Deep Learning-Based Computer-Aided Detection and Diagnosis System for Malignant Biliary Stricture (With Video)

Abstract

Simple Summary Malignant biliary stricture is difficult to detect early, and current tests often miss it or give uncertain answers. Doctors can pass a thin camera directly into the bile duct to inspect narrowed areas, but reading these images is challenging and varies from one specialist to another. We developed a two-stage artificial intelligence system that highlights predefined suspicious features, classifies images, and shows which regions influenced its prediction. The system was evaluated retrospectively using images and archived videos from three hospitals. External sensitivity was lower for individual frames, while patient-level estimates from only 25 external patients were imprecise. Prospective testing on uncurated live procedures is therefore required before clinical use.