Development and Validation of an Artificial Intelligence-Powered Platform for Prostate Cancer Grading and...

Area of Science:

• Urological pathology
• Medical artificial intelligence
• Cancer diagnostics

Background:

• The Gleason grading system is crucial for prostate cancer prognosis but suffers from interobserver variability.
• This variability negatively impacts risk assessment and clinical management of prostate cancer patients.

Purpose of the Study:

• To evaluate the impact of an artificial intelligence (AI)-assisted approach on prostate cancer grading and quantification.
• To assess the accuracy of an AI platform in detecting and grading prostate cancer from biopsy slides.

Main Methods:

• A deep convolutional neural network-based AI platform was developed and validated using 1000 digital whole-slide images from 589 men with prostate cancer.
• Three urological pathologists graded and quantified prostate cancer manually and with AI assistance.
• The AI platform's accuracy was assessed at the patch-pixel and slide levels.

Main Results:

• The AI system achieved high accuracy in detecting prostate cancer (AUC 0.92) and near-perfect agreement with a training pathologist for detection (κ=0.97) and grading (κ=0.98).
• AI-assisted grading and quantification significantly improved concordance among pathologists (e.g., 90.1% agreement with AI vs. 84.0% manually).
• The AI-assisted method led to significantly higher interobserver agreement (e.g., κ=0.92 with AI vs. 0.76 manually).

Conclusions:

• An AI-powered platform demonstrates high accuracy and efficiency in detecting, grading, and quantifying prostate cancer.
• The AI platform significantly reduces interobserver variability in histopathological evaluation.
• AI assistance holds potential to transform prostate cancer histopathology, improving risk stratification and clinical management.