I-Predict

Project page: GitHub

This project aims to develop multidisease classifiers for body CT scans using a weakly supervised deep learning approach. By automatically extracting disease labels from radiology reports, the study focused on three organ systems—lungs, liver, and kidneys—and created classifiers to detect common diseases within each system. The models showed high accuracy in label extraction (91%-99%) and achieved robust performance across various disease classifications, with AUC values ranging from 0.62 to 0.97. This approach demonstrates potential for enhancing automated diagnostics across multiple organ systems in clinical radiology.