hiatus. The levator plate angle (LPA) is the angle between horizontal and the levator plate best fit line, as shown in Figure 2b [3]. Partial Correlations were performed between the PCL, M line, H line, LPA, GA, and MA using SPSSTM v25 while controlling for parity (the number of times a woman has given birth) by including it as a covariate. A p-value of less than 0.05 indicated a significant correlation.

3. Results

All correlational coefficients between variables are shown in Table 1. The H line was significantly correlated with the M line (ρ=0.575, p=0.001). In Figure 4, we see a positive correlation between the M line and GA (ρ=0.391, p=0.036). Additionally, as displayed in Figure 5, there is a 65

positive correlation between GA and MA (ρ=0.428, p=0.018). However, there is a lack of correlation between the M line and MA, as displayed in Figure 6. Lastly, the PCL and LPA were not significantly correlated with any variables.

Figure 4. M line (mm) and Gestational Age were positively correlated (ρ=0.391, p=0.036).

Figure 5. Maternal Age and Gestational Age were positively correlated (ρ=0.428, p=0.018).

Figure 6. M line and Maternal Age were not significantly correlated (ρ =-0.104)

66 Undergraduate Research at the Swanson School of Engineering

4. Discussion

The positive correlation observed between the levator hiatus descent (M line) and GA demonstrated a laxity in the levator ani muscles with increasing gestational age. Due to the increasing descent of the levator hiatus with GA, the levator ani muscles are more elongated. This could be the result of a maternal adaptation whereby these tissues remodel to minimize the risk of injury during delivery or it could indicate a response to increased intra-abdominal pressure that is known to occur with GA [3]. The adaptation/remodeling hypothesis is supported by research that has shown increases in hormones with GA that influence the muscles responsible for holding up the organs of the pelvis. The hormonal influence is likely driven by estrogen and progesterone, which both increase as gestational age increases. In literature it has been demonstrated that the levator ani muscles are activated by these two steroid hormones [5]. While it is unclear whether these are the sole cause of the decent observed in this study, the concomitant descent observed here and in other studies along with is a reported decrease in levator ani muscle contractions is suggestive [7]. Indeed, it has been observed that prolonged duration of high progesterone levels can lead to hypotonia, weak muscle tone [7]. Additionally, increases in other hormones lead to an increased ability for the levator ani muscles and connective tissue to stretch [7]. Thus, it would be speculated that these changes would correspond with changes in mechanical properties and shape of these tissues. This increase in M line is normal during gestation, however a large M line post-delivery or M line measurements that do not return to pre-pregnancy levels could be indicative of injury or an increased risk of developing a pelvic floor disorder. After the baby is delivered the steroid hormones decrease drastically. These two hormones help in reproducing the vascularity and proliferation of the cells in the muscle [5,6]. When injury occurs during vaginal delivery, the post pregnancy drop in hormones may make it harder for the muscles to recover and lead to atrophy instead of recovery. While the exact mechanisms and effects of specific hormones are not definitive, this study helps to establish the bounds for the normal changes that occur during pregnancy that can be useful for comparisons to post-delivery measurements in the future. As stated in the Results, there was no significant correlation between any variables and LPA. In past studies, the LPA has been correlated with an increase in gestational age and an increase in the M line [4]. The LPA is positively correlated in literature with the displacement of the perineal body, which is correlated with levator hiatus descent, and small bowel prolapse [4]. However, the lack of significance concerning the levator plate angle (LPA) in the current study is conflicting with the results in other literature, which could be due to the subjectivity of the definition of the horizontal line [3]. A repeat of this measurement will be done in a more objective manner in the future by measuring the LPA with respect to the PCL rather than the horizontal.

Ingenium 2021

One major limitation that existed within this study was correlation between GA and MA, meaning our younger patients tended to be imaged earlier in pregnancy and older women tended to be imaged later in pregnancy, which is not representative of pregnancy in general. Due to the retrospective and cross-sectional study design, there was no control over when the scans were taken and we only have one timepoint for each woman. In a prospective study we could control for maternal age and gestational age, where in our retrospective study these variables were random. In the future, a larger sample size and longitudinal study design should be considered.

5. Conclusions

There was a positive correlation between the levator hiatus descent (M line) and gestational age (GA), suggesting that levator ani muscle laxity increases with increasing gestational age. Due to the retrospective nature of this study, a correlation between GA and MA existed and served as a limitation. This study supported that as gestational age increases and hormones are released to prepare for childbirth, the muscles responsible for holding up the organs of the pelvic floor begin to change in their mechanical properties and shape likely in preparation for vaginal delivery.

6. Acknowledgements

The Swanson School of Engineering undergraduate research grant and NSF GRFP Grant #1747452 are acknowledged for supporting this research.

7. References

[1] “FastStats - Births and Natality.” Centers for Disease Control and Prevention, Centers for Disease Control and Prevention, 20 Jan. 2017, www.cdc.gov/nchs/fastats/ births.htm. [2] JO. Ashton-Miller, JA. Delancey. “On the Biomechanics of Vaginal Birth and Common Sequelae.” Annual Review of Biomedical Engineering, U.S. National Library of Medicine, 2009, pubmed.ncbi.nlm.nih.gov/19591614/. [3] JC. Salvador, MP. Coutinho. “Dynamic Magnetic Resonance Imaging of the Female Pelvic Floor-a Pictorial Review.” Insights into Imaging, SpringerOpen, 28 Jan. 2019, insightsimaging.springeropen.com/articles/10.1186/ s13244-019-0687-9. [4] Fielding, Julia R., et al. “Practical MR Imaging of Female Pelvic Floor Weakness.” RadioGraphics, 1 Mar. 2002, pubs.rsna.org/doi/10.1148/radiographics.22.2.g02mr25295. [5] Smith P, Heimer G, Norgren A, Ulmsten U. Steroid hormone receptors in pelvic muscles and ligaments in women. Gynecol Obstet Invest. 1990;30(1):27-30. doi: 10.1159/000293207. PMID: 2227608. [6] Marzieh Saei Ghare Naz, Fahimeh Ramezani Tehrani, Tahereh Behroozi-Lak, Farnaz Mohammadzadeh, Farhnaz Kholosi Badr and Giti Ozgoli. Journal: Research and Reports in Urology, 2020, Volume Volume 12, Page 179. DOI: 10.2147/RRU.S249611 [7] Jacobus Wijma, Annemarie E. Weis Potters, Ben T.H.M. de Wolf, Dick J. Tinga, Jan G. Aarnoudse, Anatomical and functional changes in the lower urinary tract during pregnancy, British Journal of Obstetrics and Gynaecology, Volume 108, Issue 7,2001, Pages 726-732, ISSN 03065456, https://doi.org/10.1016/S0306-5456(00)00123-6.


Development of a post-processing method for estimating solidification parameters from finite-element modeling of laser remelting in directed energy deposition of Ni-Mn-Ga magnetic shape-memory alloy single crystals Tyler Paplham, Jakub Toman, Markus Chmielus Department of Mechanical Engineering and Materials Science

Tyler Paplham

Tyler is a senior Materials Science and Engineering major from Buffalo, NY. He has worked in the Advanced Manufacturing and Magnetic Materials lab for almost two years. After graduation, he plans to pursue his PhD in Materials Science and Engineering to research innovative techniques for producing novel and existing materials for power magnetics applications.

Dr. Markus Chmielus has been an Associate Professor in the Mechanical Engineering and Materials Science Department since 2013. He previously worked at Cornell University (postdoc) and the Helmholtz Center for Materials and Energy in Germany (research scientist), and has degrees from the Markus Chmielus, Technical University of Berlin, Boise Ph.D. State University, and the University of Stuttgart in Materials Science and Aerospace Engineering. He now leads the Advanced Manufacturing and Magnetic Materials Laboratory (AM3), where his group members perform experimental additive manufacturing and post-processing research on functional magnetic and structural metals.

Significance Statement

This work aims to build on previous work in relating the user-controlled laser parameters to the thermal parameters which govern whether a single crystal can be achieved. Establishing a relationship between these two parameter sets will increase the reliability of single crystal production via directed energy deposition.

Category: Computational Research

Keywords: Finite element analysis, directed energy deposition, additive manufacturing

68 Undergraduate Research at the Swanson School of Engineering


Directed energy deposition is a subset of laser additive manufacturing which has been suggested as a feasible method for producing single crystals of metal alloys such as Ni-Mn-Ga magnetic shape-memory alloys. However, relating the thermal parameters, specifically the solidification front growth velocity V and thermal gradient G to the operational parameters, specifically the nominal laser power and travel velocity, is challenging. Therefore, this work aimed to use finite-element analysis (FEA) to create thermal models which allow for the determination of G and V for operational parameter combinations. Comparison with experimental data allowed for association of the range of G and V with the samples that best displayed single crystal growth. It was found that a high power and medium-low travel velocity is most likely to result in single crystal growth.

1. Introduction

Directed energy deposition (DED) is a type of laser additive manufacturing in which powdered material is injected via converging jets of an inert shield gas into the laser beam. This simultaneously melts the substrate and incident powder to join the two. DED is particularly attractive for production of macroscopic single crystals, which have numerous material property benefits over their polycrystalline counterparts. Of importance to predicting whether a single crystal is achievable by the usercontrolled laser parameters (e.g. laser power, laser travel velocity) are the thermal gradient G and interface growth velocity V. Mathematical models such as that developed by Gäumann et al. have become accepted as an analytical representation of melt pool resolidification. The Gäumann model utilizes the Rosenthal solution, which describes the gradient G for a moving point heat source, to analyze the columnar (corresponding to single-crystal growth) to equiaxed transition (CET) observed in alloy resolidification after melting by such a heat source [1]. For low gradient and high resolidification velocity, the microstructure is expected to be equiaxed, as stray grains will nucleate ahead of the solidification front and inhibit progression of the columnar resolidification front, which is detrimental for singlecrystal growth. However, for high thermal gradient and low resolidification front velocity, it is possible to prevent stray grains from nucleating. This allows the columnar front to advance uninhibited and a single crystal to be produced [2,3]. It should be noted that this solution is an approximate one, as it not only assumes a semi-infinite substrate but neglects the significant effects of latent heat of phase change and circulation within the melt pool, which are very difficult to model analytically.

Ingenium 2021

An important parameter used to predict columnar growth in the Gäumann model is Gn ⁄ V, where n is a material-specific constant determined from thermodynamic and kinetic analysis. This parameter is useful because it is independent of the vertical depth within the melt pool. A threshold value of Gn ⁄ V at which the phase fraction of grains nucleated ahead of the solidification front is exactly half can be used to delineate between columnar and equiaxed growth. The value of n used by Gaümann is 3.4, but the value for the Ni-Mn-Ga system is yet. However, for the present, n = 3.4 will be used, allowing the literature threshold of G 3.4 ⁄ V = 2.7 x 1024 [K3.4m-4.4s] to serve as an approximate indicator of single crystal growth. Finite element analysis (FEA) provides several distinct advantages over analytical models. Because an FEA model must have finite boundaries, the effects of finite body size and the heat transfer boundary conditions are accounted for, unlike the semi-infinite case of the analytical model. Conversely, the FEA model has important limitations. Whereas the analytical model can return the growth velocity and gradient as continuous functions of position, the FEA is by definition discrete, and thus functions are very difficult to obtain. Since many of the aforementioned complicating effects (phase change, circulation) are similarly infeasible to represent in FEA, this method cannot be expected to accurately calculate the magnitude of the growth velocity and gradient. Still, so long as only the input travel velocity and nominal power are varied between runs, the model should serve to rank the experimental parameter combinations by their solidification parameters. By comparison with the ranges posited by the analytical solutions, it may then be predicted which combinations are most likely to result in a single crystal. It is hypothesized that the value of G 3.4 ⁄ V, and thus likelihood of a single crystal, will be maximized at a parameter combination with middling values of laser power and travel velocity.

2. Methods

The experimental substrate was an austenitic Ni51 Mn24.4Ga24.6 single crystal. The substrate was a rectangular prism with approximate dimensions of 3.8 mm in the laser travel direction x 2.7 mm deep x 15 mm in the transverse direction. Eight combinations of laser power and travel velocity were used to create eight parallel tracks (see first column of Table 1). Finite-element analysis was carried out using ANSYS Workbench and ANSYS Mechanical with the

Moving Heat Source extension (version 4.1). This extension enabled the modeling of a moving Gaussian heat source. The deposition of new powder was excluded so as to determine only behavior caused by the melting and resolidification of the existing substrate. Temperature-dependent quantities for Ni2MnGa, specifically density ρ, thermal conductivity k, and specific heat Cp were obtained from Smith et al. (ρ) and Brillo et al. (k, Cp), and were input into ANSYS as tabular functions [4,5]. The model was made to be slightly larger (4 mm x 3 mm x 16 mm) than the experimental substrate to facilitate uniform meshing. The model consisted of a fine-mesh central section (0.02 mm spacing) for high accuracy in/near the melt pool surrounded by a coarse-meshed section (1 mm spacing) to reduce runtime. A beam radius of 0.285 mm was chosen based on specifications for the Optomec LENS used to carry out the experimental DED, although this is an approximation as a truly Gaussian heat distribution does not have a finite boundary. For each run, the laser power and travel velocity were chosen as in the experimental setup. The nominal powers given in Table 1 were attenuated by an absorptivity factor estimated by averaging the ratios of the experimental cross-sectional areas of each melt pool to those of models with corresponding parameters but perfect absorptivity. This resulted in an estimated absorptivity of α ≈ 0.266. Several additional boundary conditions were added as heat sinks: convection to various degrees across the top surface and sides of the model, conduction through the bottom surface to represent heat loss to the mount, and a radiation condition. After the solution was completed, a time step corresponding to the laser being approximately halfway down the track was chosen (as this most resembles a steady state) and the relevant temperature and directional heat flux data was exported to MATLAB. Since the thermal gradient is steep relative to the mesh spacing, it was assumed that the melt pool boundary could be defined to sufficient accuracy as all nodes within a small range (± 2 K) of the solidus temperature (1373 K). A Delaunay triangulation method, which is advantageous because it avoids “splinter triangles”, was used to interpolate a surface from the nodes. For each triangular element, the normal vector was calculated, normalized and dotted with the laser travel velocity to calculate local solidification front velocity (see Fig. 1). The gradient was determined by exporting the directional heat flux along each of the primary axes of the model at every node at the same time step at which the temperature data was collected. Since the exact conductivity used by the simulation at every


given point was unable to be exported, the Brillo data was directly imported to MATLAB, where linear interpolation was used to approximate the instantaneous conductivity for each given point. Since the heat flux q" and conductivity k are known, the gradient G = ∇T can be found from Fourier’s law of thermal conduction q" = -k∇T.

3. Results Laser Parameters [W – mm/s]

G × 103 [K/mm]

V [mm/s]

G3.4/V × 1024 [K3.4m-4.4s]

































Table 1: Average thermal gradients and growth velocities calculated for each of the parameter combinations, denoted in power-velocity form, along with the G3.4/V value.

It can be seen from Table 1 that the average growth velocity was only a fraction of the laser travel velocity, and that the gradient generally increased with laser travel velocity. For a given travel velocity, the gradient tended to decrease significantly with increasing laser power. The calculated values of G 3.4 ⁄ V were all found to be above the threshold 2.7 x 1024 except for 250 W, 1 mm/s, predicting that all experimental samples should display predominantly columnar growth, which indeed is what was observed. The predicted transition between vertical and horizontal growth is compared to the experimental results in Table 2 via comparison of the depth fraction of vertical growth relative to the melt pool depth. Note that no experimental fraction was determined for any of the 100 W tracks because no melt pool was observed as a result of this laser power.

Figure 1: Top view of melt pool for 250 W – 2.5 mm/s track. (a) Growth velocity V with normal vectors. Color corresponds to magnitude of growth velocity. (b) Thermal gradient G vectors at the boundary of the melt pool.

Therefore, the experimentally observed transition between vertical growth and growth parallel to laser travel direction was able to be predicted from the model by the depth in the melt pool that the x-component of the gradient is greater than the z-component. However, this transition is continuous in experiment, and as such the point where “horizontal” begins is only an estimate.

70 Undergraduate Research at the Swanson School of Engineering

Laser Power [W]

Travel Velocity [mm/s]

Sim. Fraction

Exp. Fraction

Percent Error









































*This sample had highly irregular melt pool shape likely due to proximity to edge of substrate resulting in less material to act as a heat sink. No clear vertical region was observable. Table 2: Depth fraction of vertical growth as predicted by simulation vs. the experimentally observed depth fraction. Percent error is defined here as deviation from the experimental depth.

Ingenium 2021

4. Discussion

The results in Table 1 are in strong agreement with the range of V and G values predicted for DED resolidification by Gäumann. This was expected for the growth velocity, as this is merely determined by the orientation of each infinitesimal area of the solidification front relative to the laser travel velocity. However, the fact that the thermal gradients match the predicted magnitude of 10 6 – 107 K/mm indicates that this model is a satisfactory representation of the analytical model used in the literature [2]. All eight parameter combinations resulted in a melt pool, despite the fact that the three 100 W tracks did not display any evidence of remelting in experiment. This can be attributed to the fact that the model does not factor in latent energy of fusion, i.e. without any additional energy cost for phase change. It is likely that, had this factor been able to be included in the model, no melt pool would have been observed in simulation. Even if there was some melting experimentally, the simulation results show an extremely shallow melt pool (<60 m) with a uniformly vertical thermal gradient orientation. This indicates that solidification might occur with an entirely planar interface, which would be difficult to observe under optical microscopy. High gradients were generally associated with high laser velocity. This means that there is no clear optimum highest gradient/lowest velocity combination for single crystal growth, with best results likely arising from a medium-high gradient and a medium-low velocity. If “optimal” is simply defined here as the maximum G 3.4 ⁄ V value, then 250 W – 1 mm/s is the optimum, which satisfies the prediction of a medium-high gradient and medium-low velocity. The 100 W tracks had higher G 3.4 ⁄ V values, but again this power is insufficient to induce melting at least at the travel speeds tested. As mentioned, all parameter combinations had a G 3.4 ⁄ V index greater than the threshold value of 2.7 x 1024 [K3.4m-4.4s]. The lowest reported value occurred in the 250 W – 10 mm/s run, which was also the only parameter combination which in experiment showed a transition from vertical growth not to horizontal cellular growth but to equiaxed dendritic growth. This is evidence that the G 3.4 ⁄ V index for this parameter combination has crossed some threshold value past which the growth mode has changed. A more conclusive explanation is limited by the fact that the correct value of n is not known for this particular alloy. As defined originally by Hunt and later utilized in the Gäumann model, this parameter is dependent on the linearity of the phase boundaries, the shape of the dendrite tips, and the Péclet number. While the chosen value n = 3.4 for the alloy in Gäumann is meant to be more realistic value than n = 2 used by Hunt (which assumes linear phase boundaries and hemispherical dendrite tips), the characteristics of the controlling phenomena would have to be determined for Ni51Mn24.4Ga24.6 for a more exact definition of n [2].

As can be seen, the model was relatively accurate in predicting the depth of transition at low velocities, but less so at high velocities. It is predicted that the 250 W – 10 mm/s track was associated with such a high error because the growth mode changed during processing, resulting in this metric no longer being a good predictor of transition. Still, the subjectivity of where the continuously changing growth direction is considered “horizontal” makes it incredibly easy for errors this large to arise. Additionally, while the general temperature distribution did show an elongated “tail” in the wake of the melt pool, the melt pool itself remained relatively circular. Since it is expected that the melt pool should also have a tail whose size depends on the magnitude of the velocity, this suggests that the convection and radiation boundary conditions were too high; this is unsurprising as they were somewhat arbitrarily applied.

5. Conclusion

Using the FEA model developed in ANSYS Workbench, it was found that since high thermal gradients were generally associated with high solidification velocity, there is no highest gradient/lowest velocity combination which maximizes G 3.4 ⁄ V and by extension probability of achieving a single crystal. Instead, maximal G 3.4 ⁄ V was found in the 250 W – 1 mm/s run, satisfying the hypothesis of a medium-high gradient and medium-low velocity. Future work could determine a more accurate value of n for this material system, which would enable a more reliable prediction of the parameters resulting in magnetic shape-memory alloy single crystal production.

6. Acknowledgements

This project was funded by NSF grant #1808082, the Swanson School of Engineering, and the Office of the Provost. This work was performed in part at the Materials Micro-Characterization Laboratory. Additional acknowledgement to the rest of the Advanced Manufacturing and Magnetic Materials lab for support during this project.

7. References

[1] Kurz, W., Bezençon, C., Gäumann, M. (2001) “Columnar to equiaxed transition in solidification processing”. Sci.& Tech. Adv. Mat., 2 (pp. 185-191) [2] Gäumann et al. (2001) “Single-crystal laser deposition of superalloys: processing-microstructure maps”. Acta Materialia, 49, 6 (pp. 1051-1062) [3] Gäumann et al. (1997) “Nucleation ahead of the advancing interface in directional solidification”. Mat. Sci. & Eng. A226-228 (pp. 763-769) [4] Smith et al. (2014) “Rapid actuation and response of Ni-Mn-Ga to magnetic-field induced stress”. Acta Materialia, 80 (pp. 373-379) [5] Brillo et al. (2011) “Thermophysical properties and thermal simulation of Bridgman crystal growth process of Ni-Mn-Ga magnetic shape memory alloys”. Int. J. Heat & Mass Trans. 54 (pp. 4167-4174)


Demonstrating the antibiofouling property of the Clanger cicada wing with ANSYS Fluent simulations Brady C. Pilsbury, Paul W. Leu The Laboratory for Advanced Materials at Pittsburgh, Department of Industrial Engineering Brady Pilsbury is originally from Warren, NJ. His interests in science, writing, and law drive him to become a patent agent.

