Comparative Performance of Artificial Intelligence‐Based Computer‐Aided Detection Systems for Colorectal Polyps: A Systematic Review and Network Meta‐Analysis

BACKGROUND AND AIMS: Computer-aided detection (CADe) is anticipated to enhance adenoma detection rate (ADRs). The aim of this study was to systematically collect randomized-controlled trials comparing colonoscopy with CADe to standard colonoscopy without CADe in ADRs. METHODS: We performed a Bayesian network meta-analysis of randomized-controlled trials. Three electronic databases including MEDLINE, Embase, and the Cochrane Central Register of Controlled Trials were searched. The primary outcome was the comparison of the performance of CADe systems in ADRs; the secondary outcome was the sessile serrated lesions detection rates (SSLDRs). RESULTS: A total of 48 randomized controlled trials involving 38,986 patients were included in the quantitative analysis. Several CADe systems improved ADR compared with controls that ENDO-AID (risk ratio [RR] 1.26, 95% credible interval [CrI] 1.14-1.40), CADEYE (RR 1.18, 95% CrI 1.10-1.26), and GI Genius (RR 1.15, 95% CrI 1.08-1.22) were supported by moderate confidence evidence according to the Confidence in Network Meta-Analysis (CINeMA). For SSLDR, ENDO-AID (RR 1.36, 95% CrI 1.03-1.79) and GI Genius (RR 1.25, 95% CrI 1.08-1.46) may offer improved detection compared with controls. Across multiple sensitivity analyses excluding studies by withdrawal time, conflicts of interest, limited study numbers, image-enhanced endoscopy, non-parallel design, single-center settings, operator experience, or earlier publication years, the direction and magnitude of ADR improvements with CADe systems remained largely consistent with the primary analysis. CONCLUSIONS: Based on the CINeMA framework, the certainty of evidence ranged from low to moderate, indicating that some CADe systems are likely to improve ADR.