Computer-Aided Detection Colonoscopy for Surveillance in IBD Patients: Insights from a Single-Center Experience.

Abstract

Objectives: The real-world efficacy of computer-aided detection (CADe) in improving surveillance colonoscopy performance for patients with inflammatory bowel disease (IBD) has not been established.

Methods: A retrospective, single-center study of surveillance colonoscopies in patients with IBD. Only colonoscopies indicated for surveillance, with adequate preparation and documented cecal intubation, were included. The study compared the collective adenoma detection rate (ADR) between the periods before (pre-CADe) (June 2020 to June 2021) and after (July 2021 to September 2022) the introduction of the CADe in all endoscopy units. An adjusted ADR was calculated using a multivariable logistic regression model.

Results: The study included 225 eligible colonoscopies performed during the pre-CADe period and 750 during the CADe period. Neoplastic lesions or colorectal cancer were detected in 13 (5.8%) of 225 procedures in the pre-CADe period and 27 (3.6%) of 750 procedures during the CADe period. The collective ADR was 5.2% (95% confidence interval, 3.9-6.6) in the pre-CADe period and 3.8% (95% confidence interval, 1.1-6.5) -following CADe implementation (P = .315). Subgroup analyses stratified by endoscopist experience, IBD type, and procedure timing (daytime vs after hours) corroborated a similar nonsignificant declining trend in ADR after CADe introduction.

Conclusions: In a real-world, single-center experience, the introduction of CADe did not improve neoplasms detection in patients with IBD and was associated with a nonsignificant decline in ADR. These findings call into question the utility of generic CADe systems in IBD surveillance and emphasize the need to foster IBD-specific CADe systems, as well as addressing challenges arising from physician-artificial intelligence interactions.