Brady C. Pilsbury

Dr. Leu is an Associate Professor in the Department of Industrial Engineering and the Department of Mechanical Engineering and Materials Science. He received his BS in Mechanical Engineering at Rice University in 2002, his MS from Stanford University in 2004, and his PhD from Stanford Paul W. Leu, Ph.D. University in 2008. Dr. Leu’s lab research focuses on designing and understanding advanced materials by computational modeling and experimental research.

Significance Statement

Microdroplets released upon coughing, sneezing, or speaking are one source of spread for human viruses. Simulation results illustrate how the introduction of nanostructures inspired by the Clanger cicada wing to a surface can limit microdroplet adhesion and help reduce the spread of pathogens due to contact with surfaces.

Category: Computational Research

Keywords: Nanostructures, Antibiofouling, ANSYS Fluent

72 Undergraduate Research at the Swanson School of Engineering


The Clanger cicada wing is known to have an antibiofouling effect related to the particular pattern of nanostructures on the wing surface. Understanding the mechanisms that create this effect allows for the development of effective antibiofouling nanostructured surfaces, which can help reduce the spread of pathogens. Water repellency is well established as one such mechanism, which this study seeks to validate with the use of water droplet simulations conducted in ANSYS Fluent. The 2D simulations show how the addition of Clanger cicada wing inspired nanostructures to a flat surface impacts surface hydrophobicity and water droplet contact angle hysteresis. A pre-existing empirical dynamic contact angel model was implemented to capture differences in contact angle hysteresis between the flat and nanostructured surface. The application of such a model to a nanostructured surface has not been previously attempted within ANSYS Fluent. The water droplet achieved a greater contact angle and lower contact angle hysteresis on the nanostructured surface when compared to the flat control, indicating that the drop entered the Cassie-Baxter state and would roll off the surface at low incline angles. These results support the repellence of microdroplets containing pathogens as one mechanism through which the Clanger cicada wing achieves a strong antibiofouling effect.

1. Introduction

The wings of insects like the Clanger cicada (Psaltoda claripennis) have been observed to have highly hydrophobic and antibiofouling properties due to the presence of nanopatterns on the surface of their wings [1]. The behavior of microdroplets on such nanopatterned surfaces has implications for public health, since viruses and other pathogens can be transferred through microdroplets. Surfaces designed to strongly repel and roll off microdroplets can help reduce the spread of pathogens. The water repellency of a surface is often characterized by a water droplet contact angle measurement. This angle refers to the angle formed by the gas-liquid and solid-liquid interfaces at the points along the bottom edge of the droplet where all three phases meet. Convention dictates that a surface is hydrophilic when its contact angle is less than 90°, hydrophobic when the contact angle is greater than 90°, and superhydrophobic when the contact angle exceeds 150° [2]. Nanoscale surface roughness can allow a droplet to enter the Cassie-Baxter state, which is characterized by the droplet resting atop protrusions from the surface and bridging over lower elevation regions of the surface instead of penetrating them [2]. Water droplets in the Cassie-Baxter state tend to have a high contact angle, low contact angle hysteresis, and an associated low run off angle [2]. Contact angle hysteresis refers to the

Ingenium 2021

variation in the contact angle of droplets as they move laterally across a surface [2]. Figure 1 shows a generic water droplet in the Cassie-Baxter state when atop a surface containing nanopillars and displays how the contact angle is typically measured.

considered a droplet of water 2 µm in diameter initialized as a sphere just above the surface with an initial downward velocity of 0.3 m/s. It is assumed that the general wetting behavior of the droplet scales to larger droplets, though the time-scale for droplet spreading and recoiling likely varies with droplet size. The water droplet had the following material properties: density r = 1000 kg/m3, viscosity m = 1 × 10 -3 Pa·s, and surface tension s = 7. 20 × 10 -2 N/m. Gravity was set to 9.8 m/s2 in the negative y-direction at the beginning of all simulations.

2.2 Simulated Surfaces

Figure 1. Shows a generic water droplet suspended above a set of nanopillars in the Cassie-Baxter state with the characteristic high contact angle.

Surfaces on which water droplets reside in the Cassie-Baxter state are expected to make good antibiofouling surfaces. Microdroplets that cannot penetrate into the nanostructures will tend to roll off without leaving residual liquid behind, limiting the ability of any bacteria or viruses present in the droplets to proliferate on the surface. This project seeks to demonstrate the anti-biofouling nature of the Clanger cicada wing by simulating the impact of water droplets onto an array of nanopillars inspired by the cicada wing’s nanopattern. Results were obtained through 2D simulations using the Volume of Fluid (VOF) model in ANSYS Fluent. Previous research has used the VOF model in ANSYS Fluent to study the impact of adjusting the ratios between height, width, and spacing of nanopillars on droplet wetting behavior [3]. Other researchers have developed and implemented dynamic contact angle models in ANSY Fluent that capture changes in a droplet’s contact angle as it moves across a flat surface [4]. This project builds off previous work by applying a dynamic contact angle model to compare the differences in contact angle hysteresis and general wetting behavior between a flat surface and a nanostructured surface inspired by the Clanger cicada wing. The simulations support the already well-established theory that water droplet repellency is a key mechanism by which the Clanger cicada wing achieves its antibiofouling effect and offer a succinct demonstration of how nanoscale modification of a surface can help yield an antibiofouling effect [1].

2. Methods 2.1 Experimental Set Up

All simulations were completed in 2D using the VOF model implemented in ANSYS Fluent. Each simulation

The simulations compared the behavior of a water droplet when impacting a flat surface and a nano patterned surface. The nano patterned surface consisted of an array of rectangular pillars 200 nm in height, 60 nm in width, and with 170 nm of space between the center of pillars. These dimensions correspond to the dimensions of the nanostructures found on the Clanger cicada wing, although the rectangular shape of the simulated pillars does not match the rounded cone shape of the pillars on the cicada wing [1]. The flat surface contained no protrusions.

2.3 Dynamic Contact Angle Model

A dynamic contact angle model was used during simulations and compared to a static contact angle model to verify it was functioning as intended. The dynamic contact angle model is based on the tendency of a generic droplet moving across a surface to see an increase in contact angle on the advancing side of the droplet and a decrease in contact angle on the receding side, as seen in Figure 2.

Figure 2. Schematic of a water droplet rolling across a flat surface, with a greater contact angle on the advancing side of the droplet and lower contact angle on the receding side.

Usually, the effect of droplet motion on contact angle is constrained by a maximum advancing contact angle and a minimum receding contact angle, which form upper and lower bounds around the static contact angle [4]. The values of these angles will depend on the balance of interfacial tensions associated with the water-vapor and water-solid interfaces [2]. For a flat surface, the balance of these tensions is largely determined by the chemical composition of the surface. However, textured surfaces that disrupt the liquid-solid interface will also disrupt the balance of interfacial tensions, yielding equilibrium, 73

advancing, and receding contact angles that may differ from an otherwise chemically identical flat surface [2]. The range of angles set by the maximum advancing and minimum receding contact angle is often used to characterize the contact angle hysteresis. It is possible for a droplet to achieve a contact angle at any value between the maximum and minimum bounds as it transitions between states of motion. Capturing the dynamic behavior of contact angle within that range can be difficult to simulate, but the ANSYS Act extension developed by Yokoi et. al. captures it with the equations

qd(Ucl ) = min (qe+(Ca/Ka ), qmda ) if Ucl ≥0

qd(Ucl ) = min (qe+(Ca/Kr ), qmdr ) if Ucl ≤0 Ca=mUCL /s

It was expected that a greater contact angle and lower contact hysteresis would be observed on the pillared surfaces (Models III and IV for contact angle and lower contact hysteresis, respectively), relative to the flat surface (Models I & II). Since Models I and III used a static contact angle model, they were not expected to show any contact angle hysteresis and instead maintain the same contact angle over the course of the simulation. The approximated advancing, receding, and static apparent contact angles for each Model are shown in Table 1.

(1) (2)

Model I










3. Results

and Ucl is the velocity of the contact line, qd is the dynamic contact angle, qe is the static equilibrium contact angle, qmda is the maximum advancing contact angle, qmdr is the minimum receding contact angle, Ka is the advancing material related constant, and Kr is the receding material related constant [4]. Ca is the Capillary number and a dimensionless parameter that captures the relative effect of viscous drag forces (numerator) and surface tension or capillary forces (denominator). Yokoi’s model was designed as an empirical model that retroactively fits itself to experimental data. This limits the applicability of the model, but still allows for fair analysis of the general impact that nanostructures have on contact angle hysteresis. For the purposes of this simulation, the static contact angle was set at 90°, the maximum advancing angle at 114°, the minimum receding angle at 52°, Ka at 9 x 10 -9, and Kr at 9 x 10 -8 in order to best match the experimental results found by Yokoi et al. for a water droplet on a chemically treated silicon surface [4]. This limits the applicability of any values calculated during the simulation to specialized silicon wafers, but the general effect of the nanostructures is applicable to a broader range of chemical substrates. The dynamic contact angle model was tested against a static contact angle model in which the static contact angle was set at 90°, equivalent to the static equilibrium contact angle in the dynamic model.

Advancing Contact Angle

Receding Contact Angle

Static Contact Angle









Table 1 Shows the maximum advancing contact angle, minimum receding contact angle, and static contact angle observed under each model.

Under Model I, the droplet maintained an apparent contact angle of 90° throughout the simulation as it spread and recoiled on the surface, which can be seen in Figure 3.

2.4 Simulated Conditions

Four test conditions were considered; Model I consisted of the droplet impacting a flat surface with the static contact angle model, Model II considered a flat surface with the dynamic contact angle model, Model III considered the pillared surface with the static contact angle, and Model IV considered the pillared surface with the dynamic contact angle model. The models were compared based off of the apparent contact angle of the water droplet, which refers to the angle of the liquid-gas interface relative to the horizontal, as opposed to the liquid-solid interface, which, strictly speaking, would become vertical if the liquid partially wrapped around the rectangular nanopillars. 74 Undergraduate Research at the Swanson School of Engineering

Figure 3. The shape of the droplet under each model is shown in snapshots taken every 50 ns for 300 ns under four model conditions: I – flat surface, static contact angle II – flat surface, dynamic contact angle III – nanopillar surface, static contact angle IV – nanopillar surface, dynamic contact angle

Ingenium 2021

Model II allowed the droplet to achieve an advancing contact angle of 114° as the droplet spread out over the surface, which was occurring after 50 ns. The droplet later reached a receding contact angle of approximately 80° as the droplet recoiled, which was occurring after 100 ns. The dynamic contact angle approached the static equilibrium contact angle of 90° as the droplet came to rest for the rest of the simulation. Under Model III, the droplet achieved an apparent contact angle of 140° and the droplet did not spread across the surface nearly as much as it had on the flat surface. Model IV yielded an identical static contact angle to Model III and saw limited droplet spread over the surface. Model IV also yielded a smaller range of contact angle hysteresis compared to the flat surface in Model II. No new droplet behaviors occurred in any model after 300 ns.

4. Discussion

The droplets simulated on the nanopillars demonstrated behavior characteristic of the Cassie-Baxter state, achieving high contact angles and low contact angle hysteresis. Despite the static contact angle being set to 90° in both Model I and model 3, the apparent contact angle in Model III, when the droplet was on the array of nanopillars, was estimated to be much higher at 140°. Models II and IV implemented the same dynamic contact angle model, but the droplet in Model II had a contact angle that varied by up to 32°, while the droplet on the nanopillars in Model IV maintained a high apparent contact angle with only a 12° difference between its advancing and receding angles. These results can be attributed to the gaps between pillars in Models III and IV disrupting the liquid-solid interface, which limits the droplets overall adhesion to the surface. In turn, this allows cohesive forces to dominate such that a more consistent droplet shape is maintained. The higher apparent contact angle and lower apparent contact angle hysteresis observed on the pillars is in accordance with the typical behavior of droplets in the Cassie-Baxter state, and is strongly associated with a lower run-off angle, meaning that a droplet could slide off of the pillared surface more easily [2]. This feature lends the Clanger cicada wing its self-cleaning and antibiofouling properties.

5. Conclusions

As the risks of antibiotic resistant bacteria and global virus outbreaks have become increasingly pronounced in recent years, there is a heightened importance in understanding how the spread of pathogens can be mitigated. One mechanism to limit the spread of pathogens via surface contact is through the creation of strongly hydrophobic surfaces that strongly repel microdroplets that may contain bacteria or viruses. This antibiofouling effect can be achieved via the fabrication of nanostructures on surfaces that suspend microdroplets in the Cassie Baxter state. In this state, the liquid-solid interface of the droplet is disrupted, weakening the overall adhesion of the droplet to the surface. This weak adhesion is characterized by a high contact angle, low contact angle hysteresis, and a low runoff angle. Simulations conducted in ANSYS Fluent confirmed that the nanostructures of the Clanger cicada wing are capable of drastically increasing the hydrophobicity of a surface while reducing contact angle hysteresis. This mechanism accounts for the strong antibiofouling nature of the Clanger cicada wing and can be applied when fabricating surfaces to inhibit the ability of pathogens to proliferate on them.

6. Acknowledgments

This project was funded by Swanson School of Engineering and the Office of the Provost at the University of Pittsburgh and was conducted under the guidance of Paul W Leu.

7. References

[1] S. Podogin et al. Biophysical Model of Bacterial Cell Interactions with Nanopatterned Cicada Wing Surfaces. Biophysical Journal, 103 (2013) 835-840. https:// doi.org/10.1016/j.bpj.2012.12.046 [Accessed Nov 20,2020]. [2] E. Celia et al. Recent advances in designing superhydrophobic surfaces. Journal of Colloid and Interface Science. 402 (2013) 1-18. https://doi.org/10.1016/j. jcis.2013.03.041 [Accessed Nov 20, 2020]. [3] Z. Yang et al. Numerical Study on Size Effects in the Wettability of Textured Surfaces. IOP Conf. Ser.: Earth Environ. Sci, 252 (2019). http://doi.org/10.1088/17551315/252/2/022105 [Accessed Jan 7, 2021]. [4] K. Yokoi et al. Numerical studies of the influence of the dynamic contact angle on a droplet impacting on a dry surface, Physics of Fluids, 21 (2009). http://doi. org/10.1063/1.3158468 [Accessed: Nov 20,2020].


Identifying potential mammalian heme/CO sensors through a bioinformatic analysis of the Per-Arnt-Sim (PAS) protein domain JonPaul Plesniaka, Matthew Dentb, Jason Roseb,c, Jesus Tejerob,c a

Chemical Engineering, b Heart, Lung, Blood, and Vascular Medicine Institute, c Division of Pulmonary, Allergy and Critical Care Medicine JonPaul Plesniak is a junior chemical engineering major and bioengineering minor at the University of Pittsburgh. He has a passion for innovation and discovery and wants to pursue these interests with a career in product design and development.

JonPaul Plesniak

Matt Dent is a postdoctoral researcher in Dr. Mark Gladwin’s lab in the Vascular Medicine Institute at the University of Pittsburgh. He earned his Ph.D. in chemistry at the University of WisconsinMadison in 2019, where he used biochemical techniques to characterize heme-mediated carbon monoxide Matthew Dent sensing in microbial transcription factors. Matt’s current research interests include characterizing novel CO signaling pathways in humans.

Dr. Rose received his BSE in Biomedical Engineering at the University of Michigan in 2006, his MD at Wayne State University in 2010, and MBA from Carnegie Mellon University in 2017. He completed his internal medicine residency at Duke University Medical Center and pulmonary and critical care Jason Rose, M.D. medicine fellowship at the University of Pittsburgh. He joined the University of Pittsburgh faculty in 2016 and is currently an Assistant Professor of Medicine. His research interests focus on discovering and developing new human therapeutics, particularly in medical countermeasures and inhalational respiratory diseases. Dr. Tejero received his degree in Organic Chemistry at the University of Zaragoza, Spain in 1998 and earned his PhD in Biochemistry at the University of Zaragoza in 2004. He completed postdoctoral work at the Cleveland Clinic and the University of Pittsburgh, and moved to a faculty position at the University of Jesus Tejero, Ph.D. Pittsburgh in 2011. Dr. Tejero is currently an Associate Professor in the Department of Pulmonary, Allergy and Critical Care Medicine.

Significance Statement

The role of carbon monoxide (CO) in human physiology is not well understood. While high concentrations CO are highly toxic, recent evidence points toward the importance of endogenously produced CO in cellular signaling. Few molecular targets for CO have been characterized in eukaryotes. This study has identified potential mammalian heme/CO sensors for further analysis and characterization.

Category: Computational Research

Keywords: carbon monoxide, PAS domain, heme, phylogenetic analysis

76 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021


Carbon monoxide (CO) is a gaseous molecule commonly found in the environment and also produced endogenously by organisms as a product of heme degradation. A growing body of evidence suggests that CO functions as an important signaling molecule that exhibits cytoprotective and anti-inflammatory properties; however, the precise molecular targets of CO in mammals and other higher organisms are not well characterized. The Per-Arnt-Sim (PAS) domain is a ubiquitous protein fold, found in organisms ranging from bacteria to mammals, that often facilitates sensing of environmental stimuli. Several CO-sensing PAS domains that utilize the iron-containing cofactor heme have been characterized in bacteria; however, few such domains have been studied in higher organisms. The goal of this study is to identify potential heme-based sensors in mammals through bioinformatic analysis of PAS domain proteins. Both structural and sequence alignment tools were implemented to identify phylogenetic relationships between well-studied PAS proteins and uncharacterized PAS proteins. By establishing phylogenetic relationships between PAS proteins of known function and uncharacterized proteins, we identified the mammalian PAS proteins Pask, Kcnh1-8, Pde8A, Hif/Epas1, Sim, Pasd1, and NcoA as potential heme-based sensors.

and photoreceptors, circadian clock proteins, voltage-gated ion channels, and regulators of hypoxia response [6]. Many of these proteins have multiple PAS domains, called PAS repeats, in which each PAS domain serves a different role, often environmental sensing or mediation of protein-protein interactions [7]. Heme-based sensors have been identified in many prokaryotic proteins. The direct oxygen sensor protein from E. coli (DosP) has been identified as a heme-based O2-sensing phosphodiesterase [8]. Likewise, the FixL protein has been identified in bacteria as an oxygen sensor that regulates expression of nitrogen fixation genes in response to hypoxia [9]. The regulator of CO metabolism (RcoM) protein also employs heme to sense CO and regulates expression of genes required for CO metabolism in a variety of microorganisms [10]. Recent studies have broadened the scope of heme-based sensors to include mammalian proteins [6]. A growing body of evidence points to PAS domains involved in regulation of the circadian rhythm of mammals as heme-based sensors: NPAS2 and its homolog CLOCK have been shown to bind heme and may function as heme-based CO sensors [11]. Thus, a feedback mechanism potentially exists that is directly dependent on cellular CO concentration [12].

1. Introduction

CO is produced in the atmosphere as a byproduct of incomplete combustion and is often referred to as the “silent killer” due to its toxicity and lack of color, smell, or taste. Exposure to high levels of CO negatively impacts oxygen transport and mitochondrial respiration, ultimately leading to hypoxia, tissue damage, and death [1]. However, CO is also produced endogenously as a byproduct of heme degradation, and low concentrations of endogenously produced CO play a significant role in maintaining homeostasis in humans [2]. CO has been implicated in regulating numerous physiological processes including vasodilation, mitochondrial function, ion-channel activity, and inflammation pathways [3]. At low concentrations, CO exhibits cytoprotective properties in acute lung injury animal models, demonstrating promise for CO inhalation for anti-inflammatory and therapeutic use [4]. Despite emerging roles in signaling and therapeutics, the specific molecular targets of CO have not been well characterized. The PAS domain fold is a sequence of ~100 amino acid residues that functions as an input module able to sense gases like oxygen or CO, redox potential, light, and other environmental stimuli. PAS domain secondary structure consists of a series of antiparallel β sheets and α helices in a β-β-α-α-α-β-β-β arrangement (Figure 1) [5]. While the secondary structure is well-conserved, less than 20% average pairwise sequence identity exists between amino acids of PAS proteins. Given this low sequence similarity, sequence-based alignments and bioinformatic analyses prove difficult. Organisms from all kingdoms of life contain proteins bearing a PAS domain, and families of PAS proteins include serine/threonine kinases, chemoreceptors

Figure 1: (top) Secondary structure map of the PAS domain. (bottom) Crystal structure of human CLOCK protein PAS domain (6QPJ) [11]. Structure visualized with the PyMOL Molecular Graphics System, Version 2.0 (Schrödinger, LLC).

Although several PAS domains have been characterized as heme-based sensors, we hypothesize that there may be heretofore unidentified heme-based sensors from the PAS family due to the large sequence diversity in this family. This study utilizes bioinformatics analyses to identify putative heme-based sensors in mammals that bear a PAS domain. Using pairwise sequence alignments, we identified regions of conservation between PAS proteins of known and unknown function and established phylogenetic relationships. To address the problem of low sequence identity between PAS domains, we validated these relationships using pairwise structural alignment of structurally characterized PAS domains. To identify putative heme-binding PAS domains that may act as CO sensors, we established phylogenetic connections between uncharacterized PAS proteins and proteins characterized as heme-based CO sensors.


