CADx Program Met PIVI Thresholds for Diminutive Rectosigmoid Polyps but Not in Proximal Colon
Douglas K. Rex, MD, MASGE, reviewing Rondonotti E, et al. Endoscopy 2022 May 13.
A computer-assisted diagnosis (CADx) program is artificial intelligence (AI) software that predicts histology. In a prospective evaluation of the Fujifilm CADx program (CAD EYE, Fujifilm Co, Tokyo, Japan), expert and nonexpert endoscopists categorized polyps as adenomas or nonadenomas, then used the CADx program to categorize the polyps.
There were 596 diminutive rectosigmoid polyps (DRSPs) retrieved for histology in 389 patients. A high-confidence AI-assisted optical diagnosis was reached in 92.3% of these polyps. The negative predictive value (NPV) of the AI-assisted diagnosis was 91%, with AI-assisted optical diagnosis accuracy lower for nonexperts (82.3%) than experts (91.9%).
The final diagnosis after AI assistance delayed postpolypectomy surveillance in 2.4% of patients when the surveillance interval guideline from the European Society of Gastrointestinal Endoscopy was followed and 5.3% of patients when the U.S. Multi-Society Task Force on Colorectal Cancer guideline was used. Nonexperts improved over time so that the NPV of the last 50 DRSPs was similar to that of experts.
The NPV for adenomatous histology was lower for diminutive polyps proximal to the sigmoid at 72.4%. The driver appeared to be a higher prevalence of adenomas in the proximal colon at 78% versus 43% in the rectosigmoid.
COMMENTThe important observation here is that CADx is imperfect for predicting histology, but it does perform within parameters established by the American Society for Gastrointestinal Endoscopy for the Preservation and Incorporation of Valuable endoscopic Innovations (PIVI)-1 thresholds in the distal colon. CADx also helps teach nonexperts how to meet the PIVI thresholds for AI-assisted optical diagnoses.
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)
Rondonotti E, Hassan C, Tamanini G, et al. Artificial intelligence assisted optical diagnosis for resect and discard strategy in clinical practice (Artificial intelligence BLI Characterization; ABC study). Endoscopy 2022 May 13. (Epub ahead of print) (https://doi.org/10.1055/a-1852-0330)
