BL MedTech Society
MEDICAL FUTURIST ISSUE 1 MAY 2020
Feature: The Future of Artificial Intelligence in Healthcare
Opinion: 3D Printing: The Past, Present and the Future
Contents feature 4
The Future of Artificial Intelligence in Healthcare
Wearable Doctors: Latest Developments in the Field of Portable Technology
Delivering Healthcare from Drone to Doorstep
Rise of the Robotic Surgical Revolution
Have We Solved the Blackbox Problem of Machine Learning?
3D Printing: The Past, Present and the Future
Katerina Spranger of Oxford Heartbeat
Bruce Hellman of uMotif
Pearse Keane on AI in Opthalmology
Lise Pape of Walk with Path
16 Editorial Team:
Matthieu Komorowski on Reinforcement Learning
Stephan Chee Manfredi D'Afflitto
Editor's Note Since time immemorial, the field of medicine and healthcare has been steadily advancing on an inexorable tide of innovation, driven by altruism and limited only by one’s own imagination. Perhaps it might be said that as an innovator, one only needs to ask one question – “why not?”. It does not take much for any person to realise that whilst it is easy to find themselves marvelling at the ingenuity behind every medical tool and procedure, there is still a plethora of unmet needs within every healthcare system. This is what innovation is all about – finding unique solutions for specific problems. It is about thinking beyond the bounds of current medical practice into what it could be in future where the problems of today no longer plague the societies of tomorrow. It is for this reason that the idea for this magazine was conceived. With this magazine, we hope to challenge current ideas within medicine about what is possible and what isn’t by exploring various ideas and developments within medtech. More importantly we hope that this would inspire our readers to think outside the curriculum in search of unique solutions to the problems they will encounter within healthcare. In this issue, there is a focus on artificial intelligence (AI) within healthcare as scientists and clinicians alike increasingly seek to optimise its potential, particularly that for diagnosis. However, is it really that simple and is that really all there is to it? From news on the latest medtech developments to interviews from experts in the field, we hope you enjoy this first issue.
THE FUTURE OF ARTIFICIAL INTELLIGENCE IN HEALTHCARE From Medical Education to the Practicing Doctor
Artificial intelligence (AI) was first described in 1956 by the computer WRITTEN BY
scientist John McCarthy in the first AI academic conference. AI is the term
utilised to describe a technology that is able to simulate human cognition,
intelligence and critical thinking. Since 1956, there has been extensive research in this area of computer science and its application in society. In the last years there has been a lot of fear regarding the replacement of various professions by machines, the majority of which utilise AI to some extent. Last year, Forbes Technology Council released a list of 13 jobs which will likely be automated by 2030, including customer service, delivery and banking services. The fear has nonetheless reached the medical sector with both patients and doctors worried about their medical consultations and jobs respectively. Should we be afraid of the integration of AI in medicine and how will this impact patients and doctors?
THE FUTURE OF ARTIFICIAL INTELLIGENCE IN HEALTHCARE
To the present day, AI is already used in the
of healthcare. In radiology, over the last decades
medical field with uses ranging from online
technology has improved the acquisition, storage
appointment scheduling and check-ins to drug
and reporting of radiological images. However,
dosing calculations and immunisation dates.
technology can still play a role in the future of this specialty mostly through the use of AI in diagnosing
"During a consultation, doctors spend approximately 53% of the time on their computer completing electronic healthcare records and ordering investigations." Within the consultation room, AI can revolutionise the current medical practice . The use of documentation technology such as dictation assistance and medical scribing technology allows the doctor to spend more time actually engaging with the patient. In this setting, AI is beneficial for the doctor, who spends less time on their computer and is more efficient, and for the patient
radiological images. AI has the ability to independently recognise complex patterns within images â€“ compared to a human radiologist, AI is able to provide a quantitatively rather than qualitatively assessment of radiological characteristics. Furthermore, the efficiency of AI has been portrayed by a retrospective study published this year by Kim et al. which utilised data from South Korea, the UK and the USA; AI was found to be able to consistently detect cancer accurately in around 90% of cases compared to lower and more variable detection rates (50-74%) amongst experienced radiologists. Apart from its use in the detection and diagnosis of pathology, the technology could also be used for monitoring patients. For example, for patients on oncological screening programmes, AI would be able to provide accurate information on any
who has more direct contact with the doctor.
changes to tumour characteristics over time. In the
The potential of this technology in the medical
increase in the number of imaging tests performed
field is immense with applications in every aspect
NHS, over the last decade there has been a yearly thus resulting in an incremental strain on radiologists.
Photo: Kim et al, 2020, The Lancet Digital Health
"In 2018, the NHS reported an 18.3% increase in imaging activity from the previous 5 years."
THE FUTURE OF ARTIFICIAL INTELLIGENCE IN HEALTHCARE
In this setting, the integration of AI may benefit
The Boston-based startup CompanionMX has also
both the radiologist, who will have an extra tool to
incorporated AI to develop an application that
help them with image interpretation in a more
analyses audio recordings of patients on treatment
time-efficient manner, and the patient, who may
for mental conditions such as depression and is
be able to receive more accurate results faster. The
able to deduce and monitor the patient’s mental
major question remains whether AI will eventually
health. The use of AI in chatbots is not restricted to
replace radiologists, but for the time being it
mental health, as observed by the English startup
appears that rather than replacing it, it will
Babylon which has been adopted by the NHS and
redefine the specialty. Looking at other specialties,
valued more than 2 billion dollars. The Babylon
AI has also been shown to be more effective than
chatbot helps patients to decide the best type of
dermatologists at diagnosing and classifying
care for their particular symptoms. The platform
malignant skin lesions.
