Multi-task deep-learning model standardises ulcerative-colitis endoscopy scoring

UC-MTLNet was evaluated on 405 patients and 10,269 endoscopic images from development, prospective internal-test and independent external-test cohorts. The paper reports strong agreement for the Ulcerative Colitis Endoscopic Index of Severity and faster inference than three single-task models, while using roughly one-third of their parameters and operations.

Why it matters: Multi-task learning may reduce scorer variability and deployment cost. Prospective video-based and outcome-linked validation is still needed; this is not a medical-device approval.

Paper: Scientific Reports