A computer-aided detection (CADe) software for colorectal polyps has a clinical accuracy of 96.5% and is noninferior to detection by experienced endoscopists, according to a study published in Gut.
The new CADe software for colorectal polyp detection incorporates a temporal algorithm and was validated and compared against a composite reference standard of experienced endoscopists and CADe detection, with real-time unblinding of CADe findings by a second observer. The second observer was trained in polyp recognition but was not necessarily a trained endoscopist and monitored the colonoscopy on the CADe tool.
A total of 606 polyps were endoscopically detected in 295 patients—574/606 (94.7%) were found by both the endoscopist and CADe (true positives), and 11/606 (1.8%) were additionally detected by the CADe and confirmed by the endoscopist after real-time unblinding, and initially missing them. Diagnostic accuracy was 96.5% for the CADe and 98.2% for the endoscopists, resulting in a noninferiority of the CADe with a margin of 5% (P <.001; 95% CI, –0.0378 to 0.0047).
The CADe missed 21 polyps, with 6 (28.8%) adenomas. The CADe had a polyp detection rate of 62%, and an adenoma detection rate (ADR) of 33%. The system had 1.4 obvious false-positive detections per minute clean withdrawal time, according to the study authors.
“In a challenging setting with in vivo endoscopy, our system was not inferior to endoscopists with a high ADR (>35%),” stated the study authors. “Unexpectedly, the miss rate due to failure of polyp recognition in high detectors is in fact very low (<2%) and the integration of temporal information in our CADe system resulted in a low rate of false-positive findings.”
The study design for the validation of the CADe system for polyp detection with real-time unblinding of the endoscopist allows “a better assessment of the additional clinical value of our CADe since the reference standard is based on a computer versus composite human–computer ground truth,” according to the investigators.
Among several study limitations, the endoscopist was not blinded to the CADe presence in the room, and there is potential bias of the second observer giving feedback from the CADe system to the endoscopist.
“Our study design with real-time unblinding allows [us] to set a new gold standard for testing the real additional clinical value of CADe over human detection in a real-life clinical setting,” the researchers concluded.
Disclosures: Some of the authors reported affiliations with medical technology companies. Please see the original reference for a full list of disclosures.
Sinonquel P, Eelbode T, Hassan C, et al. Real-time unblinding for validation of a new CADe tool for colorectal polyp detection. Gut. 2021;70(4):641-643.