Artificial intelligence in colonoscopy for colorectal cancer prevention: a comprehensive narrative review of technology, evidence and clinical implementation

BACKGROUND AND OBJECTIVE: Adenoma miss rates (AMRs) reduce the effectiveness of colonoscopy in preventing colorectal cancer (CRC). Artificial intelligence (AI) systems are being developed to address this limitation. This article reviews current evidence on their performance, prevention potential, real-world implementation, and cost-effectiveness, while offering practical guidance for gastroenterologists. METHODS: A comprehensive literature search was conducted across PubMed/MEDLINE, Embase and Cochrane Library databases from January 2015 to February 2026. All studies focusing on AI on colonoscopy and major society position statements were all included. Evidence was synthesized thematically with critical appraisal and clinical contextualization. KEY CONTENT AND FINDINGS: Pooled data from meta-analyses confirms an approximately 20% relative increase in adenoma detection rate (ADR) and an approximately 50% relative reduction in AMR with computer-aided detection (CADe). ENDOANGEL demonstrated the highest efficacy in one network meta-analysis [odds ratio (OR) 1.84], followed by EndoAID (OR 1.64), though system rankings vary across analyses and should be interpreted with caution. Benefits are most prominent with diminutive early lesions, with limited impact on advanced adenomas or serrated lesions. A significant efficacy-effectiveness gap can however be seen between trial and real-world settings, driven by alert fatigue, Hawthorne effects, and endoscopist variability. Cost-effectiveness has been demonstrated in modeling studies conducted in European and American settings, with incremental cost-effectiveness ratios (ICERs) ranging from EUR 23,000 to USD 122,539 per quality-adjusted life-year (QALY) depending on the healthcare context. Implementation barriers remain substantial, consisting of organizational, technical, educational, and equity factors. CONCLUSIONS: AI systems have demonstrated measurable improvements in adenoma detection metrics in controlled trial settings; however, current evidence is insufficient to support a definitive conclusion about enhanced colonoscopy quality, and guidelines reflect ongoing uncertainty given uncertainty about long-term outcomes, real-world effectiveness and the balance of benefits with harms. Prospective trials with cancer incidence endpoints are essential to guide wider adoption.