does not provide a definite diagnosis; thus, it does not replace the usual role of the doctor. However,
With an ever-changing medical field due to the use
by providing the appropriate advice, it helps the
of novel technology and AI, it is essential that the
patient in getting access to the right type of care
medical education system integrates this new
and saves time and money for the NHS by
aspect of medicine. AI can be utilised as an
providing automated screening without the need
educational tool for medical students – for
of a doctor.
instance, the DXplain platform allows students to fill in the patient’s symptoms and investigations and it provides them with a range of potential differential diagnoses, with the available scientific evidence behind them, and the recommended actions to take. More importantly, it will be crucial to integrate the concepts of AI within the medical curriculum in order to prepare and educate future clinicians on this technology and its applications in healthcare. The growing prevalence of mental illness is a call for action to improve the accessibility and
Photo: The Drum
efficiency of mental health services - AI could lead
The potential application of AI in healthcare is vast
to a revolution in both the diagnosis and treatment
with several specialties that have already began to
of mental illnesses. Through the analysis of speech
integrate it in daily practice – the major benefits of
and written words, AI may be able to predict
AI are still to be explored. For the time being, it is
depression up to 3 months prior to the medical
difficult to predict the role of AI in healthcare in 20
diagnosis. There are already several applications
years and what changes this could bring for the
which have incorporated AI for this purpose such
patient along with medical training and practice.
as Ginger, which analyses the words a person uses and is able to provide the most appropriate counselling for the patient’s needs.
WEARABLE DOCTORS: LATEST DEVELOPMENTS IN THE FIELD OF PORTABLE TECHNOLOGY
The Elvie Pump. Photo: Business Insider
BY KAROLINA WIECZOREK
Over the last few years we have seen many technological developments that are aimed at improving our health, however only some of them turned out to be specifically successful. Amongst the biggest areas of interest for investors nowadays are women’s health and sports science. These two hold a lot of potential for development as many life-changing solutions can easily be implemented with the right technological approach.
Impact on women’s health – the Elvie Pump In recent years, the industry of ‘FemTech’ - technology directed at women - has seen many diagnostic devices and applications that are specifically directed to improve women’s health. It is estimated that the market potential of such inventions is going to reach $50 billion by 2025 making it one of the significant areas of technological development. Most smartwatches offer apps synchronised with mobile devices which offer tracking of information regarding their reproductive health and family planning. Rising popularity of these interfaces leads Apple to introduce menstrual cycle tracking into their latest update for all devices.
Photo: News Beezer
One of the most important wearable devices is The Elvie Pump, a discrete breast pump that fits in women’s bras. It includes sensors that automatically stop pumping when the container is full and is connected with a phone app which allows for tracking the process. All the parts that require cleaning are dishwasher safe and make it easy for new mothers to continue with their daily routines while pumping milk. Previously, women had to rely on hands-on devices that were much less discreet and did not allow for performing other activities. A mother using the device for a long time said that ‘it gave me some of my freedom and a lot of my dignity back’. The design is, however, not flawless with women complaining about flashing lights that are visible through clothing and sound which is still noticeable in quiet spaces. Nevertheless, it is a big step towards empowerment of mothers in the working environment and an area which has a lot of potential for development.
The ECG in Apple Watch Series 4 Perhaps the most popular invention nowadays are smartwatches, particularly Apple Watches Series 4, since they were the ones to introduce a variety of health checks including: measuring the heart beat and taking an electrocardiogram (ECG), which both can be very helpful in diagnosing heart problems.
This apple technology uses an electrical heart sensor (single-lead ECG, similar to the ones doctors use) that collects results in your phone and can detect various life-threatening irregularities such as atrial fibrillation (irregular heartbeat) or very low and high heart rates. This year Apple published numerous stories about people whose lives were saved by alerts from their smartwatches. Dr Michael Spehr, a columnist with the German FAZ newspaper, explained that an Apple watch accurately diagnosed a patient’s previously undiagnosed atrial fibrillation by constantly reporting irregular heartbeat.The patient later wrote: “It’s true, the Watch contributes to prolonging my life”. However, this technology still shows rooms for improvement as it cannot detect heart attacks or other heart-related conditions. Another crucial contributing factor is the demographics of potential users of this technology as most of them are ages 25 to 34. In this age group atrial fibrillations remain a rather rare (1 - 2%) heart problem, so only by introducing this technology to people aged over 65 is there more potential for diagnosis.
Current Health Another interesting technology that has been developed in the USA is Current Health – the first AI Medical Monitoring Wearable Device that was approved by the FDA to monitor vital signs of patients home. The device measures pulse, respiration, oxygen saturation, temperature and mobility of patients and communicates with doctors via machine learning algorithms to detect any life-threatening situations. The wearable band is supplemented by a tablet that has education resources, medication reminders that aid especially older patients. This technology holds potential for preventing avoidable deaths and reduces the workload of nurses, carers and the primary care teams.
Photo: Los Angeles Times
Scientists at UC Berkley reported a promising technology in the field of sport science: wearable sweat-sensor patches which can be used by athletes to monitor electrolyte and fluid levels in the body. These analyze perspiration under the patch, and are therefore also waterproof and can be used by professional swimmers. The device is connected with the users’ smartphone and processes information about sweat biochemical markers like pH as well as body temperature. It is particularly important for high-endurance athletes who tend to become very dehydrated during training and this technology is able to calculate the amount of water they should be consuming posttraining to become rehydrated.
The field of wearable technology is a rapidly expanding industry and more innovations are said to appear. For example, the next generation of Apple Watches are rumored to feature glucometers for diabetic patients and L’Oréal is designing UVA and UVB light-detecting sensors to measure radiation levels and make people aware of dangerous levels of sun exposure. There are many areas for development that hold a lot of potential for future investors in the medical field of wearable technology.
