Does AI colonoscopy find more precancerous polyps than standard colonoscopy?

AI colonoscopy, also called computer-aided detection (CADe), raised the adenoma detection rate by about 20% in a meta-analysis of 22 randomised trials and lowered the adenoma miss rate by up to 55%. A 2025 Lancet Gastroenterology and Hepatology study also raised a warning: detection without AI fell after endoscopists got used to it.

Colonoscopy prevents colorectal cancer by finding and removing adenomas, the polyps that can turn cancerous. Doctors track quality with the adenoma detection rate (ADR), the share of colonoscopies in which at least one adenoma is found, which is associated with cancer prevention. Even skilled endoscopists miss some polyps, which is the gap AI tries to close by flagging suspicious areas on the live video in real time.

This page explains how AI colonoscopy works, what randomised trials and meta-analyses show, the deskilling concern, and what patients should ask, including those travelling for care in India, China or elsewhere.

Key facts on the evidence for computer-aided colonoscopy

Published results from meta-analyses and the deskilling study:

  • Meta-analysis of 22 parallel randomised trials (21,825 participants, Gastrointestinal Endoscopy): adenoma detection rate up 20% (risk ratio 1.20, 95% CI 1.14 to 1.27) and adenoma miss rate down 55% (risk ratio 0.45)
  • Network meta-analysis of 52 randomised trials (46,177 patients): CADe raised ADR by 7.4% compared with standard colonoscopy
  • Absolute effect: a meta-analysis of 44 randomised trials, cited in The Lancet Gastroenterology and Hepatology, reported an absolute ADR increase of about 8%
  • Deskilling study (Lancet Gastroenterology and Hepatology, 2025): ADR in non-AI colonoscopies fell from 28.4% (226 of 795) before AI exposure to 22.4% (145 of 648) after
  • ADR with AI assistance in the same study: 25.3% (186 of 734)
  • Unproven areas: effectiveness for flat or serrated adenomas has not been established, and long-term cancer outcomes were not found in the sources reviewed

Key takeaways: AI colonoscopy and polyp detection

  • In randomised trials, AI-assisted colonoscopy finds more adenomas: roughly 20% in relative terms and about 7% to 8% in absolute terms.
  • AI lowered the adenoma miss rate by up to 55% in a meta-analysis of 22 randomised trials.
  • An observational 2025 study found that endoscopists’ ADR without AI fell from 28.4% to 22.4% after exposure to AI, raising a deskilling concern.
  • The deskilling finding is observational and does not overturn the trials, but it has prompted editorial debate about training and monitoring.
  • Whether AI colonoscopy lowers colorectal cancer incidence or deaths was not shown in the sources reviewed.

How does AI-assisted colonoscopy work?

During colonoscopy, a camera on a flexible tube sends live video of the bowel wall to a monitor. CADe software, trained on large numbers of annotated polyp images, analyses the video in real time and draws a box or sounds an alert when it sees a possible polyp. The endoscopist decides whether the finding is real and whether to remove it. Some systems also offer computer-aided diagnosis that suggests whether a polyp is an adenoma, so that small harmless polyps may be left alone.

Because the AI only flags, quality still depends on bowel preparation, withdrawal technique and careful inspection. CADe systems are built on deep convolutional neural networks.

What do meta-analyses of randomised trials show about AI colonoscopy?

Evidence is unusually strong for an AI tool. A meta-analysis of 22 parallel randomised trials with 21,825 participants found that AI-assisted colonoscopy raised ADR by 20% (risk ratio 1.20) and lowered the adenoma miss rate by up to 55%. A network meta-analysis of 52 randomised trials in 46,177 patients found CADe raised ADR by 7.4% against standard colonoscopy, by 4.4% against pigmented endoscopy and by 4.1% against mucosal exposure techniques, so AI added value beyond other add-on methods.

