A second AI-guided look at the right colon found precancerous growths in 8.5% more patients

Colonoscopy is the rare screening test that also treats what it finds. The camera goes in, a polyp comes out, and a cancer that would have arrived in ten years does not.

It only works on what the operator sees. Adenomas in the right colon are likely to be missed during colonoscopy because of anatomical complexity and lesion characteristics, which is a polite way of saying that the far end of the bowel has deep folds and grows flat lesions that do not announce themselves.

Writing in the American Journal of Gastroenterology, researchers randomized 606 patients to a standard examination or to one extra step: a second forward-view pass of the right colon with real-time software watching. The right colon adenoma detection rate in the AI group was significantly higher than that in the non-AI group, 36.4% against 27.9%.

Why detection rate is the number that matters

Colonoscopy and flexible sigmoidoscopy are procedures that let your doctor look inside your rectum and colon, using scopes with a light and a tiny camera, and they let a doctor see problems such as inflamed tissue, ulcers, polyps, and cancer.

An adenoma is the polyp that matters: a growth that can turn into cancer if it stays. The proportion of patients in whom an endoscopist finds at least one is the field’s headline quality metric, and it is not an arbitrary one. Doctors with higher detection rates have patients who develop fewer cancers in the years between colonoscopies.

So an 8.5 percentage point difference is not a difference in paperwork. It is roughly one extra patient in twelve leaving with a growth removed that would otherwise have stayed in.

What the extra step actually is

The intervention has two components arriving at once, and the honest reading has to hold both.

In the AI group, a second forward-view examination of the right colon was performed using real-time AI assistance. Standard practice is to inspect on the way out, once. This adds a return trip up the right side with software flagging suspicious areas on screen as the operator goes.

The authors are explicit that they cannot separate the two ingredients: the independent contribution of AI assistance warrants further investigation. A second look with no software at all might have found some of those lesions. This trial did not include that arm.

Where the extra lesions came from

The pattern of the gain is what makes it believable.

The improvement was driven by increased detection of small, flat lesions and those in anatomically challenging regions, such as the ascending colon. Those are precisely the lesions a human eye loses on a single pass: no stalk, low profile, tucked behind a fold, seen for a fraction of a second while the scope moves.

If the AI arm had simply found more of everything, including the large obvious polyps, that would suggest a difference in how carefully the two groups were examined rather than a real detection effect. Finding more of the specific category known to be missed is the signature of the mechanism working as described.

The finding that cuts against the usual framing

Subgroup analyses showed consistent improvement in the right colon detection rate with AI assistance regardless of endoscopist experience.

The common story about medical AI is that it lifts the floor: it rescues the inexperienced and adds nothing for the expert. That is not what happened here. Experienced endoscopists improved too, which suggests the limitation being addressed is attentional rather than one of skill.

Nobody, however good, sustains perfect vigilance across a fold-by-fold inspection at the end of a long list. Software does not get bored.

What a health service has to weigh

The cost of this is minutes, and minutes are the scarcest thing in an endoscopy unit.

A second pass of the right colon lengthens every procedure in a system already rationing slots, and the trial reports the benefit without pricing the throughput. Colorectal cancer is cancer that develops in the tissues of the colon or rectum, it is common, and screening capacity is the binding constraint in most countries. Adding time per patient means fewer patients.

That calculation may still come out in favor. One extra adenoma found per twelve patients is a substantial return for a few minutes, particularly on the side of the colon where missed lesions do the most damage. But it is a calculation, not a foregone conclusion, and 606 patients at a handful of centers is where the evidence currently stops.