An Artificial Intelligent System for Prostate Cancer Diagnosis in Whole Slide Images

Area of Science:

• Digital Pathology
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Increasing demand for automated diagnostic tools in prostate cancer assessment.
• Whole slide imaging (WSI) offers a digital platform for pathology analysis.
• Need for accurate and efficient methods to detect cancer and related features in prostate biopsies.

Purpose of the Study:

• To develop and validate a machine learning system for prostate cancer assessment using WSI.
• To enable detection of perineural invasion and measurement of cancer portion for clinical reporting.
• To create a robust AI tool for digital pathology workflows.

Main Methods:

• A three-stage system: tissue detection, classification, and slide-level analysis.
• Division of WSI into patches for analysis.
• Utilized traditional machine learning for tissue detection and deep learning for cancer classification.
• Trained and validated on 2340 H&E stained slides with expert pathologist annotations.

Main Results:

• Achieved 99.53% accuracy in tissue detection (99.78% sensitivity, 99.12% specificity).
• Cancer tissue classification accuracy reached 92.80% at 5x magnification (92.61% sensitivity, 99.25% specificity).
• Performance varied with magnification, with lower accuracies at higher magnifications (e.g., 84.71% at 20x).

Conclusions:

• The developed machine learning system demonstrates high accuracy in tissue detection and cancer classification from WSI.
• The system shows potential for aiding pathologists in prostate cancer diagnosis and reporting.
• Further optimization may be needed to improve performance across different magnifications.