Commentary in The Lancet Gastroenterology and Hepatology describes CADe as one of the most extensively evaluated uses of AI in medicine, citing a meta-analysis of 44 randomised trials with an absolute ADR increase of 8%.

Key detection numbers for computer-aided polyp detection at a glance

Summary of the pooled trial evidence and the deskilling study:

  • ADR increase: relative 20% (risk ratio 1.20); absolute about 7.4% to 8%
  • Adenoma miss rate: down by up to 55% (risk ratio 0.45)
  • Versus add-on techniques: ADR higher by 4.1% to 4.4% than pigmented endoscopy or mucosal exposure devices
  • Trials pooled: 22 (21,825 participants), 44 and 52 (46,177 patients) across different meta-analyses
  • ADR in AI-assisted colonoscopies in the deskilling study: 25.3%

What did the deskilling study in the Lancet Gastroenterology and Hepatology find?

In 2025, a multicentre observational study with Yuichi Mori of the University of Oslo among its authors examined endoscopists’ performance on standard, non-AI colonoscopies before and after AI became part of routine practice. The ADR for non-AI colonoscopies fell from 28.4% (226 of 795) before AI exposure to 22.4% (145 of 648) afterwards, a 6% absolute and 20% relative reduction. In AI-assisted colonoscopies, ADR was 25.3% (186 of 734).

The authors said the results raise a question about whether earlier randomised trials capture the whole picture, because routine AI use may reduce endoscopists’ unaided skills. The study was observational, so other explanations, such as changes in patient mix or case selection, cannot be excluded.

Is endoscopist deskilling a real risk? What we know and do not know

A commentary in The Lancet Gastroenterology and Hepatology asked whether deskilling is an unintended consequence of AI-assisted colonoscopy. Other 2026 studies explore the issue, including a pragmatic prospective trial in Endoscopy on learning and deskilling effects among endoscopists with different experience, and eye-tracking studies of how endoscopists respond to false-positive prompts. We have not reviewed their full results here.

The practical lessons are plausible without being proven: audit each endoscopist’s unaided ADR, keep periods of unaided practice, train staff to question false alerts and not rely on the box, and monitor patient outcomes. For patients, the message is not to avoid AI colonoscopy but to choose a unit that measures its quality.

Does computer-aided colonoscopy reduce colorectal cancer or deaths?

ADR is a surrogate. Higher detection is associated with colorectal cancer prevention, but the trials measured adenomas found, not cancers prevented or lives saved. A review of meta-analyses notes that the effectiveness of AI for flat or serrated adenomas remains unproven. We did not find, in the sources reviewed, trial evidence that AI colonoscopy reduces interval colorectal cancer or deaths.

That does not make the technology unhelpful. Finding adenomas that would otherwise be missed is a logical step toward prevention, because removing adenomas is known to reduce cancer risk. But the size of the benefit, and whether it holds across populations, needs long-term follow-up. Screening choices should follow national guidelines.

Limits of the evidence for computer-aided detection

The randomised evidence is strong for detection and silent on long-term outcomes.

  • Trials measure adenoma detection, a surrogate, not cancer incidence or deaths
  • Added detections may be mostly small adenomas, whose individual clinical importance is debated
  • The deskilling study is observational and from specific settings
  • Performance for flat or serrated lesions is unproven, and false-positive alerts (averaging about 16.9% across studies in one review) can add polypectomies and cost

False alerts, extra polypectomies and cost: what should patients know?

AI can flag normal folds, bubbles or debris as polyps. A systematic review of meta-analyses reported a mean false-positive rate of about 16.9% across studies, ranging widely from 10.3% to 39%. The endoscopist usually dismisses false alerts quickly, but they can add seconds to the procedure and sometimes lead to the removal of harmless tissue.

Removing polyps carries small risks of bleeding or perforation, which is why endoscopists weigh each finding. AI software also adds cost, and whether health systems recover that through better prevention is not yet settled. Patients should not decline AI-assisted colonoscopy for fear of false alerts, but they can ask the unit how it audits quality.

