Even in skilled hands, a colonoscopy can miss things. Estimates from meta-analyses suggest adenoma miss rates of around 20–25% for smaller lesions: roughly one in four small polyps may not be detected on a first pass. If a colonoscopy is a human looking at a screen, could a second set of eyes, trained on tens of thousands of past procedures, help that human see more?
That second set of eyes now exists in the form of computer-aided detection (CADe). As the endoscopist advances the scope, the software analyses the same live video and highlights areas that resemble polyps, typically with an on-screen box or visual cue. It does not diagnose, remove, or decide. It prompts.
New Zealand has been part of the early evidence base. A study from Waitematā Endoscopy reported that adenoma detection rates increased from about 38.5% to 47.9% when CADe was used. A later randomised controlled trial conducted at Waitākere Hospital and published in 2026, involving roughly 776 patients, also found higher adenoma detection with AI assistance. The effect appeared more pronounced in screening colonoscopies, where detection increased from around 68% to 79% in that subgroup.
But the signal is not uniform across all settings. A 2023 systematic review of CADe trials suggests that while controlled studies often show a modest increase in detection, real-world implementation can produce smaller or inconsistent gains. In some routine practice data, detection rates differ only marginally between AI-assisted and standard procedures. Much of what CADe improves detection of are very small adenomas, and whether this translates into fewer cancers or deaths remains unproven.
That distinction matters. Higher detection rates do not automatically mean better long-term outcomes, particularly if the additional lesions are low-risk or clinically insignificant.
New Zealand, where bowel cancer remains one of the country’s most significant cancer burdens — with roughly 3,000+ diagnoses and around 1,200 deaths annually — operates the National Bowel Screening Programme. For most eligible people, screening begins with an at-home test every two years, with colonoscopy reserved for those who return a positive result.
The programme has been expanding and refining eligibility over time, though exact changes to starting age and rollout timing have been staged and remain subject to policy updates. Equity remains a key challenge, with Māori and Pacific peoples more likely to develop bowel cancer at younger ages and less likely to participate in screening.
The takeaway is not that a clever machine has solved bowel cancer. The test most of us will meet still hinges on turning up: doing the home test when it arrives, and following through if it comes back positive. The AI improves what happens next, but only for those who make it through the door.
SPONSORED
