Controversies in Computer-Assisted Detection in Colonoscopy

Artificial intelligence (AI) applications in endoscopy, particularly computer-aided detection (CADe), have shown consistent benefit in randomized controlled trials (RCTs), with improvements in adenoma detection rate (ADR) and reductions in adenoma miss rate (AMR). Despite these findings, adoption of CADe in routine colonoscopy remains controversial, with international guidelines issuing divergent recommendations.

Evidence from RCTs demonstrates that CADe increases ADR, predominantly through detection of diminutive adenomas, while its effect on advanced adenomas is limited. Real-world implementation studies show comparatively diminished benefits, likely explained by factors which are difficult to measure, such as the absence of Hawthorne effect in real-world practice, the quality of mucosal exposure and decision-making regarding diminutive polyps. Cost-effectiveness analyses generally favour CADe even with varying assumptions across healthcare systems, although these are based on the high degree of improvement in ADR seen in RCTs with CADe. Potential harms include increased polypectomy of non-neoplastic lesions, higher lifetime colonoscopy burden, and the risk of deskilling among endoscopists. Concerns remain about bridging the gap between trial efficacy and real-world effectiveness, optimizing surveillance intervals, and mitigating deskilling and human-AI interaction issues.

(1) CADe improves ADR in RCTs, but real-world effectiveness is inconsistent and often lacklustre. (2) Gains in ADR are largely derived from diminutive adenomas, and less with advanced adenomas, with uncertain impact on clinically significant outcomes such as colorectal cancer incidence and mortality. (3) Cost-effectiveness analyses are generally favourable, but dependent on assumptions about ADR improvement, CADe cost, and surveillance policies. (4) Deskilling and altered endoscopist behaviour represent important considerations that require further study. (5) Future integration of CADe with computer-aided diagnosis (CADx) and quality-assurance (CAQ) tools may maximize clinical benefit and cost-effectiveness, but evidence gaps must be addressed before widespread implementation.