2. Methods 2.1 Establishing a consensus sequence for PAS domains

645 PAS domain-containing proteins were identified by cross-referencing online databases (UniProt, InterPro and ProSite). Annotated PAS domain sequences were ~70 amino acid residues in length. To validate PAS domain sequence annotations, we reviewed available crystallographic data for PAS domain proteins and correlated sequences and secondary structural elements. The appropriate sequence length that fully encompasses the β-β-α-α-α-β-β-β secondary structure characteristic of PAS domains ranged from 100-150 amino acid residues. As a result, we manually expanded the 645 annotated PAS protein sequences to include all residues in the secondary PAS fold. We conducted this annotation process through visualization of crystallographic 3-D structures using the PyMOL Molecular Graphics System, Version 2.0 (Schrödinger, LLC).

2.2 Sequence Alignment

Given the lack of sequence identity between PAS domains, we subdivided the full group of 645 PAS domains into protein sequences from nonbacterial and bacterial organisms to facilitate sequence alignment. We further divided each of these groups based on preliminary sequence alignments, giving rise to 4 sub-groups in total. We used the ClustalW multiple sequence alignment algorithm to align groups of sequences [13]. We identified regions of conservation corresponding to secondary structural elements within each sub-group alignment and used these conserved regions to manually curate a combined alignment with all 645 proteins. From there, we generated a representative list of 219 proteins for easier visual analysis by removing homologs with identical amino acid sequences. Finally, we created a phylogenetic tree from this alignment using the MEGA-X software. MEGA-X implemented a maximum likelihood statistical method, a Jones-Taylor-Thornton substitution model, and uniform rates among sites [14].

sequence-based and structure-based phylogenetic trees to locate potential candidate proteins positioned near experimentally characterized heme-based sensors.

3. Results

To identify relationships between known heme-based sensors and uncharacterized PAS domain proteins, we compared PAS domain sequences of all known PAS domain-containing proteins. Figure 2 displays a phylogenetic tree based on pairwise sequence analysis, and proteins in the tree cluster into identifiable regions. Mammalian proteins cluster into two distinct groups that correspond to one of two PAS repeat domains. Some noteworthy exceptions to this trend are the potassium voltage channels Kcnh1-8, which cluster near the light sensors from plants, and Pask and Pde8A, which are positioned near the known bacterial heme-based sensors RcoM, DosP, and FixL. Bootstrap values are lowest for the large cluster mostly comprised of bacterial PAS domain proteins (bootstrap score < 50). This lack of distinct clustering reflects large sequence divergence amongst PAS proteins from a diverse set of microbial organisms. Regions of higher bootstrap values (bootstrap score > 50) around the mammalian PAS repeats, the light sensors, and some of the outer nodes of the bacterial regions are characteristic of higher sequence similarity. These clusters of high sequence similarity may reflect functional similarities, particularly in the mammalian repeat 1 cluster, where four PAS domain proteins have been identified that bind heme.

2.3 Structural Alignment

To validate pairwise sequence alignment, we carried out a pairwise structural analysis using available crystallographic data of 63 PAS domains (156 proteins from the representative list were not structurally characterized). We employed the DALI server to superimpose pairs of protein structures and measure distances between α carbons in pairwise alignments [15]. The DALI server generated a pairwise distance matrix, and Z-scores from pairwise alignments were used to generate a phylogenetic tree using the MEGA-X software [14].

2.4 Data Processing

We carried out a bootstrap analysis for the sequence-based tree to determine the statistical significance of each node [16]. The MEGA-X software randomly generated 1000 trees and output values correspond to the original tree nodes [14]. Bootstrap values indicate the percent of the 1000 randomly generated trees in which the arrangement from that node occurred. A higher bootstrap value corresponds with a stronger statistical support for the clade. Additionally, we qualitatively inspected 78 Undergraduate Research at the Swanson School of Engineering

Figure 2. Representative sequence-based phylogenetic tree for PAS domains (219 proteins). Characterized heme-based sensors are highlighted in light red, and putative heme-based sensors are highlighted in blue. Bootstrap values for individual nodes are highlighted in purple (<50), cyan (50-75), or yellow (75-99).

Pairwise alignments of PAS domain protein structures reveal clustering similar to that observed in the phylogenetic tree derived from pairwise sequence alignments. Specifically, the structure-based phylogenetic tree displays clustering of two mammalian PAS repeats and a group comprised primarily of light sensors (Figure 3). Consistent

Ingenium 2021

with the sequence-based tree, several bacterial PAS proteins emerge without distinct phylogenetic clustering in the structure-based tree. The similarities in PAS domain clustering observed between structure- and sequence-based phylogenetic trees serve to validate our approach of identifying putative mammalian heme sensors from clusters of proteins with sequence homology and structural similarities.

5. Conclusion

A growing body of research suggests that endogenously-produced CO functions as signaling molecule with cytoprotective and anti-inflammatory properties in mammals. Despite these well-documented effects, the molecular mechanisms by which CO acts as a signaling molecule are poorly understood. In this study, we used sequence and structural alignment techniques to establish phylogenetic relationships among proteins bearing a PAS domain. Several PAS proteins, found in organisms ranging from bacteria to humans, have been characterized as hemoprotein-based gas sensors. We identified a number of poorly-characterized PAS proteins that exhibit sequence and structural similarities to known heme-binding gas sensors: Pask, Kcnh1-8, Pde8A, Hif/Epas1, Sim, Pasd1, and NcoA. In our ongoing research, we are expressing and purifying these soluble PAS domains to homogeneity and probing heme binding properties using biochemical and spectroscopic techniques. For those proteins that exhibit heme binding properties, we will probe CO-dependent changes in protein function in order to identify bona fide hemoprotein-based CO sensors. By identifying specific molecular targets of CO, we may be able to better characterize mammalian CO signaling pathways relevant to human health and disease.

6. Acknowledgments Figure 3. Structure-based phylogenetic tree for PAS domains (63 proteins). Characterized heme-based sensors are highlighted in light red, and functional clusters are outlined.

4. Discussion

Analysis of sequence- and structure-based phylogenetic trees reveals the evolutionary origins and functional clustering of mammalian PAS domains. The mammalian proteins Pask, Kcnh1-8, and Pde8A do not cluster with other mammalian proteins and instead cluster near plantbased light sensors and bacterial PAS domains in the sequence-based tree. That these mammalian proteins cluster with lower organisms suggests that Pask, Kcnh1-8, and Pde8A may be the most ancient mammalian PAS domains. Most of the other mammalian PAS domains cluster into one of two groups of PAS repeats in both sequence- and structure-based trees. Many uncharacterized proteins in mammalian PAS repeat 1 fall near known heme-based sensors of the circadian rhythm, suggesting that some of these uncharacterized proteins may bind heme. Pde8A and Pask also lie near the known heme-based, bacterial sensors DosP, FixL, and RcoM, suggesting that these mammalian proteins may also interact with heme. Based on their positioning near known hemebinding PAS proteins, the following mammalian proteins may act as heme-based sensors: Pask, Kcnh1-8, Pde8A, Hif/Epas1, Sim, Pasd1, and NcoA. Excitingly, a study recently demonstrated that one of the Kcnh proteins, Kcnh7, exhibits heme-binding characteristics, supporting our predictive approach [17].

This project was funded by the University of Pittsburgh Swanson School of Engineering and the Office of the Provost. Project opportunity was provided by the Summer Undergraduate Research Internship (SURI). The authors thank Dr. Mark Gladwin for providing his insight into the biological relevance of putative CO sensors and for use of his lab space.

7. References

[1] Rose, J. J., Wang, L., Xu, Q., McTiernan, C. F., Shiva, S., Tejero, J., and Gladwin, M. T. (2017) Carbon Monoxide Poisoning: Pathogenesis, Management, and Future Directions of Therapy. Am. J. Respir. Crit. Care Med. 195, 596-606 [2] Wu, L., and Wang, R. (2005) Carbon Monoxide: Endogenous Production, Physiological Functions, and Pharmacological Applications. Pharmacol. Rev. 57, 585630 [3] Ryter, S. W., Ma, K. C., and Choi, A. M. K. (2018) Carbon monoxide in lung cell physiology and disease. American Journal of Physiology-Cell Physiology 314, C211-C227 [4] Motterlini, R., and Otterbein, L. E. (2010) The therapeutic potential of carbon monoxide. Nature Reviews Drug Discovery 9, 728-743 [5] Henry, J. T., and Crosson, S. (2011) Ligand-binding PAS domains in a genomic, cellular, and structural context. Annu. Rev. Microbiol. 65, 261-286 [6] McIntosh, B. E., Hogenesch, J. B., and Bradfield, C. A. (2010) Mammalian Per-Arnt-Sim Proteins in Environmental Adaptation. Annu. Rev. Physiol. 72, 625-645 79

[7] Ryter, S. W., Alam, J., and Choi, A. M. K. (2006) Heme Oxygenase-1/Carbon Monoxide: From Basic Science to Therapeutic Applications. Physiol. Rev. 86, 583-650 [8] Shimizu, T. (2013) The Heme-Based Oxygen-Sensor Phosphodiesterase Ec DOS (DosP): Structure-Function Relationships. Biosensors (Basel) 3, 211-237 [9] Gong, W., Hao, B., Mansy, S. S., Gonzalez, G., Gilles-Gonzalez, M. A., and Chan, M. K. (1998) Structure of a biological oxygen sensor: a new mechanism for heme-driven signal transduction. Proc Natl Acad Sci U S A 95, 15177-15182 [10] Kerby, R. L., Youn, H., and Roberts, G. P. (2008) RcoM: A New Single-Component Transcriptional Regulator of CO Metabolism in Bacteria. J. Bacteriol. 190, 3336-3343 [11] Freeman, S. L., Kwon, H., Portolano, N., Parkin, G., Venkatraman Girija, U., Basran, J., Fielding, A. J., Fairall, L., Svistunenko, D. A., Moody, P. C. E., Schwabe, J. W. R., Kyriacou, C. P., and Raven, E. L. (2019) Heme binding to human CLOCK affects interactions with the E-box. Proc Natl Acad Sci U S A 116, 19911-19916 [12] Airola, M. V., Du, J., Dawson, J. H., and Crane, B. R. (2010) Heme binding to the Mammalian circadian clock protein period 2 is nonspecific. Biochemistry 49, 43274338 [13] Thompson, J. D., Gibson, T. J., and Higgins, D. G. (2002) Multiple sequence alignment using ClustalW and ClustalX. Curr Protoc Bioinformatics Chapter 2, Unit 2.3 [14] Stecher, G., Tamura, K., and Kumar, S. (2020) Molecular Evolutionary Genetics Analysis (MEGA) for macOS. Mol. Biol. Evol. 37, 1237-1239 [15] Holm, L., and Rosenström, P. (2010) Dali server: conservation mapping in 3D. Nucleic Acids Res. 38, W545 - W549 [16] Dopazo, J. (1994) Estimating errors and confidence intervals for branch lengths in phylogenetic trees by a bootstrap approach. J. Mol. Evol. 38, 300-304 [17] Burton, M. J., Cresser-Brown, J., Thomas, M., Portolano, N., Basran, J., Freeman, S. L., Kwon, H., Bottrill, A. R., Llansola-Portoles, M. J., Pascal, A. A., Jukes-Jones, R., Chernova, T., Schmid, R., Davies, N. W., Storey, N. M., Dorlet, P., Moody, P. C. E., Mitcheson, J. S., and Raven, E. L. (2020) Discovery of a heme-binding domain in a neuronal voltage-gated potassium channel. The Journal of biological chemistry 295, 13277-13286

80 Undergraduate Research at the Swanson School of Engineering

Topological descriptor selection for a quantitative structure-activity relationship (QSAR) model to assess PAH mutagenicity Caitlin Sextona, Trevor Sleight, P.E.a,b, Carla Ng, Ph.D.b, Leanne Gilbertson, Ph.D.a Gilbertson Lab Group, bNg Lab Group, Department of Civil and Environmental Engineering a

Caitlin Sexton

Caitlin Sexton is a junior chemical engineering student originally from Allentown, PA. Her interests include the various aspects of sustainability in application to the chemical industry. She aims to use her research experience in environmental hazards and data analysis as a guide in her post-grad career. Trevor Sleight is a 3rd year Ph.D. student, co-advised by Dr. Ng and Dr. Gilbertson. His research interest include environmental health, data analysis and biodegradation.

Trevor Sleight


Polycyclic aromatic hydrocarbons (PAHs) are an abundant byproduct of industrial and natural pyrogenic processes. PAHs tend to persist in soil, providing a rich nutrient source for degrading bacteria. The degradation process may lead to the formation of toxic metabolites, however, there is limited research examining the hazards of transformation products in soil. Currently, the Environmental Protection Agency (EPA) classifies 16 toxic PAHs as priority pollutants without addressing the harmful metabolites. This project aims to select descriptors for a quantitative structure-activity relationship (QSAR) model based upon a data set containing TA98 Ames test known mutagens and non-mutagens. A logistic regression model determined 20 significant descriptors representing the molecular features linked to mutagenicity classification. Of these 20 descriptors, the number of rings larger than 12 members containing oxygen, (nFG12HeteroRing), the average centered Broto-Moreau autocorrelation (AATSC6c), and the z-modified information content index (ZMIC4), had the most significant link to mutagenicity classification based upon assessment of the corresponding logistic regression coefficients. These descriptors highlight the molecular structures that contribute to mutagenicity of PAHs within biodegradation pathways.

1. Introduction

Carla Ng is an assistant professor of Civil & Environmental Engineering. Her group focuses on understanding and predicting the biological impacts of chemicals in the environment.

Carla Ng, Ph.D.

Leanne Gilbertson, Ph.D.

Ingenium 2021

Dr. Gilbertson is an Assistant Professor in the Department of Civil and Environmental Engineering at the University of Pittsburgh. Her research group at the University of Pittsburgh is currently engaged in projects aimed at informing sustainable design of emerging materials and technologies proposed for use in areas at the nexus of the environment and public health.

Significance Statement

This study aims to assess the mutagenicity, and therefore environmental hazard, of PAHs in soil using a computational QSAR model. PAH mutagenicity is difficult to assess due to the variety of metabolites which can result from many possible biodegradation pathways. Descriptors discussed in this study will be the basis of the QSAR model.

Category: Computational Research Keywords: PAH, Mutagenicity, QSAR, Logistic regression

Polycyclic Aromatic Hydrocarbons (PAHs) contain at least two aromatic rings and often result as a byproduct of various natural and industrial processes, including forest fires, extraction and burning of fossil fuels, and plastic manufacturing. PAHs are commonly found in the atmosphere and soil of surrounding ecosystems due to these processes. However, PAHs in soil tend to be more persistent and are a possible source of carbon for degrading bacteria [1]. The Environmental Protection Agency (EPA) currently classifies 16 PAHs as priority pollutants [2]. There is increasing concern over the mutagenic properties of some PAHs in the human body. However, the degrading bacteria transform these parent PAHs into various metabolites via a large multitude of pathways, which may have different toxic properties from the parent PAHs, making it difficult to thoroughly assess the potential hazard in the laboratory setting. Quantitative structure-activity relationships (QSARs) have proven to be a useful tool for characterizing the toxicity of large chemical datasets, including those from PAH biodegradation, because of their predictive power [3]. QSARs can classify the endpoint toxic potential of input data based on a foundation of empirical training data and relevant structural and electronic descriptors. This foundation provides reproducible predictive ability for input data containing chemically similar compounds. The training dataset and chosen descriptors used to build a QSAR are crucial to its applicability [4]. There are currently a variety of powerful QSARs available to the public which can predict narcotic toxicity but lack the specificity to classify mutagenic metabolites of PAHs in soil. 81

One such example is the Ecosar module of the Estimation Programs Interface (EPI) Suite, provided by the U.S. EPA, designed to estimate the toxicity as LC50 for a variety of organisms [5]. This system relies heavily upon log KOW, the logarithm of the octanol-water coefficient, to classify the toxicity of molecules. However, mutagenicity of PAH metabolites is not linearly related to log KOW, so the accuracy of this QSAR is limited in predicting mutagenicity [6]. There are also QSARs available based on nitro-PAHs, however, PAH nitrogenation only occurs in the atmosphere, making these QSARs unsuitable for soil and water ecological toxicity assessment [7]. Due to the narrow applicability of current QSARs there is a gap in accurate data for PAHs related to the potential mechanisms of toxicity via biodegradation in soil [8]. The goal of this study is to identify the relevant structural and electronic descriptors based on available empirical training data to build a QSAR tailored to PAHs and their potential metabolites in soil and water environments. To do so, this study aims to identify the features of hydrocarbon PAHs which result in either mutagenic or non-mutagenic tendency. Overall, this QSAR aims to bridge the gap in current toxicity classifiers that lack the capability to accurately sort the metabolites that result from biodegradation.

2. Methods 2.1 Data Refining

The empirical training data set was composed of data from the TA98 Ames test strains, obtained from the Chemical Carcinogenesis Research Information System [9]. The Ames test is a common reverse mutation bioassay used to assess mutagenicity of a compound, a useful toxic endpoint due to its connection with genotoxicity and carcinogenicity [10]. The TA98 Ames test strain is commonly used for hydrocarbon mutagenicity assessment. Prior to utilizing the data to identify relevant structural descriptors, the empirical data set was refined to remove any outliers that did not represent PAH degradation well. PAH degradation in soil typically follows a pathway involving oxygenation then a ring opening reaction. In order to best reflect compounds likely to occur in a natural environment, the dataset was limited to a molecular weight of less than 500 amu and fewer than 5 rings. The Ames test often uses enzymatic activation to represent an organism’s internal metabolism, however for this study, only data that did not use enzymatic activation, known as direct acting mutagens, was used in order to best represent a natural environment.

water as a solvent using the conductor-like polarizable continuum model (CPCM) [12]. The functional M06-2X [13] called M06 and M06-2X. The M06 functional is parametrized including both transition metals and nonmetals, whereas the M06-2X functional is a high-nonlocality functional with double the amount of nonlocal exchange (2X was used with basis set 6-311G(d,p) [14]. M06-2X includes dispersion correction in its formula and performs well with water-based solvation calculations. [15]. Gibbs free energy, Highest Occupied Molecular Orbital (HOMO) energy, Lowest Unoccupied Molecular Orbital (LUMO) energy, ionization potential, and electron affinity for each molecule in the data set was obtained. Ionization potential and electron affinity were performed at STP in the gas phase to agree with NIST. Approximately 1495 topological descriptors were added via the PaDEL-Descriptor software [16]. Recursive feature elimination was used to select the most relevant descriptors. Variance inflation factor (<2.5) was then used to eliminate descriptors with close correlations, reducing the data set to the 20 most relevant descriptors. To then assess the accuracy of classification for the empirical data with these chosen descriptors, a receiver-operating characteristic (ROC) curve was generated using 10 iterations of 3-fold cross validation. On each iteration, two thirds of the data were used for training, and one third for testing. Each iteration the data was randomized to select different values for testing and training. Code used for this analysis is available on github: https://github.com/ngLabGroup/ ta98_mutagen_qsar.

3. Results

Using these descriptors, Figure 1 details the current Receiver Operating Characteristic (ROC) curve training assessment. With an area under the curve of 0.930, this suggests a high level of accurate classification of the testing data. The overall accuracy of this run was 0.876, precision was 0.659, sensitivity was 0.747, and the F1 was 0.700.

2.2 Descriptor Selection

The refined data set was then used to identify the relevant descriptors that may indicate a mutagen. Kohn-Sham density functional theory (DFT) calculations were performed in Gaussian 09 [11]. Neutral structures were optimized and confirmed to minima through the absence of imaginary frequencies. In order to simulate environmental conditions, calculations were performed at 20 °C with

82 Undergraduate Research at the Swanson School of Engineering

Figure 1: ROC curve representing 20 most relevant topological descriptors

Ingenium 2021

These logistic regression coefficients of the descriptors indicate the relationship between the descriptor and mutagen classification, as shown in Figure 2. The strongest relationships, either positive or negative, are expressed in the logistic regression coefficients for descriptors relating the number of rings larger than 12 members containing oxygen (nFG12HeteroRing), the average centered Broto-Moreau autocorrelation (AATSC6c), and the z-modified information content index (ZMIC4), which have values of 1.85, -1.41, and -1.04 respectively.

Figure 2: Coefficients of logistic regression for 20 most relevant topological descriptors and descriptions of the 3 strongest relationships for PAH mutagenicity classification

4. Discussion

Establishing this set of relevant descriptors to the training data set is the first step to building a QSAR able to assess novel data. These initial assessments of the empirical data have narrowed the descriptor data set to 20 descriptors from the original 1000. nFG12HeteroRing, AATSC6c, and ZMIC4 have the most significant effect on classification according to Figure 2. The strong positive regression coefficient for nFG12HeteroRing suggests that oxygen atoms within the ring are significant to the mechanism to reach mutagenic metabolites. The molecules with the highest nFG12HeteroRing count share epoxide rings as a common feature, agreeing with literature that

epoxide rings in specific structural positions may influence mutagenicity [17]. AATSC6c relates charge to mutagenicity by assessing similar charges 6 atoms away which is in relation to oxygen groups for this study. The negative regression coefficient suggests that atoms with a large distance between oxygen groups are less likely to be mutagens. Finally, ZMIC4 refers to the symmetry and branching substituents of the molecule. The negative regression coefficient suggests that the number of branching substituents inversely relates to the mutagenicity. Overall, the influence of these descriptors in classification suggests that oxygen groups are a critical component in the mechanisms which drive mutagenicity. The accuracy value from the ROC curve suggests accurate overall classification of mutagens vs. non-mutagens for this QSAR model. Additionally, the sensitivity value from the ROC curve indicates that a number of mutagens are incorrectly classified as non-mutagens in the training data set. The dataset is unbalanced with about five times as many non-mutagens as mutagens. A threshold for classifying mutagens vs non mutagens was manually chosen to maximize the F1 score and the precision value for this analysis. Precision is another crucial measure to understanding how many structures are correctly classified based on the total number of mutagenic classifications. Future work should select this threshold mathematically based on the dataset and desired optimal performance of specific classifier metrics. Ideally, the precision and specificity values are maximized to ensure every mutagen is correctly classified.