DELIVERING HEALTHCARE FROM DRONE TO DOORSTEP byÂ Stephan Chee
Reports of unmanned drones delivering lethal
hospitals without valuable blood supplies which are
payloads to eliminate people and buildings is
in themselves logistically challenging given their
nothing new and much less spectacular now
short shelf-lives. Today, about 40 percent of blood
compared to when it was first employed as a
products delivered to hospitals in Rwanda is done
novel military weapon. In recent years however,
entirely by these drones.
what has conventionally been used to take lives is now being considered through the lens of altruism by the healthcare industry.
The entire delivery process starts with an order placed via email, text message or a phone call from the hospital to an operator at the distribution
In 2016, Jonathan Ledgardâ€™s Zipline launched the first
centre. The operator then selects the correct blood
ever fully autonomous drone delivery network for
product from the various types stored at the centre
blood products in Rwanda. Traditionally, healthcare
and packages it into the loading bay of a drone. Its
products are transported by motor vehicles along the
wings are fitted, a new battery placed in, and the
many dirt roads in the country, which are vulnerable
drone positioned onto a catapult before being
to the heavy rains during the monsoon seasons that
launched into the air where it then flies along a
Rwanda experiences biannually. This can leave
pre-set route to its destination. Once directly over
its drop-off point, it releases its package which gently descends to the ground with the aid of a parachute and picked up by a hospital staff member. The drone then returns to the distribution centre where it is disassembled and stored until needed for another delivery. Following on with the projectâ€™s success in Rwanda, Zipline has since expanded its operations to Ghana and in 2019 trialled the delivery of vaccines, blood products and other life-saving medications to local hospitals by these drones. The company also continues to refine the droneâ€™s design, with its latest model capable of delivering payloads of 1.75kg over distances up to 161km (100 miles). Meanwhile in Tanzania, DHL is working with the local authorities to utilize the vertical take-off and landing (VTOL) DHL Parcelcopter 4.0 to deliver medical supplies to hard-to-reach places. In 2018, the drone was successfully used to deliver medical supplies to Ukerewe District Hospital (UDH) in Nansio from the regional capital of Mwanza as part of a trial.
40 percent of blood products delivered to hospitals in Rwanda is done entirely
severely impedes the prompt delivery of emergency medical supplies,
temperature-sensitive products like vaccines and test samples. In
contrast, the drone consistently traversed between the capital and UDH in 40 minutes. In more developed countries like the US, the logistics company UPS has launched a new revenue drone delivery system last year in North Carolina. The programme utilises the Matternet M2 Quadcopters to deliver up to 5lbs (over 2.3kg) of medical samples at a local hospital. Earlier, the company had also partnered with CyPhy Works to successfully deliver an asthma inhaler to a child on an island inaccessible by land vehicles as part of a trial in 2016.
could cut the
In Europe, a drone is currently being developed by Delft University of
Technology to rapidly deliver an automatic external defibrillator (AED) where needed. Along with its in-built AED, the prototype contains a
taken to deliver
camera, microphone and speakers through which an operator can
an AED to a
communicate instructions to by-standers at the scene. Alec Momont, who leads the project, predicts that the drone could cut the average time taken to deliver an AED to a sufferer from 10 minutes to just 1 minute. It seems therefore that drone technology is set to change the entire face of medical logistics around the world in the coming decades. With some companies like Boeing, Sabrewing and Elroy currently developing drones capable of delivering large payloads over long distances, there is huge potential for remedying acute and chronic shortages in medical supplies worldwide, and closing the health inequality gap that much more.
sufferer from 10 minutes to just 1 minute"
News: Delivering Healthcare from Drone to Doorstep
Albeit receiving regular supplies via its overland route, the 6 hour journey
RISE OF THE ROBOTIC SURGICAL REVOLUTION By Catrin Sohrabi
There has been a rise in the prevalence of robotics and automation across all industrial sectors, and its introduction into that of the field of healthcare is not new. Indeed, healthcare has moved beyond the traditional format of simple treatment and surgical procedures, and has begun to experiment more and more with the arsenal of technological tools that have become available to doctors and surgeons alike over the past decade. As it stands, robotic surgery is slowly becoming increasingly common within the hospital setting, particularly within the fields of thoracic and cardiovascular surgery (e.g. as part of mitral valve repair, atrial septal defect closure, and multivessel minimally invasive coronary artery bypass grafting) as
well as general surgery. Surgical robotics are used due to their high level of precision and ability to execute fine and delicate manoeuvres that would otherwise not be possible using the dexterity of the human hand alone. Robotics has transformed the way and speed at which common surgical procedures are performed. Indeed, there are many benefits to be reaped from using these highly advanced technological tools. Regarding benefits to the patient, most robotics-based procedures are minimally invasive. This thereby enables smaller, more discrete incisions to be made, avoiding any chances of inadvertent punctures and providing better outcomes for the patients. Patients often also benefit from Photo: DesignNews
Opinion: Rise of the Robotic Surgical Revolution
faster recovery, enabling them to resume normal work more quickly, and to spend more quality time with their family and friends. Less pain and risk of major blood loss is also endured by the patient, reducing their dependency on unnecessary painkillers. Moreover, robotic surgery carries with it a lower risk of post-op infections compared to traditional surgery. In addition to the benefits to be reaped by the patient, there are many favourable aspects of using robotics from the perspective of the doctor and surgical team. For example, the use of robotics has been associated with better surgical performance and confidence. Surgeons may feel more comfortable utilising a robotics-based approach since such technologies provide precision and accuracy through the use of 3D cameras and hands that can be expertly and calmly controlled. Indeed, studies evaluating surgeon and surgical team perspectives on the implementation of robotics in the operating room have reported largely positive feedback, with surgeons having praised the ergonomics. Moreover, the imaging available also provides better depth perception. In addition, the
ability to use robotic instrumentation across a broad range of surgical specialities is a geatly attractive feature of its users. For example, surgical robotics may be used across a variety of surgical subspecialities such as gynaecology, colorectal and urological surgeries, ENT, orthopaedics, and neurosurgery, to facilitate the execution of high levels of surgical precision. Moreover, robotics provides a solution to the traditionally lengthy surgical procedure, which may induce strain, shoulder, and back pain in the unfortunate surgical trainee. Thus, robotics enables the procedure itself to be less physically demanding, since the entire operation can be performed from the comfort of a seat or console located elsewhere within the operating theatre. However, various limitations exist. For example, surgical robots bring with them a high cost of installation, are costly to maintain (e.g. the da Vinci Robot Surgical System), and their operation requires additional training which may also contribute to the overall expense. Moreover, technical complications and risks may include mechanical