Who could ask a doctor about AI-assisted colonoscopy?

Screening ages and intervals follow national guidelines. These are situations to raise with a doctor:

  • You are due for screening colonoscopy under your country’s guidelines
  • You have had polyps removed before and need surveillance colonoscopy
  • You have blood in the stool, a change in bowel habit, unexplained weight loss or iron-deficiency anaemia, which need diagnostic evaluation
  • A close relative had colorectal cancer, or an inherited syndrome such as Lynch syndrome is suspected
  • You want to know whether the unit uses AI and measures its adenoma detection rate
  • You want a second opinion on a colonoscopy or biopsy report

How does computer-aided detection compare with other ways to improve polyp detection?

Other add-ons include mucosal exposure devices such as Endocuff-type attachments, dye-based pigmented endoscopy and advanced imaging. The network meta-analysis found CADe raised ADR by 7.4% against standard colonoscopy, by 4.4% against pigmented endoscopy and by 4.1% against mucosal exposure techniques, so AI performed at least as well as these alternatives. A separate meta-analysis has examined combining CADe with an Endocuff device, although that evidence is newer.

The basics still matter most: a well-cleaned bowel, adequate withdrawal time and an experienced endoscopist. AI is an adjunct to good technique, not a substitute.

Colonoscopy AI evidence and access by region

What has been published or registered, with gaps stated:

  • Randomised evidence: trials from several countries pooled in meta-analyses of 22, 44 and 52 trials
  • Population screening: a 2026 randomised trial in a population-based colorectal cancer screening programme tested computer-assisted colonoscopy for adenoma detection (Endoscopy)
  • Developing countries: a registered randomised trial (NCT07066046) notes a lack of studies of AI-assisted colonoscopy in developing countries
  • India and China: we have not verified which centres use AI-assisted colonoscopy or which systems are approved locally

Is AI colonoscopy available in India or China?

AI polyp-detection systems are sold in many countries, but we have not verified which hospitals or screening programmes in India or China use them, which systems they use or whether those systems hold local regulatory approval. A registered trial in a developing-country setting states that studies there are lacking, so evidence from the trials above should be applied with care.

If you are having colonoscopy abroad, ask whether AI is used, whether the unit records its adenoma detection rate, and how quickly pathology will be reported.

What happens next for computer-aided colonoscopy research?

Next steps are outcome-focused: trials in population screening programmes, long-term follow-up for interval colorectal cancer, studies of training and unaided performance, and computer-aided diagnosis that could reduce unnecessary polyp removal. A 2026 Endoscopy trial examined learning and deskilling effects across experience levels, and eye-tracking studies are exploring how endoscopists respond to false alerts.

Better-targeted evaluation should show whether a 7% to 8% absolute gain in ADR translates into fewer cancers and deaths, and whether endoscopists’ skills can be preserved. Related efforts, such as AI clinical trial matching, aim to help patients find clinical trials of new approaches once a diagnosis is made.

Questions to ask before an AI-assisted colonoscopy

  • Does this unit use AI polyp detection, and does the endoscopist still inspect the whole bowel wall?
  • What is the unit’s adenoma detection rate with and without AI?
  • How will you decide which polyps to remove, and what are the risks of removal?
  • If polyps or cancer are found, how soon will pathology be available and who will explain it?
  • When should I have my next colonoscopy based on today’s findings and my family history?
  • If cancer is found, can my images and pathology be reviewed for a second opinion?

