AI-supported detection systems in colonoscopies: Help or hype?
Current guidelines advise against the routine use of AI-supported systems for detecting polyps. Why is this the case, and what is the data situation?
Uncertain benefits and negative effects on skills
- Current guidelines do not recommend the routine use of computer-aided detection (CAD) systems in colonoscopies, as their use could lead to more follow-up colonoscopies and overdiagnosis 3,4
- Recent data suggest that the implementation of AI may lead to an unintended reduction in the detection rate (ADR) in AI-assisted standard colonoscopies, as heavy reliance on automated processes is accompanied by a loss of competence among staff (‘deskilling effect’) 2
Current limitations and lack of evidence of benefit
After evaluating 44 studies on CAD systems in polyp detection, a panel of experts spoke out against their routine use. Although the adenoma detection rate (ADR) increased by 8% compared to standard colonoscopy, there was no evidence of any relevant clinical benefit, for example in terms of the incidence of colorectal cancer or mortality.
At the same time, a modelling study suggested that AI-assisted colonoscopies increase the burden on the healthcare system, as 6.4% more people need to be referred for follow-up examinations after screening with CAD than after screening without CAD, according to a recent article in the British Medical Journal.1,5 A randomised clinical trial is currently underway to quantify the effects and estimates.
Although improvements in the rates of detected and missed adenomas have been reported for AI CAD systems, particularly in the detection of smaller polyps, the efficiency in detecting advanced adenomas or sessile serrated lesions is modest, as a recent review in “Cancers” points out.2
Unintended effects on the quality of care
Another major challenge being discussed is the increasing dependence on automated processes and the resulting loss of expertise among specialist staff. ‘Operator deskilling’ refers to the decline in technical and diagnostic skills due to familiarity with computer-assisted solutions. Deskilling can already be observed in various medical fields following the widespread and rapid introduction of AI in healthcare.2
There is also concrete data on the dequalification of endoscopists, which was able to demonstrate this effect relatively quickly. A comparative evaluation of 1,443 conventional (non-AI-assisted) colonoscopies performed in the phase before (795) and after (648) the introduction of CAD systems showed a significant deterioration in the adenoma detection rate (ADR) from 28.4% before CAD to 22.4% afterwards, which corresponds to an absolute reduction of about 6% and a relative reduction of 22%.2,6
Conclusion
Some counter the last point on de-skilling by arguing that AI-supported systems simply need to be used and switched on at all times. However, this argument is invalid because AI is not capable of replacing a thorough examination by an experienced clinician.
It is often presented as a supportive tool for refining or improving performance, but whether it can achieve this in the context of the challenges of colorectal cancer screening is as unclear as the question of its long-term benefits in terms of colorectal cancer incidence and mortality rates.2
- Linhares, S. M., Schultz, K. S. & Mongiu, A. K. Computer aided polyp detection has limited clinical efficacy. BMJ 389, r732 (2025).
- Spadaccini, M. et al. AI and Polyp Detection During Colonoscopy. Cancers (Basel) 17, 797 (2025).
- KI bei Darmspiegelungen: Neue Leitlinie rät von Routineeinsatz ab. https://www.esanum.de/today/posts/neue-leitlinie-ki-bei-darmspiegelungen-wird-nicht-routinemaessig-empfohlen.
- Foroutan, F. et al. Computer aided detection and diagnosis of polyps in adult patients undergoing colonoscopy: a living clinical practice guideline. BMJ 388, e082656 (2025).
- Halvorsen, N. et al. Benefits, burden, and harms of computer aided polyp detection with artificial intelligence in colorectal cancer screening: microsimulation modelling study. bmjmed 4, (2025).
- Budzyń, K. et al. Endoscopist De-Skilling after Exposure to Artificial Intelligence in Colonoscopy: A Multicenter Observational Study. SSRN Scholarly Paper at https://doi.org/10.2139/ssrn.5070304 (2024).