Multiple artificial intelligence (AI) computer-aided detection (CADe) systems have been developed to improve adenoma detection rate (ADR) during colonoscopy. However, their comparative effectiveness remains uncertain. This network meta-analysis aims to compare the efficacy of different AI-based CADe systems in increasing ADR.

We systematically searched MEDLINE, Embase, Scopus, and the Cochrane Library for randomized controlled trials (RCTs) comparing any CADe system with standard high-definition colonoscopy or with another CADe system. The primary outcome was ADR, expressed as risk ratio (RR) with 95% confidence interval (CI). A random-effects network meta-analysis (DerSimonian and Laird method) was performed to synthesize direct and indirect evidence. The relative ranking of all interventions was estimated using the surface under the cumulative ranking curve (SUCRA). Methodological quality was assessed with the Cochrane risk-of-bias (RoB2) tool.

The network included standard colonoscopy and 12 CADe systems (CAD EYE FujiFilm, CADe Henan, CADe LPIXEL, Deep2 FujiFilm, Eagle-Eye, ENDO-AID Olympus, Endoangel, Endoscreener, Endovigilant, GI Genius, MAGENTIQ-COLO, SKOUT, YOLOv3). Compared with standard colonoscopy, most CADe systems significantly increased ADR, with the highest effect estimates observed for YOLOv3 (RR 1.76; 95%CI 1.35–2.30), CADe Henan (RR 1.64; 95%CI 1.31–2.04), Endoangel (RR 1.55; 95%CI 1.15–2.08), Deep2 FujiFilm (RR 1.37; 95%CI 1.06–1.76), ENDO-AID Olympus (RR 1.25; 95%CI 1.14–1.38) and GI Genius (RR 1.13; 95%CI 1.06–1.21). On the other hand, Eagle-Eye, SKOUT, CADe LPIXEL, MAGENTIQ-COLO, and Endovigilant did not reach statistical significance. SUCRA-based ranking showed YOLOv3, CADe Henan, Endoangel, Deep2 FujiFilm, and ENDO-AID Olympus as the top-performing systems, while Endovigilant ranked lowest. Most RCTs were of moderate or high risk of bias due to inability to blind endoscopists to the intervention.

This network meta-analysis provides a comprehensive comparative assessment of the relative efficacy across all currently evaluated CADe systems, with YOLOv3, CADe Henan, Endoangel, Deep2 FujiFilm, and ENDO-AID Olympus emerging as the most effective. These findings may help endoscopy units prioritize AI solutions according to expected diagnostic yield and comparative performance.