5. Conclusions

One goal with this QSAR is to create a transparent computational model with easily interpretable results. 20 descriptors are simply too many factors to interpret individual effect, however, this study significantly narrows down which descriptors are most influential on mutagenicity classification. Additional methods of statistical elimination are necessary to both validate and refine the current results. Such steps include determining which descriptors overlap significantly in the classification of data. These next steps will aid the QSAR in providing a transparent and reliable prediction method of PAH metabolite mutagenicity, efficiently reducing the amount of lab work needed to experimentally determine the hazardous properties.


6. Acknowledgements

Mentorship was under Trevor Sleight, P.E, Dr. Carla Ng, and Dr. Leanne Gilbertson. Funding provided by the Swanson School of Engineering Summer Undergraduate Research Internship (SURI) program and the Office of the Provost. Special thanks for Dr. Ioannis Bourmpakis for assistance with the Gaussian calculations.

7. References

[1] H. I. Abdel-Shafy and M. S. M. Mansour, “A review on polycyclic aromatic hydrocarbons: Source, environmental impact, effect on human health and remediation,” Egypt. J. Pet. 25 (2016) 107–23. [2] U. S. EPA, “Priority Pollutant List.” [3] A. K. Debnath et al. “Quantitative structureactivity relationship investigation of the role of hydrophobicity in regulating mutagenicity in the Ames test: 2. Mutagenicity of aromatic and heteroaromatic nitro compounds in Salmonella typhimurium TA100,” Environ. Mol. Mutagen. 19 (1992) 53–70. [4] M. T. D. Cronin et al. “Use of QSARs in international decision-making frameworks to predict ecologic effects and environmental fate of chemical substances,” Environ. Health Perspect. 111 (2003) 1376–90. [5] O. US EPA, “EPI SuiteTM -Estimation Program Interface,” US EPA, 2015. https://www.epa.gov/tsca-screening-tools/ecological-structure-activity-relationships-ecosar-predictive-model (accessed Jun. 22, 2020). [6] D. Ghosal et al. “Current State of Knowledge in Microbial Degradation of Polycyclic Aromatic Hydrocarbons (PAHs): A Review,” Front. Microbiol. 7 (2016). [7] A. K. Debnath et al. “Structure-Activity Relationship of Mutagenic Aromatic and Heteroaromatic Nitro Compounds. Correlation with Molecular Orbital Energies and Hydrophobicity,” J. Med. Chem. 34 (1991) 786–97. [8] P. Gramatica et al. “Quantitative structure–activity relationship modeling of polycyclic aromatic hydrocarbon mutagenicity by classification methods based on holistic theoretical molecular descriptors,” Ecotoxicol. Environ. Saf. 66 (2007) 353–61. [9] National Library of Medicine, “Chemical Carcinogenesis Research Information System,” https://www.nlm. nih.gov/databases/download/ccris.html. [10] B. N. Ames et al. “An improved bacterial test system for the detection and classification of mutagens and carcinogens,” Proc. Natl. Acad. Sci. U. S. A. 70 (1973) 782–86. [11] Frisch, M. J. et al. “Gaussian 09,” Wallingford CT: Gaussian, Inc. 2013. [12] V. Barone and M. Cossi, “Quantum Calculation of Molecular Energies and Energy Gradients in Solution by a Conductor Solvent Model,” J. Phys. Chem. A. 102 (1998) 1995–2001. [13] Y. Zhao and D. G. Truhlar, “The M06 suite of density functionals for main group thermochemistry, thermochemical kinetics, noncovalent interactions, excited states, and transition elements: two new functionals and systematic testing of four M06-class functionals and 12 other functionals,” Theor. Chem. Acc. 120 (2008) 215–41. 84 Undergraduate Research at the Swanson School of Engineering

[14] R. Krishnan et al. “Self-consistent molecular orbital methods. XX. A basis set for correlated wave functions,” J. Chem. Phys. 72 (1980) 650–54. [15] Z. L. Seeger and E. I. Izgorodina, “A Systematic Study of DFT Performance for Geometry Optimizations of Ionic Liquid Clusters,” J. Chem. Theory Comput. 16 (2020) 6735–53. [16] C. W. Yap, “PaDEL-descriptor: An open source software to calculate molecular descriptors and fingerprints,” J. Comput. Chem. 32 (2011) 1466–74. [17] A. W. Wood et al. “Inhibition of the mutagenicity of bay-region diol epoxides of polycyclic aromatic hydrocarbons by naturally occurring plant phenols: Exceptional activity of ellagic acid,” Proc. Natl. Acad. Sci. 79 (1982) 5513–17.

Ingenium 2021 † Innovations for reimagined educational experiences in the face of a pandemic

Sophie Shapiroa,d, Stuart McCutchenb,d, Zeve Cohenc,d, Elizabeth Gilmanb,d, William Clarkb,d Department of Bioengineering, bDepartment of Electrical Engineering, cDepartment of Mechanical Engineering, dInnovation, Product Design, and Entrepreneurship Program a

Sophie Shapiro

Stuart McCutchen

Sophie Shapiro is a bioengineering major and mechanical engineering minor at the Swanson School of Engineering. Her interests include the solid and fluid biomechanics of pathology and treatment, product design, and entrepreneurship. She intends to use her interdisciplinary training to deliver innovative regenerative and restorative therapies to cardiovascular medicine. Stuart McCutchen is a junior electrical engineering student. He is passionate about product design and development, particularly in the tech industry. His interests consist of digital electronics, IoT, signal processing, and machine learning. He currently plans to pursue his masters in electrical engineering.

Zeve Cohen is a senior Mechanical Engineering student from Philadelphia. Driven by his innate creativity, Zeve is passionate about innovative design. His involvement in Pitt’s Makerspace has help fuel his love for design and fabrication. Zeve plans to work at the intersection of mechanical engineering Zeve Cohen and industrial design with a focus on product ideation and rapid prototyping. His broad interests could lead him to numerous industries, but his primary goal is to create products that help others.

Elizabeth Gilman has always had a passion for engineering and working with others. In the future she hopes to design and build devices that can improve people’s lives and the world around them. Elizabeth Gilman

Professor William Clark joined the Mechanical Engineering and Material Science Department in 1992 and has an active research and teaching program in dynamic systems and controls. He participated in an SSOE faculty effort to create the Innovation, Product Design, and Entrepreneurship certificate proWilliam Clark, Ph.D. gram starting in 2014, which lead to the development of the Makerspaces in 2016 as well as subsequent student innovation opportunities. He currently directs the program, which invites students to participate in intensive design and innovation experiences in the summer. The paper presents the results of one of those teams.

Significance Statement

Preventing the spread of SARS-CoV-2 has resulted in research activity being significantly diminished at several institutions. To help mitigate exposure risks with in-person work, three individual product designs were developed with the cumulative goal of creating a safer environment as educational and research efforts adapt.

Category: Device Design

Key Words: SARS-CoV-2, engineering education, product design, prototyping †

Editors’ Choice



The onset of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic significantly disrupted the engineering research community, as academia reluctantly transitioned to an all-remote learning environment. With many research and development endeavors being time-sensitive, the University of Pittsburgh looks to slowly restart and ramp-up research operations while establishing effective safety measures. Working within Pitt’s plan to reopen facilities, we propose three isolated product designs that will jointly reduce the risk of exposure. By utilizing new advances in additive manufacturing and software development, our product designs are aimed at helping researchers mitigate risk as essential in-person work resumes. Following several design iterations, the projects resulted in the following, fully functional prototypes: RFID-enhanced contact tracing system, mass customization facepiece workflow, and an open-sourced LED Interactive wall. The current state of our projects has collectively worked to manage the immediate risk of SARS-CoV-2 in a scalable and compliant-driven manner, while providing a foundation for future opportunities for these designs to grow.

1. Introduction

The novel virus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), emerged out of Wuhan, China in December 2019. As of June 2020, there have been over 10 million reported cases of SARS-CoV-2, and its infectious nature paired with a lack of curative solutions has led to a worldwide pandemic [1]. Due to the apparent danger, 107 countries had implemented national school closures [2]. The University of Pittsburgh was amongst many other higher educational institutions that were obligated to shut down and switch to an all-remote learning environment for the safety of students and faculty. These operating constraints significantly disrupted the engineering research community. Laboratories, specifically ones handling living organisms, could not transform to a virtual platform as quickly or successfully as their academic counterparts; many finding it difficult to ramp down their time-sensitive research to reduced personnel and essential work. With campus activity initially restricted to essential work, the University of Pittsburgh’s Research Restart Working Group crafted policies and protocols intended to help boost research operations via implementing effective safety precautions [3]. Adjunctively, groups such as the Veser Laboratory and a joint effort by Pitt Engineering and Carnegie Mellon University helped to make 75% alcohol content hand-sanitizers and donate unused personal protective equipment (PPE) to essential workers, respectively [4]. As result of the combined response across Pitt’s and neighboring universities’ campuses, reopening has substantially increased with newly established safety measures. Working within Pitt’s plan to restart facilities, we propose three isolated product designs that will jointly contribute to the mitigation of exposure risks. The 86 Undergraduate Research at the Swanson School of Engineering

product designs help to improve the effectiveness of current, in-person safety measures. Our RFID-enhanced contact tracing system, mass customization facepiece workflow, and LED Interactive wall will collectively contribute to the ongoing effort to manage the risks of SARSCoV-2 in a scalable and compliant-driven manner, while providing a foundation for future iterations.

2. Methods 2.1 Contact tracing 2.1.1 Hardware design The urgency to contact trace more effectively in Pitt research facilities led to an early-stage device design equipped to track and store the following user-specific information: personal identification, date and time of entry/ exit, and location. Working to increase usability through transparency and ease, the project was created as an open-source, easy to manufacture and assemble device. The multi-part device was modeled in Fusion 360 (2020, Autodesk, San Rafeal, California) and designed for additive manufacturing. The device requires users to scan their Pitt identification (ID) card to which it can then retrieve the necessary information about the user. Utilizing infrared light (IR) sensors, the device differentiates between the entry and exit of users via the placement of the user’s ID card. To reinforce this, the device’s front panel is marked with an “in” and “out” section, corresponding to entry or exit. For user feedback, two Adafruit LEDs (New York City, New York, Cat # 302) and an Adafruit Piezo Buzzer (New York City, New York, Cat # 160) inside the device display green or red and sound a two-note rising or falling note interval, respectively, dependent on entrance or exit; the complete feedback loop takes no more than five seconds. 2.1.2 RFID and software development Within the actual device is an Adafruit Feather HUZZAH ESP8266 (New York City, New York) which processes data, controls outputs, and sends data over WIFI to a remote database. Its small build, WIFI capabilities, low price, and easy sourcing made it an effective and scalable solution. The remote database is a host on a shared network with the contact tracing devices via a Raspberry Pi 4b. The database is in MariaDB (2019, MariaDB Corporation Ab, Espoo, Finland), and the database interface is phpMyAdmin (2020, phpMyAdmin Project, Brownsville, Texas). The database records the date and time of which the data was received, what room it is from, the Pitt ID, and whether they were entering or exiting the room. Data older than 28 days automatically deletes, eliminating the need for manual database updates.

2.2 Mass customization facepiece workflow 2.2.1 Facemask design Following a user-discovery phase with medical personnel, the concept to create customized facemasks was pursued to ensure extended user comfort and a reliable seal. Within Fusion 360 software, a 3-D modeled

Ingenium 2021

half-mask respirator was designed to accurately conform to any facial structure. The mask was optimized for efficient airflow while minimizing required material to avoid a cumbersome facemask.

3. Results 3.1 RFID-enhanced contact tracing

2.2.2 Simulation-based deformities In the 3-D procedural generation software, Houdini (2019, Side Effects Software, Toronto, Ontario), the mask body’s seal was controllably deformed to various facial 3-D models using a built-in gravity simulator. This allowed the mask’s seal to conform to the test subject’s unique facial features and facial landmark locations. Mesh parameters of the modeled mask were revised multiple times until negligible breaches between the user-to-seal interface and symmetric positioning of the mask body was attained. Breaches were distances less than 0.5 millimeters between the seal and face, and positioning was with respect to the midsagittal plane of the face. 2.2.3 Manufacturing and post-processing Post deformation, the now customized mask body is fabricated using stereolithography (SLA) (Formlabs, Somerville, Massachusetts). 50A Elastic Resin (Formlabs, Somerville, Massachusetts, Cat # RS-F2-ELCL-01) was chosen for the first physical prototypes for its Shore durometer ranking. A universally-sized filter housing was attached to the mask’s universal interface via mechanical mechanisms.

2.3 Open-source LED Interactive wall 2.3.1 Software Development The LED Wall is a 1200 RGB LED display made for the basement of Benedum Hall at the University of Pittsburgh. It involved the use of Python, primary data processor, and Java, primary graphics, to transfer data from a camera to the LEDs. Python (Version 3.9.0) was preferred for image processing due to the OpenCV library. To transfer image processing data to Java, a WebSocket was used to convert Python to a data package recognizable by Java. A silhouette mirroring interaction was created using image differencing which drew a point wherever the camera detected a person’s location. The camera would start with an initial frame to identify the changes from the current frame and convert into coordinate points for the Java program. The program would then draw a shape based on the coordinates sent to it and generate an outline of a person. The software created preview graphics that reflected what the actual pixelization of the silhouette would be if the program was physically being ran on the wall. This makes software development easier for future designers to develop and test code remotely. 2.3.3 Electrical Design Requirements Two power supplies were needed to meet the power demand and maintain a reliable source of 5-volt power. The rewiring would allow each power supply to control half of the LEDs so providing more power to fewer LEDs. This decreased flickering and incorrect color changes in the LEDs. It also powered the peripheral devices such as the camera and Jetson-Nano (NVIDIA, Santa Clara, California).

Figure 1: Contact tracing housing. 3D-Printed contact tracing housing consisting of three pieces and intended to enclose all necessary hardware.

The contact tracing electronics enclosure is a threepiece, screw-shut housing. Via additive manufacturing, it can be fabricated in 23 hours using low-cost Fused Deposition Modeling 3-D printers (Creality, Shenzhen, China). The database is accessed via phpMyAdmin and is password-protected for security purposes. As more labs utilize this system, its value will grow exponentially as more data is collected and used to isolate and track exposure on campus.

3.2 Mass customization facepiece workflow

Figure 2: Customized facemask. Negligible breaches between the user-seal interface.

The semi-automated workflow allowed for customization for 6 users in one day. The utilization of SLA printing enabled rapid manufacturing times, giving a complete timeline for fabrication of approximately 9.5 hours. The filter housing includes an area for strap connections and 87

can accommodate a range of different filter materials. Further material selection and manufacturing methods are still being explored for both the mask body and filter housing, to increase mass customization.

3.3 Open-source LED Interactive wall

Figure 3: LED Interactive wall. Image differencing showing the sillhoute of someone in the camera’s view.

The LED Interactive wall is made up of 1200 LEDs. The camera and software involved enable graphics utilizing computer vision. The interaction demonstrated in Figure 3 shows the wall’s ability to mirror the silhouettes of passing pedestrians. Interactions such as these are the building blocks to developing even better computer vision programs, such as social-distancing and mask-wearing detection for students and faculty.

4. Discussion

The three functioning prototypes serve to promote the recommended safety precautions and catalyze the reopening of research facilities, all while providing a baseline for further design refinement and optimization. Both the contact tracing system and facepiece workflow significantly reduce the university’s dependency on third party compliance-related needs. Our contact tracing system provides an accessible and customizable alternative for universities to better streamline the tracking of SARS-CoV-2 amongst their campus. Currently, it is being tested at reopening research facilities at the University of Pittsburgh. The future entails reiterating on user feedback and scaling up the system to be successful in more labs and possibly, classrooms.

88 Undergraduate Research at the Swanson School of Engineering

The facepiece workflow works jointly with our contact tracing system to encourage compliance within the workplace. An increase in individuals wearing PPE has the power to significantly reduce and manage the risk of exposure as campus activity increases. Future direction entails further streamlining the mask body design and fully automating the workflow to increase facepiece output. Having the ability to output predictable face fittings of masks both, digitally and remotely, has been an exciting breakthrough for enhancing the user-to-seal interface and subsequentially, student and faculty health. In parallel with the mentioned endeavors, our LED Interactive wall reached the first of many breakthroughs in regard to helping manage the SARS-CoV-2 exposure risk. In achieving full body detection and tracking, it has paved a path for further expansion on this technology. With refinement and customization, various modules can be added to help drive compliance, such as social distancing and mask-wearing recognition programs. The LED Interactive wall not only serves as its own artform, but as a tool for reinforcing the precautions necessary for keeping in-person workers safe. Additionally, the wall was designed to be open-source and encourages further collaboration to achieve these next compliance-driven goals.

5. Conclusion

With little promise of an imminent eradication of SARS-CoV-2, it is imperative for institutions to adapt quickly. Student and faculty safety during this dangerous and uncertain era is of utmost importance to universities. Hence, in order to continue educating, researching, and fully operating, it is crucial to mitigate health risks. The three functional prototypes introduced in this paper present in-house tools that Pitt and other institutions can adopt to help manage those risks. The contact tracing system allows universities to localize the data that can help contain the spread of the virus on their campus. The facepiece workflow delivers a fast and personalized solution that addresses a fundamental challenge of increasing compliance with PPE. The LED Interactive Wall acts as an impartial advocator of compliance, accountability, and awareness. These are all scalable and customizable approaches that universities can take to safely and effectively reopen facilities in the face of a pandemic.

6. Acknowledgments

We thank all of our mentors for their outstanding support. We thank the Swanson School of Engineering for funding our endeavors.

Ingenium 2021

7. References

[1] Viner RM, Russell SJ, Croker H, et al. School closure and management practices during coronavirus outbreaks including COVID-19: a rapid systematic review. Lancet Child Adolesc Health. 2020;4(5):397-404. doi:10.1016/S2352-4642(20)30095-X [2] Wiersinga WJ, Rhodes A, Cheng AC, Peacock SJ, Prescott HC. Pathophysiology, Transmission, Diagnosis, and Treatment of Coronavirus Disease 2019 (COVID-19): A Review. JAMA. 2020;324(8):782-793. doi:10.1001/ jama.2020.12839 [3] Swanson School of Engineering. 2020. Managing—Not Avoiding—Risks. [online] Available at: <https:// www.engineering.pitt.edu/News/2020/Research-Restart/> [Accessed November 2020]. [4] University Times. 2020. How the University community is helping out its neighbors. [online] Available at: <https://www.utimes.pitt.edu/news/how-university-community> [Accessed December 2020].


User-specific factors associated with ladder handrail use Jami Stuckeya, Michelle Tsizhovkina, Kurt Beschornera, Erika Plinerb, Daina Sturnieksb, Stephen Lordb Human Movement and Balance Laboratory, Department of Bioengineering, bFalls, Balance, and Injury Research Centre, Neuroscience Research Australia a

Jami Stuckey

Jami Stuckey is an undergraduate biomedical engineering student at the University of Pittsburgh. Her research interests include biomechanics and rehabilitation engineering. After her time at the Human Movement and Balance Laboratories, she accepted a co-op program with the Human Engineering Research Laboratories. Kurt Beschorner is an Associate Professor of Bioengineering at the University of Pittsburgh. He utilizes techniques in tribology, biomechanics and ergonomics to generate understanding and interventions for falling accidents.

Kurt Beschorner, Ph.D.

Significance Statement

Understanding user-specific factors of ladder fall risk in older adults is important since ladder falls are a frequent cause of injury and fatality among this population. This study found greater risk-taking, as measured by a survey tool, was associated with less use of a safety handrail. Therefore, promoting less risk-taking behaviors may promote safer use of ladders.

Category: Experimental Research

Keywords: Ladder safety, fall risk, handrail use, user-specific factors, individual factors Abbreviations: Physiological Profile Assessment (PPA), Mini-Mental State Exam (MMSE)

90 Undergraduate Research at the Swanson School of Engineering


Unintentional injuries and fatalities caused by ladder falls are especially common among retirement-age adults living independently at home. Past research has focused on environmental factors that increase one’s fall risk, but there is a lack of knowledge on user-specific factors that promote safe ladder use. This study recruited older adult participants to determine the user-specific factors that influence an individual’s ladder handrail use during a domestic ladder task. Two trials of the ladder task were completed, one without a cognitive distraction (single task) and one while naming animals (dual task). Investigated user-specific factors comprised of physical ability, measured by the Physiological Profile Assessment (PPA); global cognition, measured by the Mini-Mental State Exam (MMSE); and risk-taking, measured by a self-assessment. Through Spearman’s ρ correlation analyses, this study found greater risk-taking to be associated with less use of the ladder safety handrail (ρsingle = -0.418, p < 0.0001; ρdual = -0.466, p < 0.0001). Focusing interventions to reduce propensity for risk-taking may promote safer use of ladders.

