Duke-NUS Quantitative Biology and Medicine Ph.D. candidate Yuan Han has developed a new method to improve the accuracy of AI when it is used to diagnose a collapsed lung, or pneumothorax, by limiting the area of a chest X-ray that AI uses when interpreting the image.
Working with radiologists to understand how they read such X-rays, Yuan created a simple filter or template that is applied over the X-ray to direct the AI’s analysis. With its sharpened focus, Yuan’s model reduced the false positive rate, thereby improving the performance across twelve benchmark scenarios, demonstrating the value of incorporating clinical knowledge into AI-powered diagnostic methods.
Under the mentorship of Associate Professor Liu Nan from the Center for Quantitative Medicine, who leads the Duke-NUS AI + Medical Sciences Initiative (DAISI), Yuan published the findings in the Journal of Biomedical Informatics.
“I was happy and a bit surprised. The initial decision came in just one month, granting us an acceptance in principle, which was unexpected given that this journal typically takes half a year on average to accept a paper. This swift decision indicates that the integration of clinical knowledge and machine learning captured the interest of both the editors and the reviewers,” said Yuan.
More information:
Han Yuan et al, Clinical domain knowledge-derived template improves post hoc AI explanations in pneumothorax classification, Journal of Biomedical Informatics (2024). DOI: 10.1016/j.jbi.2024.104673
Citation:
Training AI’s view on clinically relevant areas improves medical image analysis (2024, July 11)
retrieved 12 July 2024
from https://medicalxpress.com/news/2024-07-ai-view-clinically-relevant-areas.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.