Inspection time during colonoscopy withdrawal as measured by a second-generation artificial intelligence program

Abstract

Background and aims: Withdrawal time (WT) is a recommended quality measure for colonoscopy and is associated with adenoma detection and cancer prevention in retrospective studies. WT is intended to measure the time spent searching the colon for polyps but includes the time spent washing, suctioning, taking biopsy samples, and inspecting and resecting colon polyps. ColonPro (Medtronic; Minneapolis, Minn, USA) is a second-generation artificial intelligence (AI) program that reports inspection time (IT). We sought to determine whether the IT reported by ColonPro reflects the actual time searching for polyps. Methods: We prospectively recorded the procedure time, WT, and IT measured by ColonPro in 100 colonoscopies performed by 11 endoscopists at 2 centers. A trained research assistant measured IT manually with a stopwatch. Results: Total procedure time and WT were similar when measured manually or by the AI program. AI-measured IT was 2.1 minutes longer than the actual stopwatch-measured IT (9.6 [2.8] vs 7.4 [2.6] minutes, P < .01). The AI-measured IT was longer for all endoscopists and for procedures with or without polypectomy. Conclusions: Actual IT is not accurately measured by a new commercially available AI program.