Improving Colorectal Cancer Detection with AI-Assisted Colonoscopy: A Systematic Review and Meta-Analysis of 38 RCTs with GRADE Assessment.

BACKGROUND

Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide. Early detection of precancerous lesions such as adenomas and polyps is vital for prevention, yet standard colonoscopy may miss up to 26% of adenomas. Artificial intelligence (AI)-assisted colonoscopy has emerged as a promising tool to enhance lesion detection. This meta-analysis provides an updated synthesis of randomized controlled trials (RCTs) evaluating the diagnostic performance of AI-assisted versus standard colonoscopy.

METHODS

Following PRISMA guidelines, a comprehensive search of PubMed, Cochrane, Web of Science, and Scopus identified 38 RCTs including 29,745 participants. Pooled risk ratios (RRs) or mean differences (MDs) with 95% confidence intervals (CIs) were calculated using random-effects models. Subgroup analyses by geographical region and AI manufacturer were performed to explore heterogeneity. Certainty of evidence was rated using the GRADE approach, and publication bias was assessed through funnel plots and Egger's test.

RESULTS

AI-assisted colonoscopy significantly improved adenoma detection rate (ADR: RR = 1.19, 95% CI 1.14-1.25) and polyp detection rate (PDR: RR = 1.19, 95% CI 1.13-1.25), while reducing adenoma miss rate (AMR: RR = 0.52, 95% CI 0.38-0.72) and polyp miss rate (PMR: RR = 0.53, 95% CI 0.33-0.88). The mean number of adenomas per colonoscopy increased (MD = 0.20, 95% CI 0.14-0.25). AI modestly prolonged withdrawal time (MD = 0.46 min, 95% CI 0.26-0.66). No significant difference was observed for carcinoma detection (RR = 1.12) or advanced adenoma miss rate (RR = 0.67). Sessile serrated lesion detection showed a nonsignificant trend toward improvement (RR = 1.13), becoming significant after outlier exclusion (RR = 1.23). Subgroup analyses revealed higher ADR improvement in East Asian studies and with specific AI systems (e.g., Wision AI). Egger's test suggested minor small-study effects for ADR and withdrawal time.

CONCLUSIONS

AI-assisted colonoscopy substantially enhances adenoma and polyp detection while reducing miss rates. However, its benefit for carcinoma and sessile serrated lesion detection remains uncertain. Further standardized RCTs with long-term follow-up are warranted to confirm its role in colorectal cancer screening.