Clinical Efficacy of Real-Time Artificial Intelligence-Assisted Colonoscopy in Colorectal Polyp Detection: A Prospective Multicenter Randomized Controlled Trial.

BACKGROUND/AIMS: Early detection and removal of colon polyps are critical for preventing colorectal cancer. Computer-aided detection (CADe) systems have been introduced to increase the polyp detection rate (PDR) during colonoscopy, potentially enhancing its effectiveness. This study aimed to evaluate the efficacy of a CADe system in colorectal neoplasm detection.

METHODS

This prospective, randomized controlled trial was conducted at two tertiary centers (May 2023 to April 2025). Patients were randomly assigned to CADe or conventional colonoscopy and underwent screening, surveillance, or diagnostic colonoscopy. The primary endpoint was the adenoma detection rate (ADR), while the secondary endpoints were the PDR, relative risk (RR) of polyp detection, adenomas per colonoscopy (APC), and factors influencing adenoma detection.

RESULTS

Of 1,004 enrolled patients, 998 were randomly allocated into CADe and conventional colonoscopy groups (497 CADe system and 501 conventional colonoscopy). The CADe group had greater polyp counts (2.2 per colonoscopy vs 1.4 per colonoscopy; p<0.001) and APC values (1.2 vs 0.8; p<0.001). The CADe group showed significantly higher PDRs (72.2% vs 54.5%; p<0.001; RR, 2.173; 95% confidence interval [CI], 1.669 to 2.828) and ADRs (52.3% vs 36.1%; p<0.001; RR, 1.940; 95% CI, 1.505 to 2.499). CADe also significantly increased the detection rate of hyperplastic polyps (p=0.007; RR, 1.474; 95% CI, 1.113 to 1.952) and increased the detection rates across all sizes and locations. In multivariable analysis, CADe use was the strongest independent predictor of adenoma detection (odds ratio, 1.914; 95% CI, 1.467 to 2.496), outweighing male sex, older age, diagnostic indication, and withdrawal time.

CONCLUSIONS

Real-time CADe-assisted colonoscopy significantly increased PDR and ADR and proved to be a strong independent predictor of adenoma detection (cris.nih.go.kr, KCT0009664).