1. Introduction

Ladder falls are a frequent cause of unintentional injuries and fatalities; the estimated annual cost of ladder injuries in the U.S. is $24 billion [1,2]. Ladder injuries are particularly common among older adults at home (aged 65+ years) [1], but past ladder safety research has focused on falls of adults occurring at the workplace [3-5]. Similar investigation of ladder falls in older people working from home is emerging, yet gaps in our understanding remain. Furthermore, previous research has primarily studied the influence of environmental factors on ladder fall risk [6,7], and there is limited knowledge on user-specific factors that influence safe ladder use. Handrail use is a presumed safety strategy that can improve balance and enhance a person’s response to a destabilizing balance perturbation [8]. Not all individuals use the ladder handrails while on a stepladder, but there is a lack of knowledge characterizing handrail use. Relevant factors of general fall risk in older adults include physiological, cognitive, and psychological traits and abilities [9]. Thus, these user-specific factors may also be relevant to handrail use. This study quantifies the relationship of sensorimotor function (physiological factor), global cognition (cognitive factor), and risk taking (psychological factor) with handrail use of older adults when changing a lightbulb on a common household stepladder. The goal of this study is to improve current strategies for preventing unintentional ladder falls. We hypothesized that the user-specific factors would be associated with handrail use.

Ingenium 2021

2. Methods

2.2 Data Collection

2.1 Measures

104 older adult participants (Table 1) were recruited for this study. To be eligible for this study, participants were older than 65 years old, living independently at home or in a retirement village, and willing to change a light bulb while standing on the second step of a stepladder. Male (52)

Female (52)

Age (yrs)

65 – 89

65 – 87

Height (m)

1.57 – 1.93

1.51 – 1.82

Weight (kg)

55.3 – 113.2

46.5 – 93.3

Participants were asked to climb a stepladder, replace a lightbulb, and descend the ladder (Figure 1). Use of the handrail at the top of the ladder was optional and participants were not given any instruction regarding whether to use the handrails. The height of the light bulb was positioned to the participant’s hand height when standing on the second ladder step with 90° shoulder and elbow flexion.

Table 1: Participant Demographics

Exclusion criteria included the use of a mobility aid inside the home, a neurological disorder (e.g. Parkinson’s disease, multiple sclerosis, dementia/Alzheimer’s), weight over 120 kg, and the inability to change a light bulb on a ladder without pain. This study was approved by the Human Studies Ethics Committee at the University of New South Wales and informed consent was obtained prior to participation. Participants completed several clinical assessments. The Physiological Profile Assessment (PPA) [10], Mini-Mental State Exam (MMSE) [11], and an assessment of everyday risk-taking [12] were selected as a priori individual measures for this study to quantify the contribution of physical and cognitive ability and psychological state on safe ladder use. The PPA, measured by a summative Z-Score, captured measures of sensorimotor function (visual, reaction time, proprioception, strength, balance) that are predictive of fall risk [10]. This study used the PPA measure as a metric of physical ability, where a high Z-Score indicates reduced sensorimotor function. Higher scores on the MMSE indicate better global cognition [11]. Higher scores on the risk-taking assessment reflect a higher propensity to take risks in everyday activities (e.g. crossing the street after the Do Not Walk sign starts flashing) [12].

Figure 1: Ladder Setup. The lightbulb fixture was fixed to a wood and aluminum frame with an adjustable height. The stepladder had a tray and a handrail. The tray was used to hold the replacement lightbulb.


Two trials of this task for each participant were collected: one without a cognitive distraction (single task) and one while naming animals (dual task). The two different tasks were chosen to assess the influence of cognitive demand; however, the focus of this manuscript will be the userspecific factors. Participants practiced changing the lightbulb at ground level prior to the trials. These two trials were in a random order for each participant and participants were asked to complete the task “as quickly and safely as possible.” Videos were captured of participants completing the ladder tasks. These videos were analyzed to quantify handrail use. The duration of handrail use was recorded, called “Rail Time.” This was measured across two independent observers and averaged.

2.3 Statistical Analysis

Non-parametric correlation analyses (Spearman’s ρ) were performed because rail time was bimodal. Correlations were performed between the rail time and user-specific measures of physical ability (PPA Z-score), global cognition (MMSE score), and risk taking. Separate analyses were performed for the two tasks. (single task, dual task) using JMP Pro 14. An alpha value of 0.05 was used for statistical significance.

3. Results

The median rail time during the single task was 1.56 s (IQR: 0-7.58 s). The median rail time during dual task was 1.02 s (IQR: 0-12.89 s). Increased rail time use was weakly correlated with reduced physical ability, lower global cognition, and moderately associated with less risk-taking (Table 2 and Figure 2). Dependent Variable

Independent Variable

Rail Time


Single Task

Dual Task

Spearman’s ρ


Spearman’s ρ






Rail Time







Rail Time






Physical Profile Assessment measures physiological factor Mini-mental state exam score measures cognitive factor 3 Risk taking score measures psychological factor * significant p-value (< 0.05) 1 2

Table 2: Spearman’s ρ correlation coefficients for each individual measure by task. Figure 2 (right): User-Specific Factors vs. Rail Time. Scatter plot of Single Task and Dual Task rail use time with linear trend lines for a) Physiological Factor (PPA z-score) b) Cognitive Factor (MMSE score), and c) Psychological Factor (risk-taking score)

92 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

4. Discussion

The results of the correlation analyses support the hypothesis that user-specific factors correlate with handrail use. Reduced physical ability, lower cognition, and decreased risk-taking are associated with increased rail time. Individuals may use the handrail to compensate for their reduced physical and cognitive abilities. Furthermore, reduced physical and cognitive function is predictive of a greater fall risk in older adults [13]. Therefore, handrail use may reduce the fall risk of these individuals during ladder use. Individuals with higher global cognition and better physical ability showed reduced handrail use. These individuals may not require the supplementary assistance to safely complete the task. This is supported by another study in which older adults’ confidence in their stair climbing stability was assessed found that individuals with lower confidence used the handrail to a greater extent and positioned themselves closer to the rail [14]. Additionally, self-identified risk-takers are less likely to use the handrail. This is supported by other studies in which risk-taking / riskier behavior was associated with stair falls [14,15]. Possible limitations of this study include that one specific factor was tested and generalized for each broader factor (physiological, cognitive, and psychological) when perhaps the PPA, MMSE, and the risk-taking assessment may not be representative of an individual’s entire physical ability, mental capacity, or psychological state. Additionally, only task performance of individuals changing a light bulb on a household stepladder was tested. Future research is required to determine individuals measures that influence ladder task performance across different tasks and ladder designs.

5. Conclusions

This research found that rail grasping time is correlated with user-specific factors, including physical ability (ρsingle = 0.235, p < 0.024; ρdual = 0.179, p < 0.095) and global cognition (ρsingle = -0.209, p < 0.038; ρdual = -0.229, p < 0.042). Rail grasping time was also strongly associated with risk-taking (ρsingle = -0.418, p < 0.0001; ρdual = -0.466, p < 0.0001). This suggests that user-specific factors influence handrail use, especially psychological factors. Thus, focusing on the ladder task and reducing risk taking propensity may promote safer use of ladders.

6. Acknowledgements

This research was funded by the Whitaker International Program and the National Institute for Occupational Safety and Health (R01OH011799).

7. References

[1] Faergemann C., Larsen L.B., Non-occupational ladder and scaffold fall injuries, Accident Analysis & Prevention. 32 (2000) 475-750. [2] US Consumer Product Safety Commission, Injury Cost Model. (2014). [3] Axelsson P.O., Carter N., Measures to prevent portable ladder accidents in the construction industry, Ergonomics. 38 (1995) 250-259. [4] López M.A., Ritzel D.O., González I., Alcántara O.J., Occupational accidents with ladders in Spain: Risk factors, Safety Research. 42 (2011) 391-398. [5] Socias C.M., Menéndez C.K., Collins J.W., Simeonov P., Occupational Ladder Fall Injuries — United States, 2011, Morbidity and Mortality Weekly Report. 63 (2014) 341-346. [6] Chang W.R., Chang C.C., Matz S., Available friction of ladder shoes and slip potential for climbing on a straight ladder, Ergonomics. 48 (2005) 1169-1182. [7] Chang W.R., Huang Y.H., Chang C.C., Brunette C., Fallentin N., Straight ladder inclined angle in a field environment: the relationship among actual angle, method of set-up and knowledge, Ergonomics. 59 (2016) 1100-1108. [8] Maki B.E., Perry S.D., Mcllroy W.E., Efficacy of handrails in preventing stairway falls: a new experimental approach, Safety Science. 28 (1998) 189-206. [9] Pliner E.M., Doctoral Dissertation, University of Pittsburgh. (2020). [10] Lord S.R., Menz H.B., Tiedemann A., A Physiological Profile Approach to Falls Risk Assessment and Prevention, Physical Therapy. 83 (2003) 237-252. [11] Tombaugh T.N., McIntyre N.J., The Mini‐Mental State Examination: A Comprehensive Review, American Geriatrics Society. 40 (1992) 922-935. [12] Butler A.A., Lord S.R., Taylor J.L., Fitzpatrick R.C., Ability versus hazard: risk-taking and falls in older people, Gerontology. 70 (2015) 628-634. [13] Pieruccini-Faria F., Sarquis-Adamson Y., Montero-Odasso M., Mild Cognitive Impairment Affects Obstacle Negotiation in Older Adults: Results from “Gait and Brain Study”, Geronology. 65 (2019) 164-173. [14] Hamel K.A., Cavanaugh P.R., Stair Performance in People Aged 75 and Older, American Geriatrics Society.52 (2004) 563-567. [15] Jacobs J.V., A review of stairway falls and stair negotiation: Lessons learned and future needs to reduce injury, Gait & Posture, Vol 49 (2016) 159-167.


Relationship between center of pressure and contact region during gait: implications for shoe wear and slipping risk Joseph Sukinik, Rosh Bharthi, Sarah Hemler, Kurt Beschorner Human Movement and Balance Laboratory, Department of Bioengineering Joseph Sukinik is a senior bioengineering student originally from North Bethesda, Maryland. He is interested in biomechanics and medical device development. He hopes to attend medical school in the future. Joseph Sukinik

Kurt Beschorner is an Associate Professor of Bioengineering at the University of Pittsburgh. He utilizes techniques in tribology, biomechanics and ergonomics to generate understanding and interventions for falling accidents. Kurt Beschorner, Ph.D.

Significance Statement

The shoe contact region is relevant to its friction and the shoe’s response to wear. This study aimed to determine if ground reaction forces could predict this region.

Category: Experimental Research

Keywords: COF, contact region, slip and fall accidents

94 Undergraduate Research at the Swanson School of Engineering


One of the leading sources of occupational injuries, and injuries among the elderly, are slip and fall accidents. These accidents are initiated by a low coefficient of friction (COF) between the shoes and floor. Flooring and footwear products that offer good friction can improve safety and yield a competitive advantage for companies. As shoes become worn, the tread blocks degrade, reducing their friction capability and increasing slip risk. However, wear is uneven, and some parts remain completely intact. This study focuses on generating a method to guide shoe tread design and reduce slip risk through prediction of the contact region with ground reaction forces. Using force plates and a frustrated total internal reflection (FTIR) device embedded in the floor, we generated ground reaction force data and shoe contact images to measure center of pressure (COP) and the contact region centroid, respectively. The contact region (from FTIR) and COP (from the force plate) was quantified during the moment of maximum friction requirements. Correlation analysis was performed. Mediolateral COP was shown to have a strong correlation with the mediolateral centroid (r = 0.933, p < 0.001), while the anteroposterior COP showed a weak correlation to the anteroposterior centroid (r = 0.397, p = 0.115). This study provides rationale that the COP can be used to predict the contact region in the mediolateral direction. Results from this study can be used to expand this analysis to consider additional types of shoes and generalize this method of determining wear location on any footwear.

1. Introduction

Slip and fall accidents are a serious problem that plagues both occupational environments and the lives of the elderly. In general, falls (which are commonly caused by slipping) are defined as “an event which results in a person coming to rest unintentionally on the ground or other lower level, not as a result of a major intrinsic event (such as stroke) or overwhelming hazard” [1]. In 2012 alone, the Liberty Mutual Workplace Safety Index showed that the cost of disabling injuries due to falls in the workplace is around $9.19 billion and accounts for approximately 15% of total injury costs in the US [2]. While falls typically yield worse outcomes for elderly patients, falls occur at similar rates regardless of age [3, 4, 5]. One of the leading causes of falls is slip, which can occur as treads begin to wear on the shoes. According to Courtney et al. approximately 40-50% of all fall-related injuries involved slipperiness [6]. Additionally, Berg et al. found that among the elderly, approximately 59% of falls were due to slipping, and approximately 5% resulted in fractures, while an additional 9% caused significant soft tissue damage [7]. Lastly, previous work has shown good linearity between predicted contact pressures from contact mechanics and measured contact pressure from FTIR [8]. Thus, previous research studies show the potential of this technique to measure mechanics and demonstrate the need for increased research into tread design.

Ingenium 2021

Slipping risk is often the result of a reduced coefficient of friction (COF) between the shoe tread and the floor. This risk can be influenced by the condition of the floors (e.g., when wet after being mopped) or the shoes (e.g., worn tread on shoe outsoles). The contact region of shoe tread influences its friction performance [9, 10]. As the shoe tread becomes worn, their friction decreases, and slip risk increases [11]. However, understanding of shoe tread wear and its interaction with the individual is emerging. As such, footwear products that offer good and durable friction performance can improve safety and yield a competitive advantage for companies. To properly develop footwear that can provide good traction, the relationship between the biomechanics of gait and wear region must be determined. The goal of this project is to quantify whether the location of shoe tread contact can be predicted based on a person’s center of pressure during stance. This will be accomplished by attempting to correlate COP to the contact region at the point believed to contribute most to wear, the point where the required coefficient of friction (RCOF) occurs.

2. Materials and Methods

2.2 Data Processing

Data processing, image analysis, and statistical analysis was performed using custom code (MATLAB Version 2019a, Mathworks, Natwick, MA). Anteroposterior (AP) and mediolateral (ML) COP points were extracted from the force plate data at the time of RCOF since this is the time of the gait cycle believed to be most relevant to slipping [12] Both COP and centroid location were identified based on their relative location to the most posterior heel marker. Data was processed for the right and left shoes individually to avoid creating any confounding data with regards to differences in gait between the right and left legs. RCOF can be defined as the maximum friction point that halts forward motion and was calculated as the peak ratio of the vector sum of the shear forces to the normal force when other criteria were also met [11]. Only the first 200 ms after heel contact were considered in identifying the time of RCOF. Furthermore, only braking force that halts forward motion was considered [13], and the normal force needed to be greater than 100 N (Figure 2). Heel contact was defined as the point where the normal force first exceeded 10 N.

2.1 Data Acquisition

Seven participants (6 male, 1 female) were recruited to participate in the study, approved by the Institutional Review Board Protocol (University of Pittsburgh #19050285). Participants were instructed to wear a pair of slip resistant shoes (SR Max, Greensboro, NC) for an initial gait assessment and then in their workplace for one month at a time. Inclusion criteria included regularly wearing treaded shoes in the workplace, working on their feet for at least four hours in a typical day, having a BMI below 35, and no recent history of musculoskeletal injury in the last two years. For this study, an additional criterion for data collection is that participants have walked at least 100 km, measured via a pedometer, and guarantees adequate wear for visual comparison of contact region to wear region for an adjacent study. Participants provided informed consent prior to study participation. Participant demographics are stated in Table 1. Males



45.8 ± 12.3 years

26 ± 0.0 years


185.3 ± 10.0 cm

175 ± 0.0 cm


100.9 ± 8.3 kg

76.8 ± 0.0 kg

Shoe Size

12.4 ± 1.7 Men’s US

7.5 ± 0.0 Men’s US

Figure 1: Procession of transforming an FTIR image into an analyzable image. (A) is the original image, (B) is the grayscale image isolated contact region, (C) is black and white converted image, and (D) is the final image with contact region centroid (red) marked relative to the posterior heel marker (blue). This demonstrates the method used to isolate contact region centroids.

Table 1: Demographic data for participants involved in the study.

Each participant was instructed to walk across a walkway that was instrumented with a force plate to collect COP in the x- and y-directions and three-dimensional force data. They each completed five trials. They were also instructed to complete five trials of walking over a clear plexiglass plate that collected Frustrated Total Internal Reflection (FTIR) images to capture the contact region of the shoe tread (Figure 1). 95

2.3 Image Processing

The individual FTIR video frame at the time of RCOF was isolated, converted to grayscale, and the pixels were dichotomized into contact/non-contact states based on the brightness of the pixel. These pixels were selected to include the range of brightness in the illuminated tread blocks while attempting to exclude any pixels that might be associated with the non-contact region or any scratches on the plexiglass [14]. The contact region centroid was determined, and its center was calculated relative to the heel marker.

Descriptive statistics are presented as mean (standard deviation) and data maxima and minima, as depicted in Table 2. The ML locations between COP and center of contact were well correlated (r = 0.933, t15 = 10.1, p < 0.001) (Figure 3A). The AP correlation of the contact region centroid was not well correlated with the AP COP location at RCOF (r = 0.397, t15 = 1.7, p = 0.115) (Figure 3B), where the t subscript represents the degrees of freedom. Here, since we would expect the centroid values and COP values to be equivalent, as they both reference the same point (the heel marker), a slope close to one would indicate greater accuracy. The ML slope (0.796) is much closer to one than the AP slope (0.467). Thus, the AP COP is a poor predictor of the AP contact centroid while the ML COP is a much better predictor of the ML contact centroid.

Figure 2: Plot depicts the ratio of friction to normal forces during 200ms following heel contact. RCOF is indicated by the red ‘X’ at local maximum and used for data processing. This forms the basis for which time points should be isolated and indicates other potential options, such as the peak around 50 ms.

2.4 Statistical Analysis

The relationship between the contact region center location and the COP was assessed using Pearson’s bivariate correlation analyses. One linear regression was performed comparing the ML location of the COP and the center of contact and another using the AP location. Centroids and COP values were averaged across trials for each participant-shoe. Only trials where the shoe was near the center of the image were included to minimize edge-distortion effects, meaning that 2-5 FTIR trials were averaged for each foot of each participant. An alpha value of 0.05 was used for the correlation.

3. Results Data Type







Mean (Standard Deviation)




86.4 (10.7) mm

69.2 mm

109.3 mm


83.8 (16.6) mm

60.6 mm

115.1 mm


15.7 (8.55) mm

0.3 mm

32.2 mm


-14.8 (8.24) mm

-26.9 mm

1.2 mm


71.7 (16.7) mm

52.6 mm

93.74 mm


67.1 (15.8) mm

47.0 mm

93.9 mm


18.5 (5.5) mm

9.8 mm

27.2 mm


-11.83 (8.2) mm

-25.3 mm

-1.9 mm

Table 2: Data collected from both FTIR and COP collection techniques prior to bivariate correlation.

96 Undergraduate Research at the Swanson School of Engineering

Figure 3: COP vs Contact Region Centroid at the time of RCOF for the ML position (A) and the AP position (B). R-values are included in the bottom right corner of each plot to quantify fit.

Ingenium 2021

4. Discussion

The results show that the ML contact region is better captured by the COP position than the AP component. This indicates that the COP in the ML direction may be more useful for predicting the contact location at the time of RCOF than its position in the AP direction. Most participants had a lateral COP and contact indicating that their feet were inverted at the time of RCOF. A few limitations should be stated as to why the center of the FTIR images may be inconsistent with the center of pressure. First, due to the nature of the RCOF and the size of the camera lens, often the foremost portion of the tread was cropped out of the image. Thus, this region did not factor into the center of contact. Also, the center of contact from the FTIR images were not weighted by pressure on each tread. Force is differentially distributed across shoe tread, which contributes to center of pressure but not our calculation of center of area [15]. The contact region is defined as the portion of the shoe tread that is touching the ground at the specified time point. As such, we can assume that the contact region is also the same region that will eventually begin to wear away. This makes contact area a necessary component for predicting the wear region. As we could not find any previous literature to compare contact region during gait with center of pressure or contact region versus wear area, this study will form the basis for future work in predicting shoe tread wear locations and any improvements will be based off the limitations of this work. This study provides the rationale for expanding the use of FTIR to capture medial-lateral COP or to use COP to capture the medial-lateral center of the contact region. We aim to expand this analysis to consider additional types of slip-resistant and non-slip-resistant shoes to establish the generalizability of the results. We also intend to expand this method to consider other time points that may be relevant to slipping or the occurrence of wear.

5. Conclusion

Overall, the AP and ML COP locations were compared to the contact region centroid. While the ML COP locations were well correlated with the contact centroid (r = 0.933, p < 0.001), a similar relationship was not observed in the AP direction (r = 0.397, p = 0.115). A more comprehensive study is required in which we will assess numerous different time points for their ability to predict wear location as well as find a good time point that minimizes the risk for lens distortion in the AP direction. Based on these results, we believe that the COP can be used to estimate the ML contact region and presumably the ML location of tread wear. This may enable the use of COP in estimating the critical contact region of the shoe. In future work, we hope to accurately predict contact region and wear region over time. Wear prediction for footwear remains an emerging area and predictions from this paper may assist in when center of pressure may offer value in predicting contact region and subsequent wear locations.

6. Acknowledgements

We would like to thank the Swanson School of Engineering, the Office of the Provost at the University of Pittsburgh, and the National Institute for Occupational Safety and Health (R01OH010940) for providing the funding for this project.

