External Validation of an AI-Based Colonoscopy Quality Assessment Tool in Japanese Clinical Practice: A Pilot Study

While AI-assisted polyp detection has advanced significantly, comprehensive AI-based colonoscopy quality assessment tools remain limited. This study aimed to externally validate an AI-based colonoscopy quality assessment tool (AI-CQ) in Japanese clinical practice.

This retrospective study analyzed 40 screening colonoscopy procedures at a tertiary endoscopy hospital in Japan. All videos were recorded using Olympus equipment from anal insertion to complete withdrawal. AI-CQ measurements were compared with manual assessments by a certified endoscopist. Metrics evaluated included insertion time (IT), withdrawal time (WT), withdrawal time minus polypectomy time, and high-quality withdrawal time (HQ-WT).

AI-CQ successfully identified cecal intubation in 97.5% of cases. Correlations between manual and AI measurements were moderate for IT (ρ = 0.48, p = 0.0025) and strong for WT (ρ = 0.70, p < 0.001). HQ-WT differed significantly by endoscopist experience (388 vs. 273 s, p = 0.025) and polyp detection status (497 vs. 298 s, p < 0.001). These associations remained significant even after adjusting for manual polyp observation time.

AI-CQ demonstrated feasibility for measuring colonoscopy quality metrics in Japanese practice, though with reduced correlation compared to US validation. Given the pilot nature of this study, these findings should be interpreted as exploratory. Novel metrics like HQ-WT showed promising associations with procedural quality indicators, warranting further validation in larger cohorts.