Artificial Intelligence for Detection During Colonoscopy Works in the U.S. Too
Douglas K. Rex, MD, MASGE, reviewing Glissen Brown JR, et al. Clin Gastroenterol Hepatol 2021 Sep 13.
Artificial intelligence programs that detect lesions are called computer-aided detection (CADe) systems. Randomized controlled trials of CADe for colonoscopy have, thus far, mostly been performed outside the U.S. This study is a randomized tandem study involving 223 patients at 4 U.S. academic centers. Patients were randomized to undergo CADe first or high-definition white-light colonoscopy first, immediately followed by the other procedure.
The adenoma miss rate (AMR) was lower with the patients randomized to CADe colonoscopy first at 20.12% versus 31.25% in the patients randomized to high-definition white-light colonoscopy first. The miss rate for sessile serrated lesions was also lower with CADe first (7.14% vs 42.11%), and adenomas per colonoscopy were higher with CADe first (1.19 vs 0.90). First-pass ADR was 50.44% with CADe versus 43.64% without CADe, but this difference did not reach significance (study not powered for this difference). All the endoscopists were experienced, and it was noted in a post hoc assessment of videos that lesions missed by CADe were generally not exposed.
COMMENTThis study confirms that CADe is effective in the U.S., where detection rates for gastroenterologists are generally high. CADe should be viewed as an adjunct to detection, given the AMR with CADe was 20%. Excellent bowel preparation and complete mucosal exposure by meticulous examination techniques are still needed to optimize detection when using CADe.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
CITATION(S)
Glissen Brown JR, Mansour NM, Wang P, et al. Deep learning computer-aided polyp detection reduces adenoma miss rate: a U.S. multi-center randomized tandem colonoscopy study (CADeT-CS Trial). Clin Gastroenterol Hepatol 2021 Sep 13. (Epub ahead of print) (https://doi.org/10.1016/j.cgh.2021.09.009)
