Performance of artificial intelligence software: EndoAngel, EndoAID, CAD-EYE, GI Genius, and EndoScreener in adenoma detection: an extended network meta-analysis.
Colorectal cancer is a major global health burden, with most cases arising from adenomatous polyps. Although colonoscopy is the gold standard for detection, its effectiveness is operator-dependent. Artificial intelligence-assisted systems have been developed to improve adenoma detection, but their comparative performance remains unclear.
We performed a systematic review and Bayesian network meta-analysis of randomized controlled trials comparing artificial intelligence-assisted with standard colonoscopy. PubMed, Scopus, and Google Scholar were searched up to 4 November 2025. Eligible studies included adults undergoing colonoscopy and reporting adenoma detection rate (ADR) and adenomas per colonoscopy (APC). Secondary outcomes included withdrawal time and detection of advanced and sessile serrated lesions. Risk of bias was assessed using Cochrane RoB 2.0, and certainty of evidence was evaluated with CINeMA.
A total of 48 randomized controlled trials (34 106 participants) were included. Artificial intelligence-assisted colonoscopy significantly improved ADR compared with standard colonoscopy. EndoAngel showed the greatest effect [odds ratio (OR): 1.84, surface under the cumulative ranking curve (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). For APC, EndoAID demonstrated the largest benefit (mean difference: 0.62). EndoAngel modestly increased withdrawal time (mean difference: 1.14 minutes). No system significantly improved detection of advanced or sessile serrated lesions. Heterogeneity was low, and certainty of evidence was moderate.
Artificial intelligence-assisted colonoscopy improves adenoma detection; however, differences between systems are small, and benefits for high-risk lesions remain uncertain. Further head-to-head trials and cost-effectiveness studies are needed.
We performed a systematic review and Bayesian network meta-analysis of randomized controlled trials comparing artificial intelligence-assisted with standard colonoscopy. PubMed, Scopus, and Google Scholar were searched up to 4 November 2025. Eligible studies included adults undergoing colonoscopy and reporting adenoma detection rate (ADR) and adenomas per colonoscopy (APC). Secondary outcomes included withdrawal time and detection of advanced and sessile serrated lesions. Risk of bias was assessed using Cochrane RoB 2.0, and certainty of evidence was evaluated with CINeMA.
A total of 48 randomized controlled trials (34 106 participants) were included. Artificial intelligence-assisted colonoscopy significantly improved ADR compared with standard colonoscopy. EndoAngel showed the greatest effect [odds ratio (OR): 1.84, surface under the cumulative ranking curve (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). For APC, EndoAID demonstrated the largest benefit (mean difference: 0.62). EndoAngel modestly increased withdrawal time (mean difference: 1.14 minutes). No system significantly improved detection of advanced or sessile serrated lesions. Heterogeneity was low, and certainty of evidence was moderate.
Artificial intelligence-assisted colonoscopy improves adenoma detection; however, differences between systems are small, and benefits for high-risk lesions remain uncertain. Further head-to-head trials and cost-effectiveness studies are needed.
Authors
Rafi Rafi, Yasmin Yasmin, Ghani Ghani, Siddiqui Siddiqui, Mahmood Mahmood, Ali Ali, Khalid Khalid, Khanzada Khanzada, Idrees Idrees, Huzaifa Huzaifa, Riaz Riaz
View on Pubmed