AI-Assisted Colonoscopy Linked to Increased Benign Lesion Removal

PHILADELPHIA — Artificial intelligence–assisted colonoscopy (AIAC) with computer-aided detection (CADe) technology may improve adenoma detection rate (ADR), but it’s also associated with higher detection and removal of non-neoplastic lesions, according to a study presented at the American College of Gastroenterology (ACG) 2024 Annual Scientific Meeting.

In particular, AIAC led to a statistically and clinically significant increase in the proportion of exams that detected lesions that after resection were all found to be benign, compared with unassisted colonoscopy.

“The potential implications include increased procedural risks, as well as costs, such as pathology costs and other healthcare expenditures, without any additional colorectal cancer prevention benefit,” said lead author Tessa Herman, MD, chief resident of internal medicine at the University of Minnesota and Minneapolis Veterans Affairs Health Care System.

In a previous implementation trial at the Minneapolis VA Medical Center, Herman and colleagues compared ADR between a group of patients undergoing AIAC and a historical cohort of patients who had non–AI-assisted colonoscopy.

In this subsequent study, the research team conducted an ad-hoc analysis of data from the previous trial to determine the proportion of colonoscopies for screening, surveillance, and positive fecal immunochemical tests which detect lesions that after resection are all found to be benign. They excluded colonoscopies conducted for diagnostic indications or inflammatory bowel disease, as well as incomplete colonoscopies, and for those with inadequate bowel preparation.

Overall, they studied 441 non-AIAC colonoscopies (between November 2022 and April 2023) and 599 AIAC colonoscopies (between May 2023 and October 2023). The groups were balanced, and there were no significant differences in patient demographics, endoscopists, AI technology, procedure time, or average number of polyps detected.

In the non-AIAC cohort, 37 cases (8.4%) had polypectomies that revealed only benign lesions, as compared with 74 cases (12.4%) in the AIAC cohort. The most common resected lesions were benign colonic mucosa, lymphoid aggregates, and hyperplastic polyps.

Applied to the 15 million colonoscopies conducted in the United States per year, the findings indicate that full adoption of AIAC could result in about 600,000 more colonoscopies in which only benign, nonadenomatous lesions are removed, compared with traditional colonoscopy, Herman said.

More study of AIAC is needed, said Daniel Pambianco, MD, managing partner of GastroHealth-Charlottesville and the 2023 ACG president. “This technology is in a fledging stage, and the more data we have, the more helpful it’ll be to know if we’re removing the right lesions at a better rate.”

“There’s a hope that assistance will improve detection, removal of polyps, and ultimately, colon cancer,” added Pambianco, who co-moderated the session on colorectal cancer prevention.

Future longitudinal studies should monitor both ADR and benign lesion resection rates with AIAC, and modeling studies could determine the benefits and costs of the technology, Herman said. In addition, development of hybrid CADe and computer-aided diagnosis systems could mitigate concerns about excessive benign lesion resection with AI tools, she said.

Clinicians already are able to find colon mucosa that are polypoid or lymphoid aggregates during colonoscopy without AI assistance, said the session’s co-moderator, Sita Chokhavatia, MD, a gastroenterologist with Valley Medical Group in Ridgewood, New Jersey. 

“Instead, we need a tool that can help us to not remove these polyps that are not neoplastic,” she said. “With future developments, we may be able to take it to the next step where the algorithm tells us that it’s benign and not to touch it.”

The study was named an ACG Newsworthy Abstract. Herman, Pambianco, and Chokhavatia reported no relevant disclosures.