7. References

[1] Tinetti ME, et al., Risk factors for falls among elderly persons living in the community. N Engl J Med. 1988 Dec 29; 319(26):1701-7. [2] Chang, Wen-Ruey et al. State of science: occupational slips, trips and falls on the same level. Ergonomics. 2016 vol. 59,7: 861-83. [3] Rubenstein LZ, et al. The epidemiology of falls and syncope. Clin Geriatr Med. 2002 May; 18(2):141-58. [4] Dionyssiotis Y, et al. Analyzing the problem of falls among older people. Int J Gen Med. 2012;5:805-813 [5] Lockhart TE, et al. Effects of aging on the biomechanics of slips and falls. Hum Factors. 2005;47(4):708729. [6] Courtney, T. K., et al. “Occupational slip, trip, and fall-related injuries can the contribution of slipperiness be isolated?.” Ergonomics 44, no. 13 (2001): 1118-1137. [7] Berg, W. P., et al. “Circumstances and consequences of falls in independent community-dwelling older adults.” Age and ageing 26, no. 4 (1997): 261-268. [8] Sharp, J.S., et al., 2018. Optical measurement of contact forces using frustrated total internal reflection. Physical Review Applied, 10(3), p.034051. Appl. Ergon., 2001, 32, pp. 127-134 [9] Moghaddam, S. R. M., et al. (2018). Predictive multiscale computational model of shoe-floor coefficient of friction. Journal of biomechanics, 66, 145-152. [10] Iraqi, A., et al., 2020. Prediction of coefficient of friction based on footwear outsole features. Applied Ergonomics, 82, p.102963. [11] Sundaram, et al., Journal of Biomechanics, Vol. 105, pp. 109797. [12] Iraqi, A., et al., 2018. Coefficient of friction testing parameters influence the prediction of human slips. Applied ergonomics, 70, pp.118-126. [13] W. Chang, et al. The Effect of Transverse Shear Force on the Required Coefficient of Friction for Level Walking. Human Factors. Vol 53, No. 5, pp.461-473. 2011. [14] R. Bharthi, et al. Abstract: Correlation Between the Locations of Shoe Wear Region and Center of the Contact Region at Peak Friction. BMES National Conference. 2020. [15] Yamaguchi, T. Distribution of the local required coefficient of friction in the shoe–floor contact area during straight walking: A pilot study. Biotribology, 19, 100101. 2019.


Analysis of stride segmentation methods to identify heel strike Victoria Turchicka,b, Yulia Yatsenkoa,b, and Mark Redferna,b a b

Human Movement and Balance Laboratory, Department of Bioengineering Victoria Turchick was born and raised in Pittsburgh. Her research interests are focused on the biomechanics of human movement, and she hopes to pursue a career in the rehabilitation device industry.

Victoria Turchick

Dr. Redfern is the William Kepler Whiteford Professor in the Department of Bioengineering., with secondary appointments in the Departments of Physical Therapy, Otolaryngology, and Health and Rehabilitation Sciences. Dr. Redfern is a graduate of the University of Michigan where he earned Mark Redfern, Ph.D. his bachelor of science in engineering science and applied mechanics, and Ph.D. in bioengineering. He was a postdoctoral fellow at the University of Michigan Center for Ergonomics. Dr. Redfern’s research interests include two broad areas: the biomechanics of human movement and occupational biomechanics.

Significance statement

Three methods of stride segmentation from IMU data were compared for their ability to identify heel strike. All methods were able to accurately segment stride in all but one subject, while only two accurately identified heel strike.

Category: Methods

Keywords: inertial measurement units, stride segmentation, gait biomechanics

98 Undergraduate Research at the Swanson School of Engineering


The purpose of this study was to compare two methods of stride segmentation to a third ‘gold standard’ method for their ability to identify heel strike from inertial measurement units (IMU). Specifically, the acceleration data from IMUs placed on the shins and the pelvis of participants were used. These data were previously collected by the University of Rome. Methods 1 and 2 used the accelerometer signals from an IMU located on the low back (which is often used in evaluating gait with IMUs), while Method 3, the ‘gold standard’, used accelerometers on the right and left shins. On average, Method 1 identified heel strike as occurring 27 milliseconds prior to heel strike identified by Method 3, and Method 2 identified heel strike 160 milliseconds prior to Method 3. Individual differences in identifying heel strike were found and the effect of performing turns compared to straight walking were also analyzed. The findings of this study show that all methods were able to accurately segment stride, but only Methods 1 and 3 were able to correctly identify heel strike in all but one subject, making them better suited for gait analysis. None of the methods were able to accurately segment stride while subjects performed turns. Future studies should explore more robust algorithms for stride segmentation to perform accurate assessments of gait from IMU data.

1. Introduction

IMU data are becoming an increasingly popular way to measure kinematics, joint angles, and analyze gait. As an alternative to traditional motion sensor systems, IMUs can operate independent of a larger in-lab system. Containing accelerometers and gyroscopes along three axes and a magnetometer, IMUs can provide three-dimensional data outside of a traditional lab environment. This allows motion data to be captured in a more natural setting. The accelerometers collect linear accelerations values in 3 orthogonal directions, the gyroscope collects angular velocity data about those same axes, and the magnetometer allows for proper orientation of collected data within Earth’s magnetic field. This study was conducted as part of a larger project to investigate ways to process IMU data that is collected during walking. A stride, or two steps, can be identified using acceleration data from these IMUs. Stride segmentation is needed to provide insight into gait symmetry, cadence, force attenuation, and individual differences in stride. By analyzing multiple steps, several strides can be analyzed and the phases within each stride can be identified. This includes heel strike, heel strike transient, toe-off, stance phase, and swing phase. The purpose of this study was to determine the differences among three methods of stride segmentation that identify the time that heel contact occurs from IMU acceleration data. This study was motivated by the observation that most current literature using IMUs to characterize gait use only one sensor, placed on the low back. Therefore, it is vital that algorithms using accelerometer data from this sensor can accurately segment stride and identify gait

Ingenium 2021

events. In this study, two methods using accelerometer data from the low back (Methods 1 and 2) were compared to the ‘gold standard’, Method 3, which uses accelerometer data from the shins. These methods were compared by calculating the differences in the time they each identified heel strike. Methods 1 and 2 were based upon algorithms developed by Sejdic, et al. [1] and Zijlstra [2], respectively.

2. Methods 2.1 Experimental setup and data collection procedures

The data were collected previously at the University of Rome “Foro Italico’s” Centre of Bioengineering of the Human Neuromusculoskeletal System [3]. Subjects were instructed to walk normally for six minutes back and forth along a walkway, wearing IMUs containing a gyroscope, a magnetometer, and an accelerometer. The subjects walked to the end of the walkway, turned around and walked to the original starting point and turned around and so on. The IMUs were placed on the head, sternum, pelvis, and the left and right shins and attached via elastic straps, adjusted to securely fit the attachment point. Data from 9 healthy subjects (age: 27 ± 4 years; stature: 1.72 ± 0.07m; body mass: 66 ± 8kg) were used in this analysis. The IMU data were made available to the authors in MATLAB format and processed using the three segmentation methods described in section 2.2.1.

2.2 Data analysis 2.2.1 Description of methods Method 1 identifies heel strike using the antero-posterior (AP) and vertical (VT) acceleration data from the IMU placed on the pelvis of the subject. The following is a summary of the methodology used to develop the algorithm for Method 1 (see Sejdic, et al. [1] for a more detailed description). To remove artifacts unrelated to walking, the data were processed before segmentation by removing the mean and low-pass filtering. In the first stage of the algorithm, local maximum values in the VT direction that are at least 0.35 seconds apart are identified as possible events of interest relating to toe-off. The second stage is focused on exactly identifying toe-offs as the local minima surrounding the regions previously identified in stage one. If two possible toe-off locations are identified within a few milliseconds of each other, one is omitted through this process. Finally, stage three identifies heel strike using the derivative of the acceleration data identified as an area of interest. Mediolateral (ML) data is used to determine right versus left toe-offs and heel strikes in stages two and three. These three stages result in the identification of times of right and left toe-offs and heel strikes.

Method 2, developed by Zijlstra [2], uses the pattern of AP acceleration signals to identify heel strike. Based on an inverted pendulum model, the forward acceleration of the trunk changes sign from positive to negative directly after contra-lateral foot contact. A previous study conducted by Zijlstra indicated that heel strike occurs at the peak forward acceleration that occurs just before this sign change. Therefore, the acceleration peak that occurs just before a change of sign (positive to negative) was taken as left or right heel strike. This code also calculates stride duration and walking speed. Method 3 identifies heel strike using the AP acceleration data from the IMUs placed on the right and left shins. This method identifies local maxima, within about a second of data (dependent on walking cadence). Heel strike is identified as the rapid accelerations that occur at heel contact, which are more apparent at the shins than they are at any other sensor.

2.3 Data processing

The same method of data processing was applied to each of the nine subjects. The acceleration data were first cropped to exclude quiet standing at the start and end of data collection. The data then included acceleration time series collected from 50 seconds to 240 seconds of the original time trial. This section of acceleration data was used in the code for all of the methods, although Methods 1 and 2 used this data from the pelvis IMU, while Method 3 used this data from the IMUs on the right and left shins. After the time of each heel strike was identified by Methods 1, 2, and 3, each data point for that time was identified as part of a walking segment or part a turning segment of data by manually identifying consistent vs inconsistent segments of data via visual inspection. The walking segments were also labeled by number. For example, a subject starts walking during ‘Walking Segment 1’, then turns around and begins ‘Walking Segment 2’. These were identified to determine if there were any differences in heel strike identification among the different walking segments, or if there were differences in heel strike identification between walking segments and turning segments. The differences between the methods were determined by subtracting the times identified as heel strikes by Method 1 and Method 2 from these same points identified by Method 3. This method of comparison was chosen because Method 3 uses acceleration data from the shins, where heel strike is readily apparent and easy to identify. To test for the significance of walking segments versus turning segments on heel strike identification, this time difference was calculated for the entire time trial of one subject (Subject 1) (50 seconds to 240 seconds of the original trial). To examine any individual differences among subjects, the difference between Methods 1 and 3, and Methods 2 and 3, the first 40 points of Walking Segment 2 were analyzed for each of the nine subjects. ANOVA with a post-hoc Tukey HSD were used for statistical comparisons.


3. Results 3.1 Walking segments 3.1.1 Segmentation times Results of ANOVA with a post-hoc Tukey HSD revealed that Subject 4 was significantly different from all other subjects at a 95% confidence level. Table 1 summarizes the average difference in heel strike identification by subject between Methods 1 and 3 and Methods 2 and 3 for the first 40 steps of Walking Segment 2. With the exclusion of Subject 4, the average difference in heel strike identification between Methods 1 and 3 was 27 ms. As opposed to Method 1, Method 2 displays a greater time difference to Method 3. Excluding Subject 4, the average time difference is 160 ms in the identification of heel strikes between Methods 2 and 3. Subject number

Method 1-3 Method 2-3










All Subjects Avg. ± s.d.

All Subjects (w/o Sub.4) Avg. ± s.d.

35 ± 28

27 ± 8

253 ± 282

160 ± 39





































3.1.2 Stride averaging Identifying an accurate method of stride segmentation aids in visualizing characteristics of stride from large data sets. This also allows for more widely applicable and accurate stride labeling. For example, a graph of ensemble averages from each point of Subject 1’s walking acceleration data at the pelvis is seen in Figure 1. The first heel strike transient of the stride occurs at 0%. Forefoot loading occurs at about 6%, toe-off for step one at around 29%, and swing phase occurs between 29% and 49%, where heel strike occurs, followed by heel strike transient for the second step at 50% of the stride. Similarly, for the second step, forefoot loading occurs at 56%, toe-off at 80%, heel strike at 98%, and the third heel strike transient at 100%, to signify the end of one stride, or two steps.

Table 1: Time differences (ms) (mean±s.d.) in heel strike identification subtracting Methods 1 and 2 from Method 3 Figure 1. Ensemble averages of vertical pelvis acceleration across all strides for Subject 1 are normalized by percent of the gait cycle. Shading represents the standard deviation.

3.2 Walking segments vs. turning segments comparison in one subject

During the trials, subjects walked in a straight line and then turned around to continue walking multiple times. When a subject turned around, the acceleration waveform was visually distinct from straight walking. The AP acceleration from turning segments are inconsistent, and the amplitudes of the accelerations at heel strikes are smaller. All the turn and straight walking segments were identified and analyzed separately.

100 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Excluding turning segments of data, the average time delay in identification of heel strike from Method 1 to Method 3 was 25 ms for the walking segments of Subject 1. The average time delay in identification of heel strike from Method 2 to Method 3 was 135 ms. The delays between methods had much greater variability during turning segments of data. These results, and the raw data averages including turning segment data, are summarized in Table 2. It is clear that a consistent difference in heel strike is not accurately, nor consistently identified during the turning segments of the acceleration data. In addition, ANOVA with a post-hoc Tukey HSD was completed to confirm these results and was simultaneously used to determine if walking segment number produced a significant effect on the difference in heel strike identification between methods. Results showed a significant difference between at least one of the time segments, with a post-hoc revealing that the only significantly different time segments are the turns. There was no significant difference between any of the walking segments. Segment

Method 1-3

Method 2-3

Walking segments only



Turning segments only



All segments



Table 2: Time differences (ms) (mean±s.d.) between methods in heel strike identification for walking and turning segments

4. Discussion

There is a clear difference in the identification of heel strike between Methods 1 and 3. Because Method 3 identifies a very consistent and obvious spike in shin acceleration data, we consider Method 3 as an accurate representation of heel strike (i.e. the ‘gold standard’). The time difference in heel strike identification between Methods 1 and 3 is consistent across subjects. Method 1 identifies true heel strike (initial heel contact), while Method 3 identifies the rapid spike in vertical acceleration just after heel contact, known as heel strike transient [4], accounting for the small delay in heel strike identification between the two methods. The difference in heel strike identification between Methods 2 and 3 is much larger than the difference between Methods 1 and 3. A small standard deviation shows that Method 2 consistently segments stride, however, it does not accurately identify heel strike. Method 2 segments stride beginning at a point just after toe-off, during swing phase. Note that the heel strike identifications for Subject 4 were significantly different compared to all the other subjects (see Table 1). Neither Method 1 nor 2 was able to correctly identify heel contact. The reason for this inability to determine heel strike is not known. However, Subject 4’s acceleration data was less variable in magnitude than any of the other subjects. All three methods rely on distinguishing local extrema of the acceleration data to distinguish

heel strike from other gait events. This may have caused the algorithms to be less effective on Subject 4’s data. Thus, when segmenting with an algorithm, a method for verifying the ability of that algorithm to properly identify heel strike (or some other time point of gait) needs to be established. During turning segments, the difference between heel strike identification becomes inconsistent. Neither of the methods used in this study were designed to handle turning data. Specifically, Method 1 was originally designed to analyze data from participants walking on a treadmill, where no turning would occur. Therefore, when using Methods 1 and 2 to segment strides from walking acceleration data, segments of turning data should first be removed. Further, this suggests that these methods of stride segmentation would be ineffective for data collected outside of a controlled environment for straight walking tasks.

5. Conclusions

Methods 1, 2, and 3 are all internally consistent methods of stride segmentation during straight walking (i.e. the standard deviations are low). However, Method 1 is able to more accurately identify heel strike compared to Method 2. Neither Method 1 or 2 is able to accurately segment stride or identify heel strike during turning, nor are they accurate for all subjects tested. Future studies should explore accurate methods of heel strike identification and stride segmentation that are more robust across people and movements so the methods can be applied to non-healthy people and people with irregular walking patterns in potentially non-traditional lab settings.

6. Acknowledgements

This study was funded by the Swanson School of Engineering, the Office of the Provost, and the Department of Bioengineering. Special thanks to our research mentor Dr. Mark Redfern, and Professors Bergamini and Vannozzi for providing the data for analysis.

7. References

[1] E. Sejdic, K.A. Lowry, J. Bellanca, S. Perera, M.S. Redfern, J.S. Brach, Extraction of Stride Events From Gait Accelerometry During Treadmill Walking, IEEE J. Transl. Eng. Health Med. 4 (2016) 1-11. [2] W. Zijlstra, Assessment of spatio-temporal parameters during unconstrained walking, Eur. J. Appl. Physiol. 92 (2004) 39-44. [3] I. Pasciuto, E. Bergamini, M. Iosa, G. Vannozzi, A. Cappozzo, Overcoming the limitations of the Harmonic Ratio for the reliable assessment of gait symmetry, J. Biomech. 53 (2017) 84-89. [4] H.B. Menz, S.R. Lord, R.C. Fitzpatrick, Acceleration patterns of the head and pelvis when walking on level and irregular surfaces, J. Gait and Posture. 18 (2003) 35-46.


Assessment of luminal fillers in nerve guidance conduits for promoting regeneration following peripheral nerve injury Carol Vinskia, Tyler Medera,b, Bryan Brown, Ph.D.a,b,c Department of Bioengineering, bMcGowan Institute for Regenerative Medicine, c Renerva, LLC


Carol Vinski is a native Pittsburgher. Her passion for the improvement of medicine and healthcare motivates her to become a biomechanical engineer in the field of prosthetics.

Carol Vinski

Dr. Bryan Brown is an Associate Professor in the Department of Bioengineering with secondary appointments in the Department of Obstetrics, Gynecology, and Reproductive Sciences and the Clinical and Translational Science Institute at the University of Pittsburgh. He is also a Bryan Brown, Ph.D. core faculty member of the McGowan Institute for Regenerative Medicine where he serves as the Director of Educational Outreach. Dr. Brown is also an Adjunct Assistant Professor of Clinical Sciences at the Cornell University College of Veterinary Medicine and Chief Technology Officer of Renerva, LLC, a Pittsburgh-based start-up company.

Significance Statement

Autologous grafting, the current clinical standard to treat peripheral nerve injury (PNI), often results in unsatisfactory nerve healing and always causes a loss of feeling at the donor site. This study analyzes the most effective laboratory nerve guidance conduits and luminal fillers to act as an alternative solution to autografting.

Category Review Paper

Keywords: luminal fillers, nerve guidance conduits, nerve repair, biomechanics of nerves

102 Undergraduate Research at the Swanson School of Engineering


Peripheral nerve injury (PNI) is a growing field of scientific research, and transected PNIs are difficult to repair due to the quick buildup of scar tissue blocking axon regeneration. Autologous grafting, the clinical standard of care, can cause loss of feeling at the donor site, and patients may not have enough viable donor tissue to complete this procedure depending on injury severity. To offer an effective alternative to autografting without loss of sensation, nerve guidance conduits (NGCs) are being explored but are not as successful as the clinical standard. Additives to NGC lumen (luminal fillers) are being researched to obtain results at least as successful as the autograft without its limitations. As of yet, no luminal fillers have entered into clinical use. The Brown Lab has previously developed a luminal filler derived from Porcine sciatic Nerve extracellular Matrix (PNM). To better understand how PNM performs in the context of NGC additives as a whole, the present study primarily examines reports of various luminal fillers to determine what characteristics predict success as a result of their application. Secondly, reports analyzing Schwann cell (SC) activity under variable biomechanical stimuli were considered as luminal fillers to provide a mechanical substrate which interacts with SCs. Overall, successful luminal fillers most often contained highly porous and permeable substrates with steep mechanical gradients for maximum SC activity. By utilizing this knowledge, successful luminal fillers and their respective conduits can be used to efficiently heal PNIs to maximize functional returns in patients.

Ingenium 2021

1. Introduction

Occurring in 3% of traumas, peripheral nerve injury (PNI) often requires intervention. Since only 50% of surgical repairs demonstrate a satisfactory recovery rate [1,2], this field of scientific research is growing quickly. Upon injury, development of scar tissue can rapidly outpace axon regrowth within the nerve, preventing functional returns and resulting in unsatisfactory PNI healing [3]. The clinical standard for repairing PNIs is using an autologous nerve graft from the patient’s own body to bridge the severed endings of the damaged nerve, but this process causes a loss of sensation in the donor site of the body and is not always an option for patients [4]. To offer an effective alternative to autografting, nerve guidance conduits (NGCs) have been developed; however, they are not as successful as autografts and thus have a reduced capacity to heal PNIs [5]. To improve their efficacy, NGCs are being explored in combination with additives within the lumen (luminal fillers) to provide a therapy of equal or greater value than autografting; examples of these are shown in Fig. 1.

However, no luminal fillers have entered into clinical use yet. The Brown Lab has previously developed a luminal filler derived from Porcine sciatic Nerve extracellular Matrix (PNM) [6]. To better understand how PNM performs in the context of NGC additives as a whole, the present study examines reports of various luminal fillers to determine which characteristics predict success as a result of their application. Important factors to consider while creating an NGC are the mechanics of the resting (quiescent) and activated nerve and Schwann cells (SCs), both of which contribute to peripheral nerve regrowth. As shown in Fig. 2, SCs are imperative to the regrowth of peripheral nerves.

Figure 2. Peripheral nerve regrowth after PNI. At the injury site, the distal end of the neuron breaks down all the way as the proximal end of the neuron breaks down a little bit. Next, SCs help the neuron vasculature to replenish, creating the Bands of Bungner. Finally, the axon regenerates along these mature vascular bridges.

Figure 1. NGCs and luminal fillers. In the center of the image, the nerve guidance conduit can be seen between the proximal and distal ends of the PNI. The hollow tube is the NGC (A), and it is filled with luminal fillers of aligned fibers (D, F), micro fillings (B, E), or hydrogel (G). The NGC is given grooves (C) or allows controlled release of NGFs via pore size or gradients (H).

This process of transitioning between the rest and active states is due to transcriptional regulators, such as c-Jun, an activator of “a repair program to support [nerve] regeneration” [7]. SCs are regulated by MAPK and other signaling systems as well, but substrate stiffness also plays an important role in SC transition between quiescence and acquiescence [7]. The purpose of this review article is to summarize research studies that focus on mechanical properties of the materials, to develop better understanding of the environment in which peripheral nerves live, and to promote manipulation of substrates in healing PNIs via NGCs and luminal fillers.

