Comparing Gleason Pattern 4 Measurement Approaches on Prostate Biopsy Using Machine Learning: A Proof-of-Principle...

Area of Science:

• Digital pathology
• Machine learning in oncology
• Prostate cancer research

Background:

• Accurate quantification of Gleason Pattern 4 (GP4) is crucial for prostate cancer prognostication.
• Machine learning (ML) offers potential for objective and reproducible assessment of GP4 on digitized slides.
• Comparing different ML-based quantification methods is essential for clinical translation.

Purpose of the Study:

• To demonstrate the feasibility of using ML to quantify GP4 on digitized prostate biopsy slides.
• To compare the prognostic performance of various ML-based GP4 quantification approaches.
• To evaluate the predictive value of GP4 quantification for adverse pathology and biochemical recurrence.

Main Methods:

• A cohort of 726 patients with prostate cancer (Grade Group 2-4) was analyzed.
• Digitized biopsy slides were processed using the PAIGE-AI ML algorithm.
• GP4 extent was quantified using 15 distinct approaches, including linear and pixel-based metrics, assessing interfocal stroma handling.

Main Results:

• Fifteen different GP4 quantification approaches were evaluated, showing varied discriminatory performance.
• A pixel-counting method achieved the highest discrimination (AUC 0.648) for predicting adverse pathology.
• GP4 quantification outperformed Grade Group (GG) in predicting adverse pathology (AUC 0.627 vs 0.608) and biochemical recurrence.

Conclusions:

• ML can reliably quantify GP4 on digitized prostate biopsy slides using multiple measurement strategies.
• Pixel-based GP4 quantification demonstrates superior prognostic value compared to traditional GG.
• Further validation in larger cohorts is warranted to establish the optimal ML-based measurement approach for predicting oncologic outcomes.