Artificial intelligence-supported polyp detection (CADe) at colonoscopy reduces the detection of high grade dysplasia and invasive cancer

10.1055/a-2661-2624

Adenoma detection rate (ADR) correlates with the rate of post-colonoscopy colorectal cancer (PCCRC) and the rate of PPCRC-related death [1]. This is probably because ADR represents a measure of the meticulousness and comprehensiveness of mucosal inspection [2]; however, the value of detecting additional diminutive adenomas within the visual field, which current artificial intelligence (AI) systems improve, is unclear. This may decouple ADR from PCCRC, reducing its utility as a key performance indicator, and allows potential gaming [3].

A recent paper reported that endoscopists may be deskilled by using AI-supported computer-aided detection (CADe), with the ADR dropping from 28.4% before AI exposure to 22.4% by 3 months after CADe exposure [4]. Okumura et al. report in Endoscopy that improvements in ADR with AI-supported CADe continued over a 3-year period when endoscopists, particularly those who were high performing, were not using CADe, suggesting deskilling may be less of an issue with continuous exposure to AI, at least in a Japanese context [5].

The groups in the Okumura study were well matched; however, while ADR was improved and maintained with or without CADe, it was striking that the detection rates of adenomas of 6–9 mm, ≥10 mm, and with high grade dysplasia, and invasive cancer were significantly higher in the non-CADe group ([Fig. 1], extracted from table 2s in the paper of Okumura et al.). An alternative interpretation of this study is that CADe distracts endoscopists by encouraging them to focus on the detection of trivial lesions (adenomas of 1–5 mm and hyperplastic polyps) at the expense of more critical advanced lesions. The rate of invasive cancer in the non-CADe group was more than double the rate in the CADe group (P = 0.001). The consistency of effect direction among ≥10-mm adenomas, high grade adenomas, and invasive cancer, with very low P values, argues against a chance effect for advanced lesion detection. It is not yet clear that an AI-supported ADR leads to a reduction in PCCRC, and in fact these data would suggest the opposite might be true.

Fig. 1 Rates of lesion detection by pathology and size with and without AI-supported computer-aided detection (CADe).

Publication History

Article published online:
20 February 2026

• References

• 1 Corley DA, Jensen CD, Marks AR. et al. Adenoma detection rate and risk of colorectal cancer and death. NEJM 2014; 370: 1298-1306
• 2 Lee RH, Tang RS, Muthusamy VR. et al. Quality of colonoscopy withdrawal technique and variability in adenoma detection rates (with videos). Gastrointest Endosc 2011; 74: 128-134
• 3 East JE, Rittscher J. Artificial intelligence for colonoscopic polyp detection: High performance versus human nature. J Gastroenterol Hepatol 2020; 35: 1663-1664
• 4 Budzyn K, Romanczyk M, Kitala D. et al. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. Lancet Gastroenterol Hepatol 2025; 10: 896-903
• 5 Okumura T, Kudo SE, Ide Y. et al. Long-term impact of computer-aided adenoma detection: a prospective observational study. Endoscopy 2025;