Photo: Healthcare Market Experts
The Da Vinci Robot Surgical System
failures and malfunctions (albeit rare), sparks, electric arcing and burns which may inadvertently cause tissue damage, and the theoretical risk of nerve palsies due to the compression of vital nerves by the robotic arm. As a whole, there are various benefits to using robotics in the wonderful and delightful field of surgery. These attractive features have evidently led to their wide-spread adoption not only on a subspecialty scale, but on one that is also worldwide. In the future, robotics may be used for remote applications such as humanitarian relief. Alongside surgery, robotics may also one day provide laboratory assistance to facilitate the location and drawing of blood with less pain and anxiety for the patient, may carry bed linens and meals from floor to floor, may help to disinfect patient rooms and operating suites (thereby reducing the risk for infection transmission), and may even be utilised in the dispensing of daily medication. It is likely that robotics will prove a game changer to the way in which surgery and healthcare as a whole is practised.
Photo: Verywell Health
Have we solved the blackbox problem of machine learning? By Alex Deighton
Over the past decade the rise of artificial intelligence has been nothing short of meteoric. Healthcare, industry and big data companies have embraced this emerging technology with open arms and it is becoming an ever more prevalent part of our lives. Realising the full potential of this technology, however, has been fraught with difficulty. Media sensationalism and extravagant promises of a machine learning revolution have brought about repeated cycles of hype and disappointment, leading to the so-called ‘AI winter’ and threatened the collapse of the AI industry before it had even begun.
Fortunately for us, the AI winter is finally over and the AI spring has dawned. Publications on the topic of machine learning have sky-rocketed in recent years, whilst the global AI industry is expected to grow from a value of $4 billion in 2016 to an impressive $169 billion in 2025. This sudden boom has driven the creation of thousands of AI startups and pushed bigger corporations to branch out, with Microsoft, Google and IBM all opening healthcare divisions within the last couple of years. In doing so, this renewed interest has cast the spotlight on problems that have been frustrating AI researchers for decades.
Opinion: Have We Solved the Blackbox Problem of Machine Learning
"AI can process vast quantities of data, predict behaviours and preempt health conditions, but it cannot tell us how, and this lack of transparency causes big problems when things go wrong."
One such challenge facing researchers is the black box problem of machine learning, the question of how rather than what. Being so used to feeding information into one side of a machine and getting results out the other, we often pay little attention to how the machine actually reached this solution and the same goes for artificial intelligence. Here this socalled ‘black box’ between inputs and outputs obscures the decision-making process and leaves researchers in the dark, preventing us from understanding how our algorithms think and leaving plenty of room for bias. Granted, this may be of little significance when predicting shopping trends or patterns of twitter use, but when AI is being used to diagnose cancers, or convict criminals, the question of how becomes rather more important. AI can process vast quantities of data, predict behaviours and preempt health conditions, but it cannot tell us how, and this lack of transparency causes big problems when things go wrong. To put this problem into perspective, take a world without this black box for a minute. A machine learning programme mistakenly diagnoses a healthy individual with cancer, but why? Without the black box, we can look at how the programme made this decision and what went wrong. Perhaps it was too reliant on external guidelines, perhaps it detected an artefact on X-ray as a malignant tumour. We can understand this error and make adjustments, but as it stands, this isn’t the case. The black box muddies the water. AI programmes are brilliant, but secretive, and this is an issue that needs solving before we can ever trust AI to make the big decisions we can’t.
Thankfully for us, some of the best and brightest have risen to the task. Google’s AI division, DeepMind, published an article early last year presenting their algorithms capable of diagnosing over 40 ophthalmic diseases from optical coherence tomography, or OTC scans. Here these algorithms were shown to match the world’s leading specialists and even make brand new connections, determining both age and gender from nothing more than retinal images. The results of the study were nothing if not impressive, but what was even more revolutionary was how Deep Mind addressed the black box problem which has plagued AI for decades. DeepMind’s new system, instead of training a single neural network, used two, compartmentalising the diagnostic process into two separate stages. Stage 1, segmentation, would identify visual structures, creating a tissue map, with labels for anatomy, pathology, and artefacts. Stage 2, classification, would then analyse these labels and diagnose or refer as appropriate. This process, reflecting a real world diagnostic pathway, allowed clinicians to inspect the AIs decision making, and is key to more safely integrating neural networks into current clinical practice. This way, researchers were given an explanation, rather than a simple diagnosis. Crucially, work didn’t stop there. This small step in making AI more transparent has been followed up with the creation of Google’s new facility, Explainable AI. Explainable AI aims to inform clients about the performance and potential shortcomings of their machine learning models, making the decision-making of artificial intelligence less mysterious, and a bit more trustworthy. Despite this, Google themselves
recognise that there is still a long way to go before artificial intelligence becomes fully transparent and unfortunately, despite these developments, the black box problem is likely to persist. Why? Because there are downsides to transparency too. If the world can figure out how your AI works, they can figure out how to make it without you and this can be expensive when youâ€™ve poured millions of dollars into developing your system. Meanwhile, other big corporations made billions at a time when AI was unregulated, creating and selling systems that are both biased and unethical. Extricating biased black box AI from the world would probably push a number of companies out of business and result in hundreds of billions of dollars lost. As a result, as AI algorithms become increasingly transparent, many of these big businesses are lawyering up, invoking intellectual property laws to avoid giving details about how their algorithms arrive at important decisions. Sadly, we may be trading a technical black box for a legal one, and somehow this seems even more wrong.