2. Methods

Papers were found in the Google Scholar database, using keywords and phrases such as “luminal fillers,” “peripheral nerve injury conduits,” “peripheral nerve regrowth,” and “Schwann cell growth and biomechanics.”


2.1 Luminal Fillers Papers

Studies applying luminal fillers in animal models were selected to compare their contributions towards nerve healing. Key properties, such as material choice and filler porosity or addition of SCs and/or neuronal growth factors (NGFs), were compared from these luminal fillers. All papers used can be found in Table 1.

2.2 C Biomechanics Papers

Studies examining SCs in response to mechanical cues were selected to find the ideal range of luminal filler mechanical properties that were conducive to PNI healing. Studies concerning SC mechanics and how SCs react to the substrate surrounding them were selected from in vitro studies with comparisons made between Young’s moduli, substrate stiffness/compliance, and mechanical gradients. All papers used can be found in Table 2.


NGC and Filler Properties

General Success

[2], [4], [6], [25], [30]

aligned fiber orientation in conduit or luminal filler

successful for travel of nutrients/waste

[12], [32], [33], [34]

NGFs in luminal filler

successful for improving PNI healing


unaligned fiber orientation in luminal filler

unsuccessful, made it harder to transport nutrients/waste and for SCs to travel on

[2], [7], [13], [20], [23], [25], [29], [31], [33]

conduit porosity

more successful with pore sizes between 10 and 40 µm

[1], [23], [25], [27], [33], [34]

luminal filler porosity

generally more successful with pores than without

[17], [11], [26]

PNM in conduit or luminal filler

more successful when PNM is used as the sole luminal filler

[4], [6], [5], [9], [10], [12], [14], [17], [18], [23], [29]

SCs added to conduit or luminal filler

more successful when aligned

[1], [4], [3], [7], [9], [11], [12], [13], [14], [15], [21], [22], [25], [30]

collagen in conduit or luminal filler

more successful when collagen is used as a supportive structure

[3], [6], [8], [11], [13], [14], [17], [19], [22], [24], [26], [29], [30], [34]

type of gel or gellan gum within conduit or luminal filler

more successful when gel is broken down to allow for controlled release of SCs/NGFs/etc.

Table 2: All References for Schwann Cell Biomechanics

[26], [28], [29]

PCL in conduit and/or luminal filler

more successful when paired with factors that contribute to SC proliferation or PNI healing

3.1 Luminal Filler

[1] Aigner, T. B. et al (2020). doi:10.1016/j.mtbio.2020.100042 [2] Cai, J. et al (2005). doi:10.1002/jbm.a.30432 [3] Carriel, V. et al (2013). doi:10.1088/1741-2560/10/2/026022 [4] Ceballos, D. et al (1999). doi:10.1006/exnr.1999.7111 [5] Dietzmeyer, N. et al (2020). doi:10.1177/0963689720910095 [6] Du, J. et al (2017). doi:10.1016/j.actbio.2017.04.010 [7] Ezra, M. et al (2016). doi:10.1089/ten.tea.2015.0354 [8] Gnavi, S. et al (2014). doi:10.1002/term.1936 [9] Gu, Y. et al (2016). doi:10.1002/term.2123 [10] Kalbermatten, D. F. et al (2008). doi:10.1016/j.bjps.2007.12.015 [11] Kohn-Polster, C. et al (2017). doi:10.3390/ijms18051104 [12] Lackington, W. A. et al (2018). doi:10.1016/j.actbio.2018.06.014 [13] Lee, Y. S. et al (2016). doi:10.1002/jbm.a.35630 [14] Lohmeyer, J. A. et al (2007). doi:10.1177/039139880703000109 [15] Matsumoto, K. et al (2000). doi:10.1016/s0006-8993(00)02207-1 [16] Matsuura, S. et al (2006). doi:10.1016/j.urology.2006.09.051 [17] Meder, Tyler et al (2019). doi: 10.1002/jbm.a.36235 [18] Meyer, C. et al (2016). doi:10.3727/096368915x688010 [19] Nakayama, K. et al (2007). doi:10.1111/j.1525-1594.2007.00418.x [20] Nectow, A. R. et al (2012). doi:10.1089/ten.teb.2011.0240 [21] Newman, K. D. et al (2006). doi:10.1177/039139880602901109 [22] Pace, L. A. et al (2013). doi:10.1089/ten.tea.2013.0084 [23] Rao, J. et al (2017). doi:10.1016/j.msec.2016.12.085 [24] Roldán-Carmona, Cristina, et al (2014). doi: 10.1039/x0xx00000x [25] Ryan, A. J. et al (2017). doi:10.1002/adhm.201700954 [26] Sun, A. X. et al (2019). doi:10.1016/j.biomaterials.2019.01.038 [27] Sun, B. et al (2017). doi:10.1021/acsami.7b06707 [28] Tonda-Turo, C. et al (2011). doi:10.1002/adem.201080099 [29] Tonda-Turo, C. et al (2014). doi:10.1002/term.1902 [30] Verdú, E. et al (2002). PMID:12515893 [31] Wang, X. et al (2005). doi:10.1093/brain/awh517 [32] Wood, M. D. et al (2009). doi:10.1016/j.actbio.2008.11.008 [33] Wu, H. et al (2018). doi:10.1016/j.jmbbm.2017.10.031 [34] Zhang, L. et al (2016). doi:10.1007/s11595-016-1545-y

Table 1: All References for Luminal Fillers

104 Undergraduate Research at the Swanson School of Engineering



[3], [8], [12]

impact of altering Young's modulus

[5], [6], [7], [10], [13]

mechanical properties of the microenvironment of healthy and injured peripheral nerves

[1], [2], [4], [11], [14]

Schwann cell growth and quiescence patterns

[1] Boerboom, A. et al (2017). doi:10.3389/fnmol.2017.00038 [2] Evans, E. et al (2018). doi:10.1186/s40824-018-0124-z [3] Gittes, F. (1993). doi:10.1083/jcb.120.4.923 [4] Gu, Y. et al (2012). doi:10.1016/j.biomaterials.2012.06.006 [5] Hunt, G. C. (2002). doi:10.1179/108331902125001888 [6] Janmey, P. A. et al (2019). doi:10.1152/physrev.00013.2019 [7] Ju, M. S. et al (2017). doi:10.1299/jbse.16-00678 [8] Manssor, N. A. S. et al (2016). doi:10.1159/000431328 [9] Meder, Tyler et al (2019). doi: 10.1002/jbm.a.36235 [10] Pfister, B. J. et al (2020). doi:10.1016/j.cobme.2020.05.009 [11] Rosso, G. et al (2017). doi:10.3389/fnmol.2017.00345 [12] Rosso, G. et al (2019). doi:10.1063/1.5108867 [13] Sundararaghavan, H. G. et al (2009).doi:10.1002/bit.22074 [14] Xu, Z. et al (2019). doi:10.1002/term.2987

3. Results Using the selection criteria in 2.1, 33 studies were examined for luminal filler efficacy. Overall, luminal fillers were found to be effective at promoting nerve healing through increased SC migration or neurite extension when containing aligned fiber orientation or added NGFs. Since SCs travel along straight fibers during nerve healing, as seen in Fig. 2, aligned fibers must produce more successfully healed PNIs than luminal fillers containing densely packed or unaligned substrates, which had negative effects on PNI healing [8]. Similarly, NGFs are neuronal growth factors and thus improve nerve regrowth, so adding more NGFs to a luminal filler or NGC can only increase PNI healing. It was observed that key predictors between successful and unsuccessful luminal filler were fiber alignment and substrate porosity [9,10]. Pores in the luminal fillers allow nutrients and NGFs in and allow waste to flow out. If the pore size is too large, scar tissue infiltrates into the space where the nerve is trying to regrow and thus keeps the nerve from reconnecting and giving functional returns. If pores are too small, waste cannot leave and nutrients and NGFs cannot enter the area to allow PNI healing [11]. Thus, NGC porosity over long gap defects is a key determining factor in the success of tissue within the lumen; for these PNIs, ideal pore size ranged from 10 to 40 μm [12]. Some gap sizes, however, are small enough that they do not require pores for removal of waste and diffusion of nutrients and NGFs because such PNIs heal much faster

Ingenium 2021

than larger gap sizes. Thus, small gap sizes do not require pore size to be a significant contributing factor in NGC and luminal filler success.

3.2 SC Biomechanics

From the 11 studies examining SCs with mechanical cues, SC function was found to be greatly affected by the Young’s moduli in the substrate surrounding them. In Gu’s study, the 1-3 day old rat sciatic nerves grew most successfully on a Young’s modulus of 7.45 kPa as opposed to 9.10 kPa or 12.04 kPa [13], and another study found that 8.67 kPa worked the best out of the other stiffnesses used [14]. Thus, an elastic modulus too high or low will result in less viable SCs, suggesting that there is a theoretical exact Young’s modulus value that will promote nerve healing better than any other number. In addition to specific Young’s moduli, SCs preferred travelling down steep gradients when they transitioned to their activated state as opposed to staying on a flat surface. The presence of the gradient pushed the SCs to their acquiescent state for PNI healing through increased spread, elongated nuclei, and increased speed and direction [15]. However, substrates continuously promoting SC activation revealed difficulty in termination of the activation response, making it difficult for the nerve to return to its quiescent state.

4. Discussion 4.1 Luminal Fillers

From 3.1, it can be inferred that the key materials for successful luminal fillers are aligned fibers over all gap sizes and specific porosity over large gap sizes. Overall, aligned fibers within NGCs and luminal fillers, as in Fig. 2.D,F, greatly improved the success of PNI healing. In addition, pore sizes large enough for nutrients to enter but small enough that scar tissue cannot invaginate into the NGCs and luminal fillers was imperative for success over large gaps. If the gap is small enough, porosity is not an issue due to the short amount of time it will take for the PNI to heal compared to the amount of time it takes for scar tissue to infiltrate the regenerating nerve.

4.2 SC Biomechanics

From 3.2, it can be inferred that it is theoretically best to use a material within the luminal filler that begins with a stiffness gradient to promote PNI healing but is remodeled over time to mechanically match native nerves and promote SC quiescence. This would best utilize SCs to help improve nerve regrowth, as seen in Fig. 2, while preventing SCs from having an elongated activation response.

5. Conclusions

The luminal filler should be used in combination with NGFs with pores of appropriate size and that allow exchange of waste, nutrients, and NGFs. When these criteria are not met, it is possible for a luminal filler to promote invagination of scar tissue or inhibit axon penetration, decreasing healing as a result. Successful luminal fillers most often contained highly porous and permeable substrates with either aligned fibers or added NGFs. Increased success could be observed with a steep initial gradient to promote SC activation and migration. As the luminal filler gradient degrades, natural tissue will deposit, leaving a suitable environment for SCs to return to their quiescent state. This process allows for increased functional returns. The conclusions made in this study could be pursued to improve patient healing of PNIs.

6. Acknowledgements

Funding was provided by the Swanson School of Engineering, the Department of Bioengineering, and members of the Brown Lab including PI, Dr. Bryan Brown, and my student mentor, Tyler Meder. Funding for the nerve project was provided by the Wallace Coulter Foundation and the PInCH program through the University of Pittsburgh.

7. References

[1] C. A. Taylor, D. Braza, J. B. Rice, T. Dillingham. The incidence of peripheral nerve injury in extremity trauma, Am J Phys Med Rehabil. 87(5) 381-5, 2008. [2] A Portincasa, G Gozzo, D Parisi, L Annacontini, A Campanale, G Basso, A Maiorella, Microsurgical treatment of injury to peripheral nerves in upper and lower limbs: a critical review of the last 8 years Microsurgery. 27(5) 455–462, 2007. [3] Ngeow et al, Scar less: a review of methods of scar reduction at sites of peripheral nerve repair, Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology, and Endodontology. 109(3) 357–366, 2010 [4] D. Grinsell, C. P. Keating, Peripheral Nerve Reconstruction after Injury: A Review of Clinical and Experimental Therapies, Biomed Res Int 2014. [5] Brattain. Magellan Medical Technology Consultants, Inc. p. 1-11, 2013. [6] T. A. Prest, E. Yeager, S. T. LoPresti, Nerve-Specific, Xenogeneic Extracellular Matrix Hydrogel Promotes Recovery Following Peripheral Nerve Injury, J Biomed Mater Res A. 106(2) 450-459, 2018.


[7] A. Boerboom,V. Dion, A. Chariot, R. Franzen, Molecular Mechanisms Involved in Schwann Cell Plasticity, Frontiers in Molecular Neuroscience, 2017. [8] Verdú E, Labrador RO, Rodríguez FJ, Ceballos D, Forés J, Navarro X, Alignment of collagen and laminin-containing gels improve nerve regeneration within silicone tubes, Restor Neurol Neurosci., 20(5):169-179, 2002. [9] Du, J., Liu, J., Yao, S., Mao, H., Peng, J., Sun, X., … Wang, X, Prompt peripheral nerve regeneration induced by a hierarchically aligned fibrin nanofiber hydrogel, Acta Biomaterialia, 55, 296–309, 2017. [10] Cai, J., Peng, X., Nelson, K. D., Eberhart, R., & Smith, G. M., Permeable guidance channels containing microfilament scaffolds enhance axon growth and maturation, Journal of Biomedical Materials Research Part A, 75A(2), 374–386, 2005. [11] Rao, J., Cheng, Y., Liu, Y., Ye, Z., Zhan, B., Quan, D., & Xu, Y., Materials Science and Engineering: C, 73, 319–332, 2017. [12] A. R. Nectow, K. G. Marra, D. L. Kaplan, Biomaterials for the Development of Peripheral Nerve Guidance Conduits, Tissue Engineering Part B: Reviews, 18(1), 40–50. [13] Gu, Y., Ji, Y., Zhao, Y., Liu, Y., Ding, F., Gu, X., Yang, Y, The influence of substrate stiffness on the behavior and functions of Schwann cells in culture, Biomaterials, 33(28), 6672–6681, 2012. [14] Xu, Z., Orkwis, J. A., DeVine, B. M., & Harris, G. M., Journal of Tissue Engineering and Regenerative Medicine, 2019. [15] Evans, E., Brady, S., Tripathi, A., Schwann cell durotaxis can be guided by physiologically relevant stiffness gradients, Biomater Res 22, 14 (2018).

106 Undergraduate Research at the Swanson School of Engineering

Ingenium 2021

Neural Network-based approximation of model predictive control applied to a flexible shaft servomechanism Yuanwu He, Yuliang Xiao and Nikhil Bajaj Department of Mechanical Engineering and Materials Science Yuliang Xiao is from Beijing, China. His interests in robotics and strong education background motivate him to take further research on Robotics and Computer Vision in graduate.

Yuliang Xiao

Yuanwu He was born and raised in Beijing, China. He is passionate about system controls and robotics, which are also the fields he would like to dive in for his career of engineering.

Yuanwu He

Nikhil Bajaj, Ph.D.

Dr. Bajaj is an Assistant Professor in the Department of Mechanical Engineering and Materials Science. He received his PhD from Purdue University in 2017 and served as a postdoctoral researcher from 2017-2019. His current research interests include nonlinear dynamics, mechatronics, advanced sensors, and advanced manufacturing.

Significance Statement

This work seeks to address the computational burden associated with Model Predictive Control (MPC) algorithms, and specifically demonstrates the improvement of the memory footprint of an explicit MPC via the use of feedforward neural networks for approximation of the controller.

Category: Computational Research

Keyword: model predictive control (MPC), machine learning, feed-forward neural network, computational cost.


In this paper, we seek to address the computational burden associated with Model Predictive Control (MPC) algorithms. MPC algorithms seek an optimal control action over a control horizon that drives a discrete time control system according to the determined reference. It is always used in the process industries in chemical plants, oil refineries, power system balancing models and power electronics. One critical limitation of MPC is the computational complexity, which constrains a system’s ability to find an optimal solution in real-time, resulting in limitations on the classes of systems for which MPC can be used. To address this problem, the major findings are to develop explicit MPC (eMPC) algorithms, which precompute the optimal solutions across the state, reference, and input space, and interpolate between optimal solutions via different methods. In general, however, eMPCs are characterized by trading off computational burden in real-time for memory allocation size [1]. In this work we present the successful use of neural networks to approximate a MPC control algorithm for a particular servomechanism and evaluate their performance and utility.

1. Introduction

Model Predictive Control (MPC) is a popular family of control algorithms in a wide and expanding variety of fields [2]. MPC finds the optimal control action over a control horizon that drives a discrete time control system output to the determined reference while honoring the system constraints. Hence, it is a good tool to handle multi-input multi-output systems that interact with inputs and outputs since designing this system by using traditional controllers like PID is challenging. MPC can help solve constraints because it has predictive capabilities, which computes the trajectories of each state in advance in order to select the best control action according to a cost function. This is why engineers often use MPC to get detailed control of each state during a dynamic process, such as chemical plants, oil refineries, and power balancing systems. The optimization problem iteratively evaluates the expected system response at each step over a prediction horizon. Compared to trajectory planning in the task space or the determination of required control commands in the state space [2], MPC is an advanced technique and can generally provide good performance. The basic MPC algorithm can be computationally expensive. When the prediction horizon is large, a limitation of this approach is its computational burden since the solutions need to be computed in real-time between control actions. This restricts the applicability of MPC to systems that have large time constants and/or high computational power [1]. Decreasing computational cost is a desirable objective, as it will allow for MPC to be practical in more systems. A common way of addressing this is with an “explicit” MPC (eMPC), which trades off memory for computational complexity. The explicit MPC is a way to manage the computational load by pre-computing the optimal control law u* = μ*(x) offline as a function of all feasible states x [3]. 107

The eMPC computes the responses offline, and then fits a sequence of hyper-surfaces to the derived responses, giving an approximation of the performance of the MPC. In this research, an alternative approach is proposed. A direct approximation of the MPC’s computed response may be represented with a combination of states and reference inputs via a feed-forward neural network. This method allows for an intermediate solution between the two, with the potential for smoother fitting of the response function due to the universal function approximation properties of the neural network. It contains a tightly connected set of non-linear neurons which approximate the function of the MPC. The new methodology will be applied to a simulated flexible shaft electromechanical servomechanism, which consists of a DC motor, a gearbox, an elastic shaft, and loads. The goal is to use Neural Network to reduce computational efforts while approximating to the performance of MPC.

2. Method

The governing differential equations representing the position servomechanism system are used to define the state-space representation for the plant model. The system has one control input which is voltage applied to the DC motor. One output gives the measurement of the angle of shaft and the torque is estimated as an unmeasured output. The model constants are substituted into the equations and implemented in a state-space form. A linear model predictive controller was designed using MATLAB’s MPC Designer [4] tool with the derived state equations. This designs a linear MPC, along with an estimator that can easily be tuned to the desired aggressiveness of the response. Although the focus of this work was on approximating the response of the MPC, the actual performance is also important. The MPC computes the input at time k, uk: uk = f MPC (x, r, uk-1)

(1) [2]

Here, x is the vector of states at time k, r is the angle reference and uk-1 is the vector of manipulated variables at time k-1 to be determined by the controller. f MPC (x, r, uk-1) is the result of the optimization problem solved by the MPC given the states, references, and prior inputs. Data was generated based on this function and used as a training set to feed into the Neural Net Fitting App. The goal of the neural network is to reproduce this function, with the neural network, e.g. f NN (x, r, uk-1) ≈ f MPC (x, r, uk-1) . After creating a neural network model, the results of outputs were derived from implementation in Simulink. Eventually, the performance of newly designed feed-forward neural network model was evaluated by doing error calculations and data comparison. Typically, MSE and plots comparison are used to evaluate approximation of neural network-based approach. Also, the computation time versus the MPC is improved. The computation time of the MPC can be drastically reduced by approximating the MPC law with a Neural Network controller [2].

108 Undergraduate Research at the Swanson School of Engineering

For neural network, the number of neurons in hidden layer has some effects on its performance. From Gnana et al., the output errors decrease with being added, where Nh is the number of neurons in the hidden layer [5]. In the book, two equations are introduced to explain the relationship between number of neurons in hidden layer and network performance:

Where N is number of training pairs, d is input dimension, and Cf is first absolute moment of Fourier magnitude distribution of target function. Also, MSE could be used to evaluate the performance of network. Based on the functions above, it is possible to construct a mapping function between Nh and MSE. As a result, there is a relationship between number of hidden neurons and MSE. In the following sections, we show results about the effects of the number of hidden neurons on neural network performance.

3. Data Processing

At first, the training samples are gained from the MPC model by using the MATLAB command “mpcmove”. “mpcmove” evaluates f MPC (x, r, u k-1 ) by providing the required states and parameters. The training data features are x ∈ ℝ4, r, uk-1 ∈ ℝ for this system. Therefore, it is rational to have six input features, and one output for the neural network. The range of the six input variables used in “mpcmove” is based on the physical constraints of the system so that the precision is maximized. Since changing the value of the interval affects the amount of training data gained from MPC, we chose a suitable value that could produce sufficient training examples while considering constraints on computation time. There were 1932612 training examples produced during the training. 70% of these were used as a training set, and 15% used in testing, and 15% in training algorithm validation. Then, a neural network tool was used to train data gained from MPC. Four network architectures, differing in their number of neurons in the hidden layer, were tested, and each of them has different testing performance in error calculations (i.e. mean square error (MSE), regression (R), etc.). The specific values of error are shown in the Table 1. 15 Neurons

25 Neurons

30 Neurons

40 Neurons











Table 1: MSE and Regression for different tests

Next, the Neural Network Fitting App provided a neural network function to implement as a Simulink block, which was used to replace the MPC block. The angle outputs and control effort were saved in the workspace for further data processing.

