Let me be honest about something. The first time I watched a CADe system flag a flat lesion I had just passed over, my reaction was not admiration — it was a flash of irritation. But that irritation passed quickly, because it had caught something I had missed.
In November 2025, NICE formalised five computer-aided detection tools — GI Genius, CAD EYE, ENDO-AID, EndoScreener, and MAGENTIQ-COLO, which are now conditionally recommended for NHS colonoscopy lists in England. Whether you are enthusiastic or sceptical, this guidance lands on our patch, and it is worth understanding what it actually says.
CADe systems watch the live video feed during withdrawal and draw a box around anything suspicious. That is essentially it. They do not remove polyps, they do not characterise them, and they do not make decisions — the endoscopist does all of that. Think of it less as artificial intelligence and more as a very vigilant second observer who never gets tired and never gets distracted by a bleep. The AI adds roughly 90 seconds to a standard procedure and helps to catch lesions that would otherwise walk out of the door with the patient.
The headline evidence comes from COLO-DETECT, published in The Lancet Gastroenterology & Hepatology in August 2024 — a pragmatic RCT run across 12 English NHS trusts, in real lists, with real patients. In the screening population, adenoma detection rate was 68% with GI Genius versus 61% with standard colonoscopy. In the symptomatic population — the one most of us spend our days in — it was 39% versus 28.5%. Sessile serrated lesion detection also improved significantly, from 8.3% to 11.6%.
NICE's endorsement is not a green light — it is a provisional one, with a four-year clock attached. The committee was candid: detecting more polyps is not the same as preventing more cancers. We do not yet have long-term outcome data showing that increased ADR from CADe actually reduces colorectal cancer incidence or mortality in the NHS population.
There is also the question of overdiagnosis. More polyps found means more surveillance colonoscopies, more pathology, more patient anxiety, and more strain on already overstretched endoscopy services. Not every extra adenoma we find would have become clinically significant. We should be honest about that trade-off.
Now what this means on Monday morning - If your unit is adopting one of these tools, a few things are worth keeping front of mind. AI does not rescue a sloppy withdrawal. Preparation quality, mucosal visualisation, and technique remain the primary variables. The units most likely to see real gains are those with high operator variability — where a distracted or fatigued endoscopist now has a prompt they would not otherwise have had.
For those of us in the MDT, expect a gradual increase in surveillance referrals and low-risk polyp findings. It is worth having that conversation with your endoscopy and pathology colleagues now, before the volume arrives.
And perhaps most importantly: the four-year evidence period is an opportunity. The NHS is, in theory, generating one of the largest real-world CADe datasets in the world. We should be involved in the research that comes from it — not as passive consumers of someone else's data, but as the clinicians who actually understand what we need to know.
My take? I think these tools are genuinely useful, and I think the NICE guidance is proportionate. But I also think we should resist the temptation to treat AI endorsement as the end of the conversation rather than the beginning of one. The question is not whether a machine can find polyps — it clearly can. The question is whether that finding translates into fewer of our patients dying of bowel cancer. We do not know that yet.
In the meantime: use the tools where you have them, contribute to the evidence where you can, and keep asking the questions that the enthusiasm around AI sometimes discourages. A second pair of eyes is welcome in any operating theatre. It does not replace the first.
Miss Kiran Altaf
Director of Equality, Diversity and Inclusion
