Artificial intelligence-assisted colonoscopy improves adenoma detection rates in routine colonoscopy practice: a single-center, retrospective, propensity score-matched study with concurrent controls

Abstract

Background/Aims This study aimed to investigate whether a real-time artificial intelligence (AI)-assisted polyp detection system can improve adenoma detection rates (ADRs) in real-world colonoscopy practice.

Methods This single-center, retrospective, propensity score-matched study collected data from consecutive patients who underwent colonoscopy—either AI-assisted or standard colonoscopy— between March 2023 and February 2024. Propensity score matching was conducted to adjust for baseline characteristics across the groups.

Results During the study period, 1,085 patients who underwent colonoscopy were eligible for inclusion. After propensity score matching, 474 patients who underwent AI-assisted colonoscopy and 474 who underwent standard colonoscopy were included in the primary analysis. The ADR was significantly higher in the AI-assisted colonoscopy group than in the standard colonoscopy group (35.9% vs. 26.4%; p = 0.002). Additionally, the number of adenomas detected per colonoscopy was significantly higher in the AI-assisted group than in the standard group (0.69 ± 1.22 vs. 0.43 ± 0.91; p < 0.001). However, the detection rates of advanced adenomas and sessile serrated lesions did not differ significantly between the two groups.

Conclusion AI-assisted colonoscopy significantly improves ADRs in real-world colonoscopy practice.