Development and Validation of a Deep Learning Algorithm to Evaluate Endoscopic Disease Activity of Ulcerative Colitis

Ulcerative colitis is a non‐specific chronic inflammation of gut characterized by referral bloody stool, diarrhea and abdominal pain. Endoscopic features of the disease severity and mucosal healing stage are strongly associated with treatment response and prognosis in the future. Currently, the Mayo endoscopic sub‐score (Mayo ES) and Ulcerative colitis endoscopic index of severity (UCEIS) are commonly recommended to guide therapeutic adjustments. However, the accuracy of these scales greatly relies on intra‐observer and inter‐observer consistency for lack of objective measurements. Recently, deep learning algorithm based on convolutional neural network (CNN) has shown multiple potential in computer‐aided detection and computer‐aided diagnose of gastrointestinal lesions. Up to now, no randomized controlled trials have been conducted to evaluate the performance of deep learning algorithm for assessing disease activity in ulcerative colitis. This study aims to train a deep learnig algorithm to assess severity and mucosal healing stage of ulcerative colitis using the Mayo ES and UCEIS scale, then testify whether the engagement of AI can improve the evaluation accuracy of physicians in a multi‐center clinical practice.