Prospective Comparison of Pathology and Computer-Aided Optical Diagnosis for Polyp Characterization Using Expert...

Histopathology is the accepted gold standard for classifying colorectal polyps. However, the complex pathway from initial polyp detection and resection to final histological diagnosis involves multiple steps (retrieval, embedding, sectioning, interpretation), each potentially introducing diagnostic errors. We aimed to investigate the diagnostic accuracy of autonomous computer-aided optical polyp diagnosis (CADx) versus histopathology, using expert consensus as the reference standard.

In this prospective study, three internationally recognized expert endoscopists independently reviewed colonoscopy images and videos of diminutive (≤ 5 mm) polyps assessed by autonomous CADx. Unanimous, high-confidence expert agreement was defined as the reference standard ('ground truth'), assumed superior due to absence of resection or specimen-management related artifacts. The primary outcome was the proportion of specimens with potential incorrect histopathology diagnosis. Secondary outcomes included comparative accuracy, sensitivity, and specificity of autonomous CADx and histopathology against the expert-defined reference standard.

Among 510 patients, experts provided unanimous, high-confidence diagnoses for 226 diminutive (≤ 5 mm) colorectal polyps, serving as the reference standard. The proportion of potentially incorrect pathology diagnoses was 23% (95% CI: (17.8,29.2)). Autonomous CADx achieved a higher accuracy of 91.2% (95% CI: 86.5-94.4%), compared with pathology at 77.0% (95% CI: 70.8-82.2%; p < 0.001).

These findings highlight that histopathological diagnosis after polyp resection is not infallible and can introduce diagnostic errors in a relevant proportion of cases. CADx diagnosis, obtained during the procedure and unaffected by specimen handling artifacts, may prevent some of these errors, suggesting it could offer superior accuracy for polyp characterization in certain cases.