Sources

Where the figures in this guide come from:

Gastrointestinal Endoscopy: Use of artificial intelligence improves colonoscopy performance in adenoma detection, a systematic review and meta-analysis03471-0/fulltext)

The Lancet Gastroenterology and Hepatology: Endoscopist deskilling, an unintended consequence of AI-assisted colonoscopy? (August 12, 2025)00164-5/abstract)

The Lancet Gastroenterology and Hepatology: Endoscopist deskilling risk after exposure to AI in colonoscopy, a multicentre observational study (2025)

EurekAlert: Routine AI assistance may lead to loss of skills in health professionals who perform colonoscopies, study suggests

International Journal of Colorectal Disease: Effectiveness of AI-assisted colonoscopy, a systematic review and network meta-analysis (2025)

World Journal of Gastrointestinal Endoscopy: Current status of AI colonoscopy on improving adenoma detection rate, a systematic review of meta-analyses

PMC: Integration of Endocuff-assisted and computer-aided colonoscopy, a meta-analysis of randomized controlled trials

ClinicalTrials.gov: AI-assisted colonoscopy in the detection and characterization of colorectal lesions, randomized trial (NCT07066046)

arXiv: Gaze responses to false-positive computer-aided detection prompts during colonoscopy, an eye-tracking study

Frequently asked questions

Does AI colonoscopy find more polyps?

Yes, in randomised trials. A meta-analysis of 22 trials with 21,825 participants found a 20% relative increase in adenoma detection and up to 55% fewer missed adenomas, and a network meta-analysis of 52 trials found a 7.4% rise against standard colonoscopy. These measure adenomas found, not cancers prevented, which has not yet been shown.

What is computer-aided detection (CADe) in colonoscopy?

CADe is software that analyses live colonoscopy video in real time and alerts the endoscopist to possible polyps with a box or sound. It is trained on many annotated polyp images and built on deep neural networks. The endoscopist decides whether a finding is real and whether to remove it, so the doctor remains responsible.

Can AI make endoscopists worse at finding polyps?

One 2025 observational study suggests it might. Adenoma detection in non-AI colonoscopies fell from 28.4% before AI exposure to 22.4% afterwards. The study was observational, so other explanations cannot be excluded, and it does not overturn the randomised trials. It has prompted debate about training, unaided practice and monitoring of each endoscopist’s performance.

Does AI colonoscopy reduce the risk of dying from colon cancer?

That has not been shown in the sources reviewed. The trials measured adenoma detection rate, which is associated with colorectal cancer prevention, rather than cancer cases or deaths. Long-term follow-up studies are needed. Regular screening according to national guidelines remains proven to reduce risk, whether or not AI is used.

Is AI-assisted colonoscopy available in India or China?

We have not verified which hospitals or screening programmes in India or China use AI-assisted colonoscopy, or which systems hold local approval. Ask your centre whether AI is used and whether it records its adenoma detection rate.

Should I avoid AI colonoscopy because of false alerts?

No. False alerts, averaging about 16.9% across studies in one review, are usually dismissed quickly by the endoscopist and add little time. The evidence shows more adenomas are found with AI. A better question is whether the unit audits its quality, including how often polyps are removed and its adenoma detection rate.

What is the adenoma detection rate?

The adenoma detection rate, or ADR, is the share of colonoscopies in which at least one adenoma is found. Adenomas are polyps that can turn cancerous, and doctors use ADR to track quality because it is associated with cancer prevention.

In randomised trials, AI raised ADR by roughly 20% in relative terms and about 7% to 8% in absolute terms. ADR is a surrogate, though: the trials measured adenomas found, not cancers prevented or lives saved. You can ask your unit what its ADR is, with and without AI.

How does AI compare with other ways to find more polyps?

Other add-ons include mucosal exposure devices, dye-based pigmented endoscopy and advanced imaging. A network meta-analysis found AI raised ADR by 7.4% against standard colonoscopy, by 4.4% against pigmented endoscopy and by 4.1% against mucosal exposure techniques.

The basics still matter most: a well-cleaned bowel, adequate withdrawal time and an experienced endoscopist. AI is an adjunct to good technique, not a substitute for it.

This page is general information, not medical advice. OncoVOIS explains and helps you prepare; your oncologist decides about your care.