Ingenium 2021

To determine which number of neurons has better performance, further calculations in relative error against MPC and against reference were implemented by using the data saved in the workspace.

4. Results

Overall, from comparing data generated in Simulink, although there are some differences between machine learning and MPC, the Neural Network could approximate the MPC to a great extent. However, not all four tested numbers of neurons are appropriate for the final design. There was a fluctuating trend in error as the number of neurons was increased.

In Figure 2 (a), it shows the absolute angle outputs error against reference for 15 neurons, 25 neurons, 30 neurons, and 40 neurons. The Figure 2 (b) shows the error of neuron network result against reference. As can be reasonable expected, the 30-neuron network has the least error of the tested architectures, and as a result, the 30-neurons Neural Network block was used in the final Simulink model to test the performance compared with MPC model.

Figure 1: Error against MPC versus time for different tests

In Figure 1, it shows how relative error against MPC model changes versus time for four tests. Here the interest mainly focuses on vertical comparison, which is the difference of errors between different numbers of neurons, since the goal is to find the most suitable number of neurons in the same condition. It can be seen from the values on vertical axis that tests with 15 neurons and 25 neurons have higher relative error than 30 neurons and 40 neurons, which means the latter two could be chosen for a final implementation. The choice between 30 neurons and 40 neurons was not intuitive since 30 neurons has relatively small error from T = 0 to T = 10, while 40 neurons have relatively small error in the rest of the time. Therefore, the error of angle outputs was calculated.

Figure 2: Angle error, and error against reference versus time.

Figure 3: Plot of comparison of MPC and Neural Network in same plant.

In Figure 3, it shows the output from original MPC and the one from the system with implementation of Neural Networks. The horizontal axis refers to the time and vertical axis refers to the output voltage. The orange line is the original output of the MPC model, whereas the blue line is result coming after the 30-neuron Neural Network block. Qualitatively, the result of the Neural Network matches MPC model to a great extent.

5. Discussion

Figure 2 gives a brief illustration on how error changes as the number of neurons increases. Since there are only four different numbers of neurons tested, the overall trend could not be fully explained. 15-neuron and 25-neuron Neural Network have larger error than 30-neuron Neural Network, while 30-neuron has less error than 40-neuron Neural Network. However, changing the number of neurons does affect the voltage output and the relative error between the original MPC model and the one with Neural Network-based approximation. Therefore, to find a suitable number of neurons is a trial-and-error task. If appropriate number of neurons is not selected, model can be overfitting and these lead to wrong results. This means that the Neural Network model gives better result at and near states and conditions associated with training examples, but it may give worse results farther away from 109

such points. In the research done by M. Yilmaz and E. Arslan [6], the authors document a situation similar to the one explored in this paper, where the appropriate number of neurons used in their research was below 30, and the output of their model was overfitting when the number of neurons is selected larger than 30. The appropriate number of neurons is application dependent and is based on a tradeoff between training complexity, on-line complexity, and achieved error performance. For this application, Neural Network-based approximation towards MPC is feasible since the error between original and tested one is small enough to accept. In Figure 3, the control effort of MPC (blue line) and machine learning (orange line) are exhibited. We can clearly conclude that the output voltage of MPC system and the system with implementation of Neural Networks are approximately in a same trace, while the latter running much faster than MPC. Due to the limitation of computer memory, higher neuron tests are not implemented, but this does not imply that higher neuron performs worse. The performance is expected to be improved further at a specific number of neurons, but this claim must be tested with more accurate experiments to draw a conclusion. Furthermore, determining the range and interval of six input variables during data process still has some scope to develop. For example, the range of angle speed may vary between some values which are based on the output of neural network. One could plot the variation of this variable when using neural network to better understand the maximum and minimum value on the plot, and then redefine the range to obtain a good training sample. As for the intervals between training examples in the input space for the neural network, decreasing it means greater numbers of training examples, which could be beneficial for neural network training. In this work, we chose a relatively large interval to perform the proof of concept, but this may have been a suboptimal set of training examples. As a result, in future research, it will be important to study the effect of sampling techniques in the state and input space to the MPC to improve the training data.

110 Undergraduate Research at the Swanson School of Engineering

6. Conclusion

Above all, the approximation of MPC via Neural Network is feasible, and its performance depends on the range of data set and the number of hidden layers. The Neural Network generates an acceptable result compared to original MPC model while shortening a great amount of computation time. For further MPC applications in industries, neural network-based approximation could decrease real-time cost in the process of dynamic controls.

7. References

[1] S. Paternain, M. Morari and A. Ribeiro, Real-Time Model Predictive Control Based on Prediction-Correction Algorithms, 2019 IEEE 58th Conference on Decision and Control (CDC), Nice, France, 2019, pp. 5285-5291, doi: 10.1109/CDC40024.2019.9029408. [2] J. Nubert, J. Köhler, V. Berenz, F. Allgöwer, S. Trimpe, Safe and Fast Tracking on a Robot Manipulator: Robust MPC and Neural Network Control (v2), 2020. [3] S. Chen et al., Approximating Explicit Model Predictive Control Using Constrained Neural Networks, 2018 Annual American Control Conference (ACC), Milwaukee, WI, 2018, pp. 1520-1527, doi: 10.23919/ACC.2018.8431275. [4] MATLAB and Model Predictive Control Toolbox Release 2020b, The Mathworks, Inc., Natick, Massachusetts, United States. [5] K. Gnana, S. N. Deepa, Review on Methods to Fix Number of Hidden Neurons in Neurons, 2013. [6] M. Yilmaz, E. Arslan, Effect of increasing number of neurons using artificial neural network to estimate geoid heights, January 10, 2011.

Finite element analysis of stents under radial compression boundary conditions with different material properties Yaoyao Xua , Jonathan P. Vande Geestb


Department of Mechanical Engineering and Material Science, Department of Bioengineering and Department of Mechanical Engineering and Material Science b

Yaoyao Xu is a mechanical engineering undergraduate at Swanson School of Engineering. Her current interests are control, finite element analysis, and computational modeling. She plans to pursue a master’s degree in control after graduation. Yaoyao Xu

Dr. Jonathan Vande Geest is a Professor in the Department of Bioengineering, Department of Mechanical Engineering and Material Science, the Department of Ophthalmology, the McGowan Institute for Regenerative Medicine, the Louis J. Fox Center for Vision Restoration, and the Vascular Medicine Institute at the Jonathan P. Vande University of Pittsburgh. He received his Geest, Ph.D. BS in Biomedical Engineering from the University of Iowa in 2000 and his PhD in Bioengineering from the University of Pittsburgh in 2005. Dr. Vande Geest began his career at the University of Arizona in the Department of Aerospace and Mechanical Engineering and joined the University of Arizona’s Department of Biomedical Engineering in 2009. Dr. Vande Geest returned to the University of Pittsburgh in January of 2016.

Significance statement

By simulating the reactions of the cardiovascular stents in-vivo with radial compression acts as the boundary conditions, the relationships between the displacements, material stiffness, and the maximum principal stress could be found, which provides references to the stent design when large deformation causes problems to the performance of the stent.

Ingenium 2021


The purpose of this work is to begin to establish a computational testing platform to assess the compressive behavior of magnesium alloy stents. This provides references to the stent design when large deformation causes problems to the performance of the stent. According to the results, the sensitivity of peak maximum principal stress to changes in imposed displacement increased as the stiffness of the stent increased. This sensitivity became nearly linear in both stiffness and imposed displacement when these values were near their maximum imposed values.

1. Introduction

The incidence of the cardiovascular and cerebrovascular disease has become higher and higher in recent years and has become the main threat to human life and health [1]. About 80% of patients with the cerebrovascular disease have the ischemic stroke, the main cause of which is the narrowing of blood vessels [2]. A cardiac stent is used to treat narrowed or blocked coronary arteries. Stent restenosis is a major problem resulting in device and treatment failure and often requires subsequent reintervention. The method of finite elements analyzes the structural characteristics that affect the support strength and safety of vascular stents. Through mechanical simulation and numerical study of the releasing and operating process of vascular stents in the stenosis model, the relevant factors that influence the effect of stent expansion can be identified, which can be provided to the physician when using stent implantation as a valuable reference. Previous research has found that restenosis within balloon-expandable endovascular stents may occur as a result of radial compression [3], and radial compression acts as one of the common boundary conditions for cardiovascular stents in-vivo [4]. Based on Hooke’s law, we assume that the maximum principal stress of the stent is inversely proportional to material stiffness with the same displacement, and as the displacement increases, the stress will also increase. The purpose of this work is to begin to establish a computational testing platform to assess the compressive behavior of magnesium alloy stents. This provides references to the stent design when large deformation causes problems to the performance of the stent.

Category: Computational Research Keywords: Simulation, Stent, Abaqus, Radial compression, Stiffness


2. Methods 2.1 Materials and Modeling

The stent geometry showed in Figure 1 was created in SolidWorks using an open-source cardiovascular stent geometry as the starting point. The diameter of the stent was created to be 3.2mm and the length is 16mm [5].

2.2 Data Processing

The negative displacement imposed on the top set is defined by how many percent of the diameter of the stent. There are four groups of displacements, 20%, 40%, 60%, and 80% of the initial diameter. In each group of displacements, four tests with different material stiffness were made. The four tests have the corresponding Young’s modulus to be 0.1MPa, 100MPa, 5000MPa, and 45000 MPa. The Poisson’s ratio was fixed to be 0.3 for all tests. The total sixteen tests give sixteen maximum principal stress as the result of the compression which was visualized using a contour plot. The maximum principal stress for each test was collected and used in this plot.

3. Results and Discussion

Figure 2 below displays the original stent and the deformed stent following compression.

Figure 1. Stent geometry

After importing the stent geometry into Abaqus CAE as a part, several parameters are defined to simulate the radial compression of the stent. The geometry of the stent did not change for any of the simulations. Magnesium represents a very attractive material for biodegradable stents since the process of its natural and gradual dissolution into the human body by a corrosion process would prevent restenosis risks and would allow the progressive transmission of the mechanical load to the surrounding tissues after several months of service [6]. Therefore, the stent material is defined to be mechanical linear elastic material of AZ31-O magnesium alloy. The Young’s Modulus will be changed as the parameter in a specific range. The model uses a second-order tetrahedral element. The approximate global element seeding size was 1. To simulate radial compression, four sets of nodes are imposed different boundary conditions. The first two sets are the left and right ends of the stent. The third set is the bottom part of the stent by selecting the nodes with the minimum Y position. The fourth set is the top part of the stent by selecting the nodes with the maximum Y position. The first two sets have fixed longitudinal displacement (Z direction) to prevent the stent from moving in the longitudinal direction. The third set is fixed in all 3 directions. The fourth set is “compressed” by imposing a negative Y displacement. The output was the peak maximum principal stress in the model. The imposed Y displacement and Young’s Modulus were varied in order to understand how the peak stress nonlinearly changes with these two parameters. A mesh convergency study was performed prior to starting the sensitivity study.

112 Undergraduate Research at the Swanson School of Engineering

Figure 2. Upper part is the initial stent geometry; the lower part is the deformation of the stent.

Table 1 below displays the maximum principal stress of different material properties and compression levels. Different mesh densities were tested with the same model and boundary conditions to make sure that the results in this table were mesh converged.

Table 1. Maximum Principal Stress of The Whole Stent

Ingenium 2021

Figure 3 is the plot of the relationship between the data in Table 1. The color changes represent the changes in stress with corresponding material and imposed Y displacement, and the values in the plot represent the maximum principal stress of the stent in MPa.

4. Conclusion

The methods in this study established a computational testing platform to assess the compressive behavior of magnesium alloy stents, and successfully analyzed the stiffness of materials that affects the reactive stress of vascular stents. Specifically, the results of this study showed that the sensitivity of peak maximum principal stress became nearly linear in both stiffness and imposed displacement when these values were near their maximum imposed values. Chiefly, though, the significance of this study is to find the relationship between the maximum principal stress, the displacement, and the material stiffness for stents under radial compression. Importantly, the results provide references to the stent design when large deformation causes problems to the performance of the stent.

5. Acknowledgments

This research is supported by the Center for Medical Innovation (CMI) (Grant No: F 274), and funded by Swanson School of Engineering, University of Pittsburgh. Figure 3. Different color represents different maximum principal stress value in MPa

According to the output from the simulation, the top part and bottom part of the stent will suffer the highest principal stress, but the specific element which has the highest stress could change with the changes of each parameter. The maximum principal stress within the stent under radial compression is proportional to both the modulus of elasticity and the percentage of radial displacement. The sensitivity of peak maximum principal stress to changes in imposed displacement increased as the stiffness of the stent increased as expected. This sensitivity became nearly linear in both stiffness and imposed displacement when these values were near their maximum imposed values. Our results indicate that stent designs utilizing stiffness values lower than 1 to 1.5 MPa will display low peak maximum stress values. The current work will need to be extended in the future to include different geometries as well as including a model depicting the degradation of the biodegradable magnesium material [7].

6. References

1. Levi F, Lucchini F, Negri E, et al Trends in mortality from cardiovascular and cerebrovascular diseases in Europe and other areas of the world Heart 2002; 88:119-124. 2. ZHOU ZM, YIN Q, XU GL, et al. Influence of vessel size and tortuosity on in-stent restenosis after stent implantation in the vertebral artery ostium[J]. Cardiovascular and Interventional Radiology, 2011, 34(3): 481-487. 3. Rosenfield, K., Schainfeld, R., Pieczek, A., Haley, L., & Isner, J. (1997). Restenosis of Endovascular Stents From Stent Compression. Journal Of The American College Of Cardiology, 29(2), 328-338. DOI: 10.1016/s07351097(96)00498-6 4. Sullivan, S.J.L., Dreher, M.L., Zheng, J. et al. Effects of Oxide Layer Composition and Radial Compression on Nickel Release in Nitinol Stents. Shap. Mem. Superelasticity 1, 319–327 (2015). https://doi.org/10.1007/s40830-0150028-x 5. LEE, C. H. et al. Differential cutoff points and clinical impact of stent parameters of various drug-eluting stents for predicting major adverse clinical events: An individual patient data pooled analysis of seven stentspecific registries and 17,068 patients. International Journal of Cardiology, [s. l.], v. 282, p. 17–23, 2019. DOI 10.1016/j.ijcard.2019. 6. Farè, Stefano, Ge, Qiang, Vedani, Maurizio, Vimercati, Gianmarco, Gastaldi, Dario, Migliavacca, Francesco, Petrini, Lorenza, & Trasatti, Stefano. (2010). Evaluation of material properties and design requirements for biodegradable magnesium stents. Matéria (Rio de Janeiro), 15(2), 96-103. 7. Moravej, M.; Mantovani, D. Biodegradable Metals for Cardiovascular Stent Application: Interests and New Opportunities. Int. J. Mol. Sci. 2011, 12, 4250-4270


Index Category: Computational Research

Category: Experimental Research

Eve C. Ayar, Michelle R. Heusser Representations of population activity during sensorimotor transformation for visually guided eye movements...................................................................7

Francis J. Fattori Hardware acceleration of k-means clustering for satellite image compression.............. 20

Benjamin Bailey Simulating the effect of different structures and materials on OLED extraction efficiency................................................................................................ 13 Peiyun Feng, Liangyan Hao Calphad modeling and prediction of magnetic phase diagram for the Fe-Cr-Ni system.......................................................................................................... 25 Eileen N. Li, Zhong Li, Ryan M. Ronczka Fluid flow simulation of microphysiological knee joint-on-a-chip................................... 49 Jacob C. McDonald Bioinformatic analysis of fibroblast-mediated therapy resistance in HER2+ breast cancer....................................................................................................... 60 Oreoluwa Odeniyi, Megan Routzong Levator Ani muscle dimension changes with gestational and maternal age................... 64 Tyler Paplham, Jakub Toman Development of a post-processing method for estimating solidification parameters from finite-element modeling of laser remelting in directed energy deposition of Ni-Mn-Ga magnetic shape-memory alloy single crystals................................................ 68 Brady C. Pilsbury Demonstrating the antibiofouling property of the Clanger cicada wing with ANSYS Fluent simulations................................................................................................ 72 JonPaul Plesniak, Matthew Dent Identifying potential mammalian heme/CO sensors through a bioinformatic analysis of the Per-Arnt-Sim (PAS) protein domain........................................................................ 76 Caitlin Sexton, Trevor Sleight Topological descriptor selection for a quantitative structure-activity relationship (QSAR) model to assess PAH mutagenicity...................................................................... 81 Yuanwu He, Yuliang Xiao Neural Network-based approximation of model predictive control applied to a flexible shaft servomechanism.................................................................................... 107 Yaoyao Xu Finite element analysis of stents under radial compression boundary conditions with different material properties................................................................................... 111

Category: Device Design Lucy Jones Testing the compressive stiffness of endovascular devices............................................. 40 Nicholas P. Lagerman Design and manufacture of two experimental test benches for the in vitro assessment of a method to perform in situ fenestration of stent grafts.......................... 44 Sophie Shapiro, Stuart McCutchen, Zeve Cohen, Elizabeth Gilman Innovations for reimagined educational experiences in the face of a pandemic............. 85

114 Undergraduate Research at the Swanson School of Engineering

Elizabeth J. Mountz, Jinghang Li Association between functional dedifferentiation and amyloid in preclinical Alzheimer’s Disease......................................................................................................... 55 Jami Stuckey, Michelle Tsizhovkin User-specific factors associated with ladder handrail use............................................... 90 Joseph Sukinik, Rosh Bharthi, Sarah Hemler Relationship between center of pressure and contact region during gait: implications for shoe wear and slipping risk............................................................................................. 94

Category: Methods Megan Black, Clara Bourrelly Spike decontamination in local field potential signals from the primate superior colliculus............................................................................................................ 16 Yuxuan Hu, Rene R. Canady Computer vision methods for tool guidance in a finger-mounted device for the blind...................................................................................................................... 30 Victoria Turchick, Yulia Yatsenko Analysis of stride segmentation methods to identify heel strike...................................... 98

Category: Review Hilda Jafarah, Isabelle Chickanosky, Alexis Nolfi Immunomodulators and macrophages in endometriosis pathogenesis and progression................................................................................................................ 35 Carol Vinski, Tyler Meder Assessment of luminal fillers in nerve guidance conduits for promoting regeneration following peripheral nerve injury ................................................................................... 102

151 Benedum Hall 3700 O’Hara Street Pittsburgh, PA 15261 412-624-9800 engineering.pitt.edu

The University of Pittsburgh is an affirmative action, equal opportunity institution. Published in cooperation with the Office of University Communications. 113032-0321

Articles inside


pages 114-115

u Neural Network-based approximation of model predictive control applied to a flexible shaft servomechanism

pages 107-110

Department of Bioengineering, McGowan Institute for Regenerative Medicine, Renerva, LLC

pages 102-106

u Finite element analysis of stents under radial compression boundary conditions with different material properties

pages 111-113

Analysis of stride segmentation methods to identify heel strike

pages 98-101

Joseph Sukinik, Rosh Bharthi, Sarah Hemler, Kurt Beschorner

pages 94-97

Human Movement and Balance Laboratory, Department of Bioengineering; Falls, Balance, and Injury Research Centre, Neuroscience Research Australia

pages 90-93

u Topological descriptor selection for a quantitative structure-activity relationship (QSAR) model to assess PAH mutagenicity

pages 81-84

Department of Bioengineering, Department of Electrical Engineering, Department of Mechanical Engineering, Innovation, Product Design, and Entrepreneurship Program

pages 85-89

Department of Chemical Engineering, Heart, Lung, Blood, and Vascular Medicine Institute Division of Pulmonary, Allergy and Critical Care Medicine

pages 76-80

u Demonstrating the antibiofouling property of the Clanger cicada wing with ANSYS Fluent simulations

pages 72-75

u Levator Ani muscle dimension changes with gestational and maternal age

pages 64-67

u Bioinformatic analysis of fibroblast-mediated therapy resistance in HER2+ breast cancer

pages 60-63

Department of Bioengineering, Department of Psychiatry, Department of Neurology, Physician Scientist Training Program, University of Pittsburgh School of Medicine

pages 55-59

u Fluid flow simulation of microphysiological knee joint-on-a-chip

pages 49-54

Department of Bioengineering, Division of Vascular Surgery, University of Pittsburgh Medical Center, Department of Surgery, Department of Cardiothoracic Surgery, and Department of Chemical and Petroleum Engineering, McGowan Institute for Regenerative Medicine, and Center for Vascular Remodeling and Regeneration

pages 44-48

Testing the compressive stiffness of endovascular devices

pages 40-43

Department of Bioengineering, Carnegie Mellon University, McGowan Institute of Regenerative Medicine

pages 35-39

Physical Metallurgy & Materials Design Laboratory, Department of Mechanical Engineering & Material Science

pages 25-29

Hardware acceleration of k-means clustering for satellite image compression

pages 20-24

Visualization and Image Analysis (VIA) Laboratory, Department of Bioengineering

pages 30-34

Spike decontamination in local field potential signals from the primate superior colliculus

pages 16-19

u Simulating the effect of different structures and materials on OLED extraction efficiency

pages 13-15

u Representations of population activity during sensorimotor transformation for visually guided eye movements

pages 7-12

Message from the Coeditors in Chief

page 5

A Message from the Associate Dean for Research

page 4
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.