Development and Application of an Artificial Intelligence-assisted Endoscopic System for Automatic and Accurate Diagnosis of Colorectal Ulcers

Crohn's disease,ulcerative colitis,intestinal Behcet's disease,intestinal tuberculosis,and primary intestinal lymphoma are common colorectal ulcers.We developed an artificial intelligence(AI)-assisted ecdoscopic diagnostic system to accurately disgnose these five diseases.This is a multi-center retrospective-prospective study.AI system was developed based on the endoscopic images using convolutional neural networks(CNNs) and transformers.The AI system was first tested at multiple centers,then compared with endoscopists,and evaluated in a prospective study.In the retrospective internal test,the system exhibited an accuracy of 96.8%.In the multi-center test,the diagnostic accuracy of model was 83.4%,which was significantly higher than that of senior endoscopists(83.4% vs 59.4%,P<0.001).The accuracy of the best comprehensive diagnostic model trained with both clinical data and endoscopic images was 76.3%.The system can effectively improve the diagnostic accuracy of colorectal ulcers.