Detection of latent cancer in prostate biopsy using Paige Prostate^® (prostate pathology diagnostic support artificial intelligence; AI): Revaluation of diagnostic accuracy based on Japan’s first trial experience

Abstract

The U.S. Food and Drug Administration (FDA) has approved Paige Prostate (Paige.US), an artificial intelligence (AI) technology that uses images to identify the area most likely to contain occult cancer in prostate biopsies. We evaluated the diagnostic accuracy of this system for detecting prostate cancer. The subjects were 108 biopsies from 10 patients with cancers, and the results of Paige Prostate analysis and HE staining diagnosis by humans were evaluated to determine: 1) Prostate’s biopsy diagnostic accuracy, 2) number of detected cancer areas, 3) Gleason pattern, and 4) reproducibility of the analysis. Prostate’s biopsy by Paige Prostate diagnostic accuracy is 100% sensitive and 82.3% specific. The concordance rate between Paige Prostate and human microscopic observation on Gleason pattern was 64.7% (44/68 areas: Patterns 3 and 4), whereas the concordance rate was 35.3% (24/68 areas: Pattern 4 glomerular-like structure, Pattern 5 signet ring cell subtype), indicating a tendency to misrecognize rare histological subtypes. Paige Prostate is a useful system that does not overlook cancer areas, In the future, further learning in non-cancerous areas and rare histological subtype areas will lead to practically useful pathological diagnosis AI with higher diagnostic accuracy.