3D PRINTING: THE PAST, PRESENT AND THE FUTURE BY DHILLON HIRANI
3D printing’s popularity has skyrocketed in
3D printing has a role to play in several
recent times amongst technology
industries such as dentistry, architecture,
enthusiasts. But why? Initially we were
aviation and lately in healthcare. There is
drawn in by its inherent coolness, but we are
currently 18% growth in the 3D printed medical
now beginning to see past that, realising just
device market per year and it is estimated that
how useful it can be. But what is it?
this market will reach a value of $3.5 billion by
3D printing is essentially an additive
implants and prosthetics which have been in
manufacturing process that involves placing
use since the early 2000s. In 2016 for the first
down thin layers of a material successively in
time 3D printing was used in the NHS when it
order to recreate a digital model in three
was combined with robotic surgery. It was
dimensions. It was once just a fantasy but with
pioneered at Guy’s and St Thomas’ where a
the technological revolution we are in, it has
cancerous prostate was modelled using MRI
become a reality with the potential to open
and then printed to help plan an intricate
many opportunities in many fields. Its
surgical removal. Surgeons have stated that
beginnings can be traced back to Charles Hull
‘having a 3D print of the anatomy gives an
in 1986 who patented stereolithography where
increased understanding of spatial orientation’.
acrylic polymers were layered and then
Studies have shown this to reduce operating
toughened with UV light – his first print being a
time, complications and need for
small eye wash cup.
reinterventions. To help surgery further, tools
2025. It began with the production of dental
such as screws and plates can also be custom printed to improve the precision of an implant.
OPINION: 3D PRINTING: THE PAST, PRESENT AND THE FUTURE
These models can then also be used for the benefits of education to understand anatomy that can be difficult to visualise. As well as helping students, printing models can help patients understand and make informed decisions about surgery as opposed to having to interpret 2D images on CT or MRI scans
"So currently, 3D printing has made significant strides in creating prosthetics, implants, anatomical models and dental apparatus. But what is the future? Why are we so excited about this technology?"
The prime industry for 3D printing to crack is tissue engineering. Currently we isolate stem Photo: KLS Martin Group
cells, grow them in labs and then place them upon scaffolds to develop into tissue. This is a
The largest healthcare industry currently
haphazard and slow process. 3D printing has
benefitting from 3D printing is dental care.
the potential to accurately create a tissue
There is a diverse use of 3D printing here
structure via inkjet-based bioprinting in which
already – spreading from its use in producing
living cells or biomaterials can be ‘dropped’ in
transparent aligners to crowns and models of
layers to form a tissue. Stem cells are first
teeth for surgery. The prints are custom, sterile,
differentiated into the organ cell type of choice
complex and low cost. This is because it can be
and then bioink ‘reservoirs’ can be created to
run on a small scale to produce bespoke small
then print the tissue of choice. This concept is
implants and prosthetics using various
still being researched and so far we have been
materials. 3D printing is resulting in faster,
able to create prototypes of a knee menisci,
cheaper treatments. Practices can design a
heart valve and an artificial ear. Printing tissue
model and print it in the same room. In fact,
is a stepping-stone to what would be the
crowns can now be replaced in less than an
ultimate application – using 3D printers to
hour as a result of 3D printing compared to the
create complex organs such as the heart or
usual several weeks. Furthermore, these
liver. These are larger and require
models and designs can be shared worldwide
vascularisation and nervous supply which
and so people around the world can instantly
complicate it more, although institutions are
produce the same model from their own
starting to print functional capillaries which is a
sign of promise.
3D PRINTING: THE PAST, PRESENT AND THE FUTURE
"IT IS ESTIMATED THAT 50% OF MEDICATIONS AREN’T TAKEN AS DIRECTED AND POLYPHARMACY IS A POTENTIAL CAUSE".
We have begun to print synthetic tissue for patients with burns and it is believed that bioprinting could potentially produce functional organs within the next 10-20 years. This would be a huge achievement in the field of medicine. Transplant waiting lists are ever increasing and the ability to bioprint organs can reduce that waiting list and provide many people with potentially lifesaving treatment. Along with its use in transplantation – 3D printed tissues have a role in research as drugs can be tested on these organs during development to attest their safety and efficacy. This can be further extended to using a patient’s stem cells, printing tissue using those stem cells and then testing the efficacy of a drug on that specific personal printed tissue. This brings a different dimension to the idea of personalised medicine. Another big step in 3D printing is its potential use in pharmacy. Currently, we are limited by how we can divide dosages – however with 3D printing, pharmacists could potentially print out and dispense a specific dosage of a medication to provide optimum treatment.
