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

Area of Science:

• Gastroenterology
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Withdrawal time (WT) is a key quality indicator in colonoscopy, linked to adenoma detection and cancer prevention.
• WT aims to quantify polyp search time but includes other procedural elements like washing and biopsies.
• ColonPro is an AI tool designed to report inspection time (IT) during colonoscopy.

Purpose of the Study:

• To evaluate if the inspection time (IT) reported by the ColonPro AI system accurately reflects the actual time spent searching for polyps during colonoscopy.

Main Methods:

• A prospective study involving 100 colonoscopies performed by 11 endoscopists across two centers.
• Procedure time, WT, and IT were recorded, with IT manually measured by a research assistant using a stopwatch for comparison with AI-generated data.

Main Results:

• The AI program's reported IT was significantly longer (9.6 minutes) than the manually measured actual IT (7.4 minutes), with a difference of 2.1 minutes (P < .01).
• This overestimation of IT by the AI was consistent across all participating endoscopists and irrespective of whether polypectomy was performed.

Conclusions:

• The commercially available ColonPro AI program does not accurately measure the actual inspection time during colonoscopy.
• The findings suggest limitations in current AI technology for precisely quantifying specific procedural components like polyp search time in colonoscopy.