[Using the PathVision.ai model of medical decision support in morphological diagnostics of prostate cancer in biopsy...

The implementation of prostate-specific antigen (PSA) screening over the past decade has led to a significant increase in the number of primary prostate biopsies, thereby imposing a growing workload on pathology departments. Globally, considerable efforts are being directed towards developing artificial intelligence (AI) models capable of assisting pathologists in the morphological diagnosis of prostate cancer (PCa).

Develop an AI-based model for detecting PCa in biopsy samples and compare its diagnostic performance with pathologists with varying levels of experience.

A training dataset comprising 470 core biopsy specimens of prostate tissue from 86 patients was utilized. The AI model employed a U-Net neural network architecture. For testing, a separate dataset of 282 scanned specimens, not included in model training, was separated in 2 sets - 1 and 2 (with 141 slides in each). This dataset consisted of 81 specimens with non-neoplastic prostate tissue and 201 specimens with adenocarcinoma structures. The study included 20 pathologists with varying levels of experience from 12 medical institutions in the Russian Federation, who evaluated each case on the «PathVision.ai» platform. The diagnostic performance and agreement between the AI model and pathologists were assessed using comparative statistical analysis.

The AI-model demonstrated high diagnostic performance, with sensitivity of 0.99 [95% CI: 0.97-1.00], specificity of 0.93 [95% CI: 0.84-0.97], and accuracy of 0.98 [95% CI: 0.95-0.99] for detecting PCa. These metrics surpassed those of junior pathologists (sensitivity: 0.92 [95% CI: 0.91-0.94]; specificity: 0.89 [95% CI: 0.86-0.91]; accuracy: 0.91 [95% CI: 0.90-0.93]) and were comparable to those of experienced pathologists (sensitivity: 0.93 [95% CI: 0.91-0.95]; specificity: 0.94 [95% CI: 0.91-0.96]; accuracy: 0.94 [95% CI: 0.92-0.95]).

Developed model «PathVision.ai» in pilot study demonstrated high sensitivity, specificity and accuracy in the diagnosis of PCa in biopsy samples.