乌鸦传媒

An artificial intelligence-based system missed fewer adenomas and found more first-pass adenomas per colonoscopy than high-definition white light (HDWL) colonoscopies, a multicenter randomized trial found.

In an analysis of 223 patients, the adenoma miss rate among those who initially underwent a computer aided detection (CADe) colonoscopy was significantly lower than those who initially underwent HDWL colonoscopy (20% vs 31%; OR 1.80, 95% CI 1.07-3.02, P=0.024), reported Jeremy R. Glissen Brown, MD, of Harvard Medical School in Boston, and colleagues.

The CADe-first group also had significantly more first-pass adenomas per colonoscopy (1.19 vs 0.90, P=0.032, respectively) and significantly higher first-pass adenoma detection rates (50% vs 43%, P=0.309), the authors wrote in .

"There's really no question artificial intelligence is a potent tool in its widespread use," Neil Hyman, MD, of the University of Chicago, who was not involved in this study, explained to 乌鸦传媒. "This is a very logical extension of this technology...to my knowledge there have been some anecdotal reports on data presented suggesting benefits of colorectal polyp detection with it."

Glissen Brown and colleagues noted that the effect of CADe during screening and surveillance colonoscopy has not been studied before in a U.S. population. The CADe system has been evaluated in only a few Chinese trials, where it utilizes a deep learning network founded by , they added.

From May 2019 to November 2020, 232 patients at four U.S. medical centers were randomized 1:1 to HDWL colonoscopy-first (n=116) or to CADe colonoscopy-first (n=116), followed by the opposite procedure "in tandem fashion."

The primary outcome assessed the adenoma miss rate, calculated from the second colonoscopy.

Patients were sedated and bowel preparation procedures (both groups scored a 9 on the Boston bowel preparation scale) were completed prior to colonoscopies. Patients had a mean age of about 61, 55% were men, and about two-thirds were Caucasian. About 60% of patients had indications for colorectal cancer screening, while the primary indication for the remaining patients was post-polypectomy surveillance.

Interestingly, there were significantly fewer sessile serrated lesions (SSLs) missed in the CADe-first group versus the HDWL group (7.1% vs 42%, respectively, P=0.048).

"HDWL did miss more adenomas than the CADe system, but the most impressive thing is the sessile serrated lesions that are much harder to identify," Hyman added. "Most of us who do a lot of colonoscopies find the sessile serrated lesions are harder to detect and far easier to miss than an adenoma."

Glissen Brown also highlighted the potential benefits of CADe in reducing the SSL miss rate, saying this was to show this potential benefit.

"Miss rates for flat adenomas and serrated polyps are higher than for traditional polyp-shaped adenomas and SSLs contribute disproportionately to the histology of post-colonoscopy colorectal cancers," Glissen Brown told 乌鸦传媒.

A multivariate regression analysis found three factors tied to missed adenomas: HDWL-first (OR 1.88, P=0.021), age 65 or younger (OR 1.74, P=0.045) and right colon versus other location (OR 1.79, P=0.044).

"Sessile serrated lesions are more often on the right side of the colon, more subtle, and harder to find in a colonoscopy," Hyman said. "The ability to identify sessile serrated lesions could go a long way."

The analysis had several limitations, the authors acknowledged. Patients were blinded, but not endoscopists, who were highly experienced with highly established baseline adenoma detection rates. Findings may not be generalizable to less experienced endoscopists.

"This is a prospective study but is the correct experimental design added to this learning system, the whole reason we do colonoscopy is to find the adenomas due to cancer and anything that increases the [adenoma detection rate] is beneficial," Hyman said.

Comment