Evidence Mounts for AI During Colonoscopy

Laird Harrison

August 08, 2022

Artificial intelligence (AI) improved the adenoma detection rate during colonoscopies to screen patients without symptoms of colorectal cancer, report researchers.

The AI slightly prolonged colonoscopies and appeared to benefit more experienced endoscopists than less experienced ones. However, the results are strong enough to encourage adoption of the new technology, said Joseph Sung, MD, PhD, dean of medicine at the Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore.

"The advantage of AI-assisted colonoscopy is quite well established," he told Medscape Medical News. "I don't see any reason not to use it, unless the equipment is very expensive."

Sung and colleagues published results from the large multicenter randomized controlled trial in Clinical Gastroenterology and Hepatology .

Testing AI Assistance in an Asymptomatic Population

Previous research has shown that conventional colonoscopies miss about a quarter of adenomas. Multiple AI systems have recently come on the market, promising to improve detection rates by overcoming human error in polyp detection.

Other comparison studies have already shown that AI-assisted colonoscopies have a higher rate of adenoma detection than conventional colonoscopies. But most of these studies have drawn their data from single centers or small populations or have not focused on people without symptoms.

To fill that gap, Sung and colleagues conducted their study in six university-affiliated endoscopy centers in Hong Kong, Beijing, Inner Mongolia, Jilin, and Xiamen, China.

They administered colonoscopies to 3059 asymptomatic persons aged 45-75 years old were eligible for screening between November 2019 and August 2021. Some had undergone fecal immunochemical tests, but they were offered colonoscopies regardless of those results.

The researchers randomly assigned participants to undergo either AI-assisted or conventional colonoscopies. The computer marked lesions it detected by a blue tracking box on the high-definition monitor of the endoscopy system in real time. The AI algorithm was trained on colonoscopy images from Zhongshan Hospital, affiliated with Xiamen University.

To standardize the procedure, the endoscopists were not allowed to use electronic image enhancing functions for polyp detection, only for polyp characterization.

More Adenomas Detected

In an intention-to-treat analysis, the adenoma detection rate (ADR) (defined as the proportion of patients with at least one adenoma detected during the colonoscopy) and the mean number of adenomas per colonoscopy (APC) were both significantly higher for the AI group than for the conventional group.


AI-Assisted Colonoscopy (n = 1519)

Conventional Colonoscopy (n = 1540)

P Value

Overall ADR (%)



<. 001

ADR for advanced adenomas (%)




ADR for nonadvanced adenomas (%)







< .001

The AI-assisted colonoscopies also detected adenomas ≥ 10 mm, adenomas < 5 mm, nonpedunculated adenomas, and adenomas in both the proximal and distal colon at a statistically significant higher rate. Findings were similar in a per protocol analysis.

Previous studies showed that AI boosted the detection of small adenomas, but these are less likely than large ones to develop into cancer.

"The unique finding is that even larger adenomas are starting to be picked up better with AI," said Sung.

There was no statistical difference between the two groups in detection of colorectal cancer, sessile serrated lesions, hyperplastic polyps, or pedunculated adenomas.

Both the more experienced and the less experienced endoscopists benefited from assistance with AI. But the ADR for the more experienced endoscopists jumped from 32.8% with conventional colonoscopy to 42.3% with AI-assisted colonoscopy, whereas it only rose from 32.1% to 37.55% for the less experienced endoscopists.

"That is a part that I find it a bit difficult to understand," said Sung, because in theory, AI should be more helpful to less experienced endoscopists. But he noted that the cutoff between more and less experienced endoscopists in the study was 5000 procedures.

"What they call less experienced colonoscopists, in fact, are fairly experienced too. Therefore, I'm not so sure about the significance of this finding," he said.

Both the intubation and withdrawal times were less than half a minute longer with the AI procedure.

Next Steps

The study makes an important contribution to understanding AI-assisted colonoscopies, said Cesare Hassan, MD, PhD, an associate professor at the Humanitas Clinical and Research Center Endoscopy Unit in Rozzano, Italy, who was not involved in it.

Hassan noted that it was a large, multicenter study with a well-characterized population. "I'm sure this is the best paper ever done in this field," he said.

Its only weakness is that the endoscopists knew whether they were using the AI, he said, but it is very hard to create a study where they could be blinded to that.

The next step in studying AI colonoscopy is to determine whether it is actually preventing cancer, Sung said. "That is the ultimate advantage."

Hassan said he and his colleagues were planning a 10-year trial with that goal in mind.

In addition, studies are under way to determine whether these systems can help separate adenomas from hyperplastic polyps.

"Every year, in every hospital, we spend a lot of money doing histology to examine whether this is a hyperplasic polyp or an adenomatous polyp," said Sung. "If you can just ignore them or just remove the polyps and discard them with confidence, then it would be a distinct advantage, and the patient wouldn't need to be followed up so frequently."

Sung and Hassan report no relevant financial relationships .

Clin Gastroenterol Hepatol. Published online July 18, 2022. Full text

Laird Harrison writes about science, health and culture. His work has appeared in national magazines, in newspapers, on public radio and on websites. He is at work on a novel about alternate realities in physics. Harrison teaches writing at the Writers Grotto. Visit him at www.lairdharrison.com or follow him on Twitter: @LairdH

For more news, follow Medscape on Facebook, Twitter, Instagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.