6 minute read
DI Europe Summer 23
Alan Barclay
AI Improves the Detection of Lung Nodules on Chest X-Rays
Typically arising from previous lung infections, pulmonary nodules are very common abnormal growths that form on the lung. While many nodules may be benign, nodules can also represent potentially life-threatening cancers, so timely identification is vital.
Chest X-ray is the imaging modality that is the most widely used in the world, and is one of the common screening methods used for identifying lung nodules – and indeed it is one of the crucial tasks in chest X-rays.
However, detection by radiologists of pulmonary nodules on chest radiographs can be quite challenging, especially when radiologists are experiencing a high volume of cases. 1
Since the adoption of artificial intelligence (AI), several AI-based Computer-Aided Detection (CAD) systems have been developed and have been reported to substantially improve radiologists’ performance in the detection of lung nodules. However, so far these evaluation studies have all been retrospective, and have several major limitations. For example the retrospective data sets used in the evaluations are often arbitrarily selected to have a disease-enriched distribution. Additionally, the performance tests were typically conducted under conditions where readers could be more focussed and sensitive than is usually possible in real time clinical routine. Finally, proper integration with a PACS system, which is necessary for use in real work place situations, was lacking in the previous evaluation studies.
To address all these issues, a pioneering randomized control trial has been carried out to identify the actual effect that AI based soft-ware has in real clinical practice on the detection of lung nodules. The results have just been published. 2
The study used a commercially available AI-based CAD software (the Lunit INSIGHT CXR package version 2.0.2.0 from the Korean company Lunit). This was embedded into a commercial PACS system (M6; Infinitt Healthcare). The Lunit INSIGHT CXR software uses a deep learning-based algorithm developed for ten common radiological abnormalities in chest X-rays (pneumothorax, mediastinal widening, pneumoperitoneum, nodule/mass, consolidation, pleural effusion, linear atelectasis, fibrosis, calcification and cardiomegaly). This software has been extensively validated. 3
In the recent study to identify the actual effect that AI has in clinical practice, researchers from the department of radiology at Seoul National University Hospital in Korea, included 10,476 patients, who underwent chest X-rays at a health screening center between June 2020 and December 2021.
“As our trial was conducted with a pragmatic approach, almost all enrolled participants were included, which is a real clinical setting,” said corresponding author Dr. JM Goo. Patients completed a self-reported health questionnaire to identify baseline characteristics such as age, sex, smoking status and past history of lung cancer. Eleven percent of the patients were current or former smokers.
Solid nodules with diameters either larger than eight millimeters or subsolid nodules with a solid portion larger than six millimeters were identified as actionable, meaning that the nodule required follow-up, according to lung cancer screening criteria.
(A) Frontal chest radiograph shows a subtle nodular opacity (arrow) in the right middle lung zone.
(B) The lesion was detected by the AIbased computer-aided detection software, with an abnormality probability of 81.1 percent. The designated radiologist reported this chest radiograph as positive.
(C) Axial, noncontrast, low-dose chest CT scan shows a 1.1-cm solid nodule (arrow) in the right lower lobe.
Solid nodules with diameters either larger than eight millimeters or subsolid nodules with a solid portion larger than six millimeters were identified as actionable, meaning that the nodule required follow-up, according to lung cancer screening criteria.
(A) Frontal chest radiograph shows a small nodular opacity (arrow) in the left upper lung zone, which was missed by the designated reporting radiologist.
(B) Axial, noncontrast, low-dose chest CT scan shows a 9-mm solid nodule (arrow) in the left upper lobe. The nodule showed low metabolism at PET and decreased in size at follow-up CT. It was confirmed to be an inflammatory nodule
Results
Lung nodules were identified in two percent of the patients. Analysis showed that the detection rate for actionable lung nodules on chest X-rays was higher when aided by AI (0.59 percent) than without AI assistance (0.25 percent). There were no differences in the false-referral rates between the AI and non-AI interpreted groups. While older age and a history of lung cancer or tuberculosis were associated with positive reports, these and the other health characteristics did not have an impact on the efficacy of the AI system. This suggests that AI may work consistently across different populations, even for those with diseased or postoperative lungs.
Conclusions
The authors concluded that in their randomized controlled trial of 10,486 health checkup participants, comparing performance of radiologists aided by AI versus non-aided radiologists, the AI-assisted group improved the detection of actionable lung nodules on chest radiographs with a similar false-referral rate between the two groups.
“Our study provided strong evidence that AI could really help in interpreting chest radiography. This will contribute to identifying chest diseases, especially lung cancer, more effectively at an earlier stage,” Dr. Goo said.
The researchers plan to conduct a similar study using chest CT, which will also identify clinical outcomes and efficiency of workflow.
REFERENCES
1 Auffermann WF. AI Nodule Detection on Chest Radiographs Using Randomized Controlled Data: The Effect on Clinical Practice. Editorial in Radiology. 2023; 223186. https://doi.org/10.1148/radiol.223186
2 Nam JG et al. AI Improves Nodule Detection on Chest Radiographs in a Health Screening Population: A Randomized Controlled Trial Radiolog. 2023 Feb 7; 221894. https://doi.org/10.1148/radiol.221894
3 Nam JG et al. Development and validation of a deep learning algorithm detecting 10 common abnormalities on chest radiographs. Eur Respir J. 202; 57(5): 2003061. https://doi.org/10.1183/13993003.03061-2020
