Two studies demonstrate the benefits and limitations of artificial intelligence in colonoscopy

Editorial: https://www.acpjournals.org/doi/10.7326/M23-2022 

New research published in Annals of Internal Medicine explores the possible benefits and limitations of using computer assistance, or artificial intelligence, in detecting lesions and cancer during colonoscopy. An editorial accompanying these articles highlights that the current gap between randomized controlled trial performance and real-world performance likely reflects both differences in clinician behavior outside of trials and the complexity of real-world clinical environments.

AI not associated with improved detection of advanced colorectal neoplasias during colonoscopy

Abstract: https://www.acpjournals.org/doi/10.7326/M22-2619 

URL goes live when the embargo lifts

A randomized controlled trial found that colonoscopy assisted by computer-aided detection (CAD) was not associated with improved detection of advanced colorectal neoplasias.

Screening for colorectal cancer has greatly improved mortality rates due to greater detection of malignant and premalignant lesions. Systems relying on artificial intelligence using deep-learning technology have been linked to improved adenoma detection rates and reduce miss rate, but there are concerns that adenoma detection rates will continue to improve due to better detection of small polyps and nonadvanced adenomas, rather than detection of advanced and clinically significant lesions.

More than 3,000 persons with a positive fecal immunochemical test (FIT) were randomly assigned to colonoscopy with or without CAD to evaluate the contribution of CAD to colonoscopic detection of advanced colorectal neoplasias, adenomas, serrated polyps, and non-polypoid and right-sided lesions. FIT-positive patients were chosen because this group has the highest prevalence of advanced colorectal neoplasias, and therefore offers the best context for investigating the ability of computer aided detection to support the diagnosis of advanced colorectal neoplasias. The researchers found no significant difference in advanced colorectal neoplasia detection rate or the mean number of advanced colorectal neoplasias detected per colonoscopy between the two groups. Small effect was observed in increasing number of nonpolypoid lesions, proximal adenomas and small lesions of 5 mm or less, either colonic polyps in general, and adenomas and serrated polyps in particular, detected per colonoscopy. These findings suggest the need for additional research and more defined detection parameters in CAD before it can be integrated into routine clinical care.

Media contacts: For an embargoed PDF, please contact Angela Collom at [email protected]. To speak with the corresponding author Rodrigo Jover, MD, PhD, please email Carolina Mangas Residente at [email protected].

—————————-

Computer-assisted colonoscopy may increase polyp detection and removal but not cancer detection

Abstract: https://www.acpjournals.org/doi/10.7326/M22-3678  URL goes live when the embargo lifts  A review of 21 randomized trials found that the use of CAD for polyp detection during colonoscopy resulted in increased detection of polyps and polyp removal, but not detection of advanced adenomas, the types of polyps at higher risk of cancer progression. 

Artificial intelligence computer-aided detection (CADe) of colorectal neoplasia during colonoscopy may increase adenoma detection rates (ADRs) and reduce adenoma miss rates, but it may increase overdiagnosis and overtreatment of nonneoplastic polyps.

Researchers from Humanitas University conducted a systematic review and meta-analysis of 21 randomized controlled trials comprising 18,232 participants. The authors found the use of CADe was associated with a 55 percent relative risk reduction in miss rate of adenoma detection, but it was also associated with an increase in the removal of nonneoplastic polyps. The authors also report that CADe was also associated with a marginal increase in mean inspection time. The authors note that the studies mostly involved experienced gastroenterologists, and CADe programs may be more helpful to less experienced endoscopists. 

Media contacts: For an embargoed PDF, please contact Angela Collom at [email protected]. To speak with the corresponding author Marco Spadaccini, MD, please email [email protected].