ProtoRadNet: An Interpretable Deep Learning Network for Robust Radiology Image Classification
The Research group of Dr. Tanmay Basu in the Department of Data Science and Engineering developed ProtoRadNet, an Artificial Intelligence (AI) based solution that helps make medical image diagnosis more transparent and trustworthy. Unlike many existing AI models that behave like “black boxes,” ProtoRadNet is designed to explain why it makes a particular decision, which is crucial for clinical decision making. It works on MRI, CT scans, X-rays and other modalities by identifying small, meaningful image regions - called prototypes - that represent important visual patterns linked to specific diseases. When it predicts a diagnosis, it highlights relevant regions that influenced its decision, allowing doctors to visually verify and interpret the result. It does not require detailed manual annotations from experts, making it practical for real-world hospital settings where such annotations are often unavailable. It has been successfully tested on multiple large standard radiology imaging datasets covering brain tumors, lung diseases, chest X-rays, and Alzheimer’s disease with high accuracy and strong interpretability. Outcomes of ProtoRadNet have been published in Artificial Intelligence, a premier journal with an impact factor of 6.2. Further plan is to deploy this solution in diagnostic labs for obtaining regulatory approval to use it in real life settings. For more details, kindly visit https://doi.org/10.1016/j.artmed.2025.103324