Polypharmacy is also becoming an issue as patients are having to take more and more medicines. 3D printing can be used to combine medications to print a single tablet to improve compliance and ultimately improve the management of chronic conditions. However, despite all this excitement about 3D printing we must be realistic about timescales. The goals of having 3D printed tissue and organs, or medications printed in the pharmacy are still just goals and we are still some way away from achieving this. We are not just limited by the technology but also factors such as healthcare regulations, copyright legislations, and even ethics. However, we are on that journey, and the 3D future looks bright.
INTERVIEW WITH KATERINA SPRANGER, CEO & FOUNDER OF OXFORD HEARTBEAT
In a few words, could you
Katerina is a Royal Academy
describe the technology behind
PreSize™ enables clinicians to better
of Engineering Fellow. She
Oxford Heartbeat and how it works?
visualise how a stent might behave
holds a PhD in Biomedical Engineering
When it comes to minimally invasive,
University of Oxford and has
high-risk surgeries, particularly those
Laboratory in Paris. Named a “founder striving to change
involving the insertion of stents into blood vessels in the brain, there is an astonishing
involved. Primarily, it’s always been
the world” by Forbes, and
“Young Innovator of the Year
anatomy before surgery. It’s equally
Conference Breakthroughs and Society.
Walls Future Science
tricky to decide exactly where the stent should be placed, because even the most minor differences in device dimensions can have a huge impact
when deployed and choose the best one for a particular patient case, bringing
What do you believe are the benefits for surgeons of using Oxford Heartbeat’s software? The
surgeons are able to select the best stent for each patient, reducing the propensity for complications during and after the procedure. PreSize™
on the patient.
We were perturbed to find out that
stents available on the market by
20% of patients who undergo brain surgeries
complications arising from ill-fitting stents, among other reasons. Given that each stent costs around £10,000, a lot of waste is incurred when
presenting them with their digital equivalents ahead of the surgery. Secondly, PreSize™ facilitates safer surgeries by enabling surgeons to rehearse
surgeries aren’t successful.
making them better prepared and
These discoveries drove us to develop
surgery, it also helps to reduce the
PreSize™ Neurovascular, a medical device software that makes brain
particular stent in an individual
intra-operative radiation exposure under which these procedures are being performed.
Finally, PreSize™ has been shown to
For a long time now, I have been
simulate stent deployment with a
fascinated and awed by the most
crucial moments of surgical decision-
making. The realisation that I could
have emphasised that knowing the
apply my skills to improve these
tool is highly accurate and safe really
high-risk procedures, and make a
helps to reduce the uncertainties
difference to the lives of patients in
and risks involved in their decision
doing so, really changed my life. I
making processes. We want to play
have continued to find inspiration
our part in making difficult medical
and meaning in this idea as Oxford
procedures less anxiety-inducing for
everyone involved. I also encountered and tackled many
In your view, aside from minimally
matters for the first time: hiring,
invasive brain surgery, in what
other types of surgery do you think
regulatory approval, setting up a
Oxford Heartbeat could be utilised
clinical trial, and the list goes on. I’m
looking forward to many more “firsts”
as we continue to grow and explore
We’re really excited to be developing
new ways to make complex surgeries
analogous to PreSize™, which can be
surgical procedures on other parts of the body. Our aim is to support complex surgeries in any way we can.
Above all, I am grateful to work with my
appreciate all the support we have received from the government, the NHS and our network of friends in
What was your experience like of founding a start-up company in the
medical field? It has been the most exciting journey so far! I have stretched myself in every possible direction and am humbled to have learnt a lot along the way. My background in computer science and engineering showed me the power of technology to challenge the status quo. My academic
experiences have also helped me to cultivate a conviction that you are able to make the world better, when you see an opportunity and work hard towards it.
INTERVIEW WITH BRUCE HELLMAN
CEO & CO-FOUNDER OF UMOTIF
Bruce is the CEO and co-founder of uMotif, a platform which focuses on collecting patient data for clinical trials. In his early career, Bruce worked on clinical trials at British Biotech and he subsequently joined the UK Civil Service Fast Stream.
In a few words, how did you come up with the idea
is highly valuable for them. Once you have an idea
for something that delivers value you need a great
Back in 2012 my co-founder and I met two people with Parkinson's disease. In speaking with those patients and understanding the challenges they faced in managing their condition, we realised that our simple idea for health tracking could potentially
it to customers and then deliver it excellently. The team is all-important, as it's often the case that you'll need to refine or adapt your offering, and a good team will be able to pivot with those changes.
help patients capture more data to support their
Is the COVID-19 pandemic fuelling more
conversations around virtual trials?
Can you briefly explain how uMotif works?
Yes, COVID-19 is creating a real sense of urgency to
Today, uMotif is a software platform used by
adopt virtual trials, both to help with vital research
pharmaceutical, academic, and health researchers to help gain new insights. Patients remain at the core of our approach and we focus on providing people with an engaging experience to capture their data. The data they capture is sent securely to the cloud,
into a vaccine or cure, as well as helping other research to continue. We are seeing delays to clinical trials due to sites closing and the social distancing measures currently in place. If some of these trials had a virtual methodology they could potentially be
where it is accessed by specific researchers for each
running without disruption.
For COVID-19 trials, virtual approaches can help
completing daily diaries, tracking symptoms, and
connecting wearable devices the data captured by
symptomatology and impacts of the virus. Weâ€™ll soon
patients helps researchers understand the efficacy of
be supporting a number of studies in this area and
a new drug, the impact of an existing drug on new
hoping to make an impact to bring the crisis to an
therapy areas, or understand how a disease can
progress. As a software platform we support multiple projects across the globe and in any therapeutic area. In your experience, what do you think are the most important components of a successful start-up? It's essential to focus on delivering value to a specific target market, who need the service/good/product the start-up offers. In our case - we help researchers undertake essential research studies and trials, which
team to create the service (or product) and package
Do you think the current COVID-19 pandemic will cause us to see more virtual trials in the future? Absolutely. The fact that coronavirus is so contagious and has caused countries the world over to close business
distancing, demonstrates the need for more virtual and hybrid trials. The world will be a changed place post Covid-19 and clinical trials will be too.
