Impact of Artificial Intelligence Use on Endoscopist Optical Diagnosis of Sessile Serrated Lesions, Traditional Serrated Adenomas, and Advanced Conventional Adenomas.

Abstract

Evaluate Computer Aided Diagnosis (CADx)-assisted and unassisted endoscopist optical diagnosis of sessile serrated lesions (SSLs), traditional serrated adenomas, and advanced adenomas. We performed an IRB-approved secondary analysis of a large prospective colonoscopy cohort. Of 2,111 polyps (1,011 patients), 116 were histopathology-proven SSLs. Sensitivity for SSL identification of CADx-assisted and unassisted optical diagnosis was 55.7% (95% CI 42.9-67.8) and 38.6% (95% CI 22.7-57.3), respectively. Endoscopists correctly identified SSLs more frequently when CADx classified these as hyperplastic compared with neoplastic (66.7% [95% CI 51.8-78.9] vs 24.5% [95% CI 10.8-46.3]; P < 0.001). CADx use did neither significantly harm nor improve optical diagnostic performance for SSLs despite CADx inability to classify SSLs.