Artificial Intelligence-Driven Colonoscopy: A Systematic Review and Network Meta-Analysis on System Performance for Colorectal Neoplasia Detection.

Abstract

<h4>Background</h4>Colorectal cancer (CRC) is a major global health issue, with adenomatous polyps as the main precursors. Standard colonoscopy is the gold standard, but its effectiveness is limited by operator variability and missed lesions. Artificial Intelligence (AI)-assisted colonoscopy is emerging to enhance detection, but a comparative performance across different AI systems is unclear.<h4>Methods</h4>We conducted a systematic review and Bayesian network meta-analysis of 48 randomized controlled trials (RCTs) (<i>N</i> = 34 106 participants), searched across PubMed, Scopus, and Google Scholar up to November 4, 2025. We evaluated five commercial systems-EndoAngel, EndoAID, CAD-EYE, GI Genius, EndoScreener-and local platforms. Primary outcomes were Adenoma Detection Rate (ADR) and Adenomas Per Colonoscopy (APC).<h4>Results</h4>All AI systems significantly improved ADR versus conventional colonoscopy. EndoAngel showed the largest effect (OR 1.84; SUCRA 0.9), followed by EndoAID (OR 1.64; SUCRA 0.7). CAD-EYE (OR 1.46; SUCRA 0.5) and GI Genius (OR 1.45; SUCRA 0.5) also showed gains. APC gains were highest with EndoAID (MD 0.62). EndoAngel modestly increased withdrawal time (MD 1.14 min). Crucially, no AI system significantly improved the detection of high-risk lesions. Evidence quality was moderate.<h4>Conclusion</h4>AI-assisted colonoscopy improves adenoma detection over conventional methods. While EndoAngel and EndoAID show the largest gains, performance differences among systems are modest. Detection of high-risk lesions remains uncertain, underscoring the need for future head-to-head trials and cost-effectiveness studies to guide optimal implementation in CRC screening.