INTERVIEW WITH PEARSE KEANE
OPTHALMOLOGIST, MOORFIELDS EYE HOSPITAL
In a few words, could you explain how optical coherence tomography (OCT) imaging works and how it is utilised? Optical coherence tomography (OCT) is a form of imaging which was first invented in 1991. It is a bit like ultrasound imaging but measures the reflections of light waves rather than the echoes of sound waves. It achieves this using interferometry. As the wavelength of light is so much less than the wavelength of sound it produces very high resolution images. The axial resolution of OCT in the retina is approximately 5 microns, an order of magnitude better than imaging modalities such as CT or MRI. Even better OCT scanning is quick and safe and easy to do. It can be done on a 5 year old or a 95 year old!
Following your work with Google DeepMind Health, how is artificial intelligence able to interpret OCT images? Working with DeepMind we trained a deep learning system to analyse retinal OCT scans. The AI system can first delineate any pathology in the retina. It then makes both a diagnosis and a triage decision on the pathology. For example, in a patient with age-related macular degeneration (the commonest cause of blindness in the UK), it might make
recommend “Urgent” referral.
What do you think are the potential benefits of using artificial intelligence to interpret OCT scans? We hope this system will allow patients with the most severe, sightthreatening retinal diseases to get seen by ophthalmologists with specialist expertise as soon as possible. By allowing earlier detection and thus earlier treatment, we think this system has the potential to
Dr Pearse Keane is a Consultant Ophthalmologist at Moorﬁelds Eye Hospital and an NIHR Clinician Scientist at the Institute of Ophthalmology, UCL. His research focuses on the application of new technologies, such as AI, virtual reality and advanced imaging, in the field of ophthalmology. His work in collaboration with Google Deepmind involved the use of machine learning to analyse optical coherence tomography (OCT) images.
Do you believe artificial intelligence could be applied to other aspects of ophthalmology? AI is already being applied to many other areas of ophthalmology, including
Ophthalmology is very much at the forefront of the AI revolution in healthcare and I think it can act as an exemplar for other medical specialties. I’m also proud that Moorfields Eye Hospital, the NHS, and indeed the UK as a whole can be central to this revolution.
WITH LISE PAPE
FOUNDER OF WALK WITH PATH LIMITED Lise is the Founder of Walk With Path which, since it was founded in 2015, has developed "Path Finder" - a visual cueing shoe attachment to aid patients with Parkinson's disease - and "Path Feel" - which is a haptic feedback shoe insole to help with balance.
In a few words, what are the major benefits of
Can you tell us more about the Path Feel
using Path Feel and Path Finder for patients with
technology and how you hope it could be used in
Vibratory feedback from Path Feel acts to augment
with Parkinson's who are struggling with their gait
muscle stimulation, leading to improved balance.
and are suffering from Freezing of Gait and
Use of a vibratory insole has already shown in a
Shuffling. The laser light attachment provides a
group of healthy elderly people to improve balance
visual cue, which triggers the brain to take the next
vibrational feedback to the bottom of the foot can
feedback to the soles of people with Parkinson's
who have balance issues and are at risk of falls.
neuropathic and stroke patients, among others.
The vibration applied to the feet of Parkinson's
With Path Feel Insole we aim to help several
disease patients has been found to be effective in
groups of vulnerable individuals with mobility,
improving gait patterns, and significantly improves
balance and gait issues on a large scale and for the
product to become a go-to tool in preventing PD from getting worse (by staying active).
How does the use of laser lights help to improve mobility and gait?
What do you think are the main applications of
In Parkinson’s disease, sensory cueing such as
design engineering in the medical field?
visual and auditory has been long proven to
Medical devices tend to be heavily engineered, to
improve gait (Lebold and Almeida, 2011). Two main
achieve the desired function, however, sometimes I
mechanisms have been suggested to underlie this
think we overlook that the persons who will be
phenomenon. It has been demonstrated that in
using them are ordinary people who also care
Parkinson’s patients, visual dependence is used to
about what they may look like or they care about
the desirability of a product in their home. I think
feedback. At the same time, attentional processing
that design engineering has a huge role to play in
is also used to alleviate automaticity in walking. It is
personalising products more and making them
desirable - to everyone.
dependence play a role in gait control (Azulay et al., 2006). Projection of a visual cue, through Path Finder, may allow the patient to by-pass deficits in their internal cueing system by augmenting both sensory and attentional abilities.
In your opinion, what are the major advantages for the NHS of introducing Path Feel and Path Finder? NHS introducing Path Finder would provide massive advantages to not only the individual suffering from PD, but also the personal carer, the nurse and the health care system. With our currently 300 users (around the world) - if these were all in the UK, we would estimate saving the UK health care system about 100 falls which amounts to costs of £648,738 in admissions to the hospital plus a potential £831,554.37 in surgery. The impact of Path Finder reaches far beyond the quality of life for the individual as with most individuals with Parkinson’s
healthcare professional helping them on a daily basis. This means that we have effectively changed the daily lives of 1200 people. This impact can reduce the amount of people accessing health care services, thus reducing the burden on healthcare professionals, and allow informal care-givers more time to engage in their community. In addition to this, some people with Parkinson’s
workforce due to Path Finder Laser Shoes. The above figures are based on 300 people. If the NHS were to introduce Path Finder Laser Shoes to a much larger percentage of the eligible population, the savings for the Health Care System would subsequently be much larger: in the UK the total PD population is 145,000. 50% of these are suffering from FoG and 65% of that subgroup would be eligible. If we estimate that 10% of the eligible population would use Path Finder due to the NHS introduction, it would affect 4,713 people. The costs saved the health care system is on average £4.934,31 per person (in admission to the hospital and for surgery), which amounts to saved costs of over £23 million for the UK health care system. And that is only 10% of the eligible population - there is a potential to save £230 million should Path Finder Laser Shoes become a standard treatment for PD patients. For Path Feel, we have less data as the product is not yet on the market, but we would estimate cost savings, efficiency rising, and improved quality of life for citizens - leading to less strains on the NHS. Long term, preventative measures could also become more common place.
