Artificial intelligence with mucosal exposure devices versus artificial intelligence alone in detection of colorectal adenomas: a systematic review and meta-analysis.

BACKGROUND AND AIMS: Both artificial intelligence (AI) and mucosal exposure devices (MEDs) have been shown to improve adenoma detection rate and reduce miss rate during colonoscopy. However, the effectiveness of AI combined with MEDs remains unclear. In this meta-analysis, we have evaluated the combined effect of AI with MEDs on enhancing detection rates of various colonic lesions. METHODS: Several databases were reviewed to identify studies reporting data on outcomes of interest from inception to December 10, 2025. Our outcomes of interest were the adenoma detection rate (ADR), adenomas per colonoscopy (APC), sessile serrated detection rate (SSDR), and advanced adenoma detection rate (aADR). Further subgroup analyses were performed based on adenoma size, location, and mucosal exposure device type. Data were analyzed using the Mantel-Haenszel random-effects model. Heterogeneity was assessed using the I2 statistic. RESULTS: Twelve studies (8 RCTs, 2 retrospective, and 2 prospective) with a total of 6155 patients (2700 AI + MED, 3455 AI alone group) met the inclusion criteria. AI + MED were associated with higher rates of ADR, RR 1.15 (95% CI 1.07-1.25; p = 0.0005) as compared to AI alone, and subgroup analysis on device type also showed higher rates for AI + Endocuff RR 1.19 (95% CI 1.08-1.32; p = 0.0006) and AI + balloon device did not show significant results for ADR, RR 1.08 (95% CI 0.93-1.27; p = 0.32) between the groups, however, the formal test for subgroup differences between device type was not statistically significant (P = 0.33). There was no significant difference in aADR, SSDR, proximal ADR, cecal intubation time, withdrawal time, ADRs for size < 5 mm, and ADRs for size 6-9 mm between AI + MED vs. AI alone. CONCLUSIONS: AI-assisted colonoscopy augmented with a mucosal exposure device, particularly Endocuff, significantly enhanced ADR compared with AI alone, while aADR, proximal ADR overall, SSDR, and procedure times were comparable between strategies. These findings suggest that the selective use of Endocuff in AI-enabled colonoscopy may optimize lesion detection.