INTERVIEW WITH MATTHIEU KOMOROWSKI HONORARY INTENSIVE CARE PHYSICIAN, CHARING CROSS HOSPITAL
In a few words, could you briefly explain what reinforcement learning is? I would argue that “AI” is mostly a buzzword, but what we actually do is machine learning. All machine learning deals with the overall same overarching goal: to generate structured information or inference from data. For example, in supervised learning, we learn the mapping between input data (e.g. patient vital signs on admission) and output data (e.g. their 28-day mortality) to build a predictive model (e.g. a mortality prediction score). Lots of different models allow to represent with more or less accuracy the relationship between input and output data, for example logistic regression, decision trees, or neural networks. Reinforcement learning represents a very different type of machine learning algorithms. Here, the goal is to optimise decisions to maximise some kind of reward. For example, the ground-breaking 2015 Nature paper by Mnih and colleagues described a reinforcement learning agent capable of playing Atari games with no prior knowledge of any game rules. Starting with a random strategy (called a policy in reinforcement learning), the virtual agent was able learn by trial-and-error to maximise the score and therefore to play most games with supra-human level of performance. Reinforcement
Dr Matthieu Komorowski is an
learning works particularly well for complex, sequential
intensive care physician and a
and stochastic (partially random) decisions. We can
former research fellow at the
apply the same framework to healthcare, where the
European Space Agency. He has
objective of physicians is to improve outcomes of our
attained his PhD at Imperial College
patients (the reward), similarly often facing complex sequential decisions, for example deciding every time we review a patient with suspected infection whether the time has come to initiate antibiotic therapy.
Laboratory for Computational
intelligence could be used in the treatment of
involving secondary analysis of
Machine learning algorithms can be classified into supervised,
reinforcement learning. These 3 types of algorithms
Physiology at MIT where he collaborates on numerous projects
with a particular focus on sepsis. He is also a visiting scholar at the
Can you tell us more about how artificial
London for his research on the use of machine learning in critical care,
have different applications and objectives. In the
Following your work at the European Space
particular example of sepsis, supervised learning can
Agency, what do you think are the major
be used to predict sepsis before the onset, which in theory
administration. Unsupervised learning has been used to identify homogeneous subgroups of patients (also called phenotypes), which could have practical implications for their treatment or for the design of randomised controlled trials. Finally, reinforcement learning could be used in theory to optimise a wide range of sequential decisions, from weaning of mechanical ventilation, to dosing of insulin, heparin, intravenous fluids and vasopressors.
in space? This is another very exciting area for healthcare innovation. However, at the moment, the need to perform invasive medical procedures in space is very limited, because space flight is confined to low Earth orbit and pre-flight medical screening is extremely stringent. The contingency plan for any serious medical condition is to evacuate the International Space Station, which is possible at any moment using a Soyuz capsule. However, plans for exploration
In your opinion, what are the potential applications of artificial intelligence in critical care?
and/or colonisation of the Moon and Mars have been laid out and will push further the limits of human experience in austere environments. Science fiction
The potential applications of AI/machine learning in intensive
applications listed above, many AI models have been proposed to quantify severity of illness and predict mortality, help with the design of new antibiotics, predict antibiotic resistance directly from amplifying the genome of bacteria isolated in blood samples, or intelligent
weaning programme, to name a few. What I would like to emphasize is that the biggest hurdle is not the availability of algorithms, but the of
intelligent medical equipment capable of diagnosing diseases, performing autonomous surgery, provide empathy and psychological support, etc. The reality will most likely be much more stark and much closer to what remote crews experience in wilderness expeditions: having to deal with a huge range of medical
specifically talk about anaesthesia, the challenges in
limitations in performing anaesthesia
something that we could term “the implementation gap”. Efforts should focus on assessing the tools in the clinical setting for safety and efficacy and putting them in the hands of clinicians to “close the loop”. Ultimately, evidence needs to come from randomised controlled trials comparing standard of care to doctors augmented by AI tools. Only a handful of
space are 3-fold: physiological (negative effects of the exposure to the space environment on the body, in particular the cardiovascular system), technical (lack of equipment and consumables) and skills (lack of onboard expertise and capability for real time telemedical support in the case of Mars). For these reasons, we have tested a range of simplified protocols and techniques in different environments, including during parabolic flights and a Mars mission simulation in the desert of Utah.
algorithms made it this far. As the field of “medical AI” develops, some have suggested that a new medical specialty might arise of clinicians specialised in data science and machine learning, best suited to navigate a learning healthcare system embedded with decision support tools built from AI algorithms. In such a system, medical decisions would not only be based on empirical evidence and RCTs, but on AI models capable of harnessing the data contained in large medical datasets, to deliver a more personalised medicine. Academic institutions now recognise this vision, for example through UKRI funded centres for doctoral training (CDT) in AI for healthcare. Imperial College London is the host of the AI4HEALTH CDT.