Prostate Cancer Risk Stratification by Digital Histopathology and Deep Learning.

Area of Science:

• Oncology
• Digital Pathology
• Machine Learning

Background:

• Prostate cancer (PCa) is heterogeneous, necessitating accurate risk assessment for treatment planning.
• Current risk stratification relies on parameters like Gleason grading, which has significant interobserver variability.
• Objective tools are needed to improve the accuracy of PCa risk stratification.

Purpose of the Study:

• To determine if machine learning (ML)-driven histopathology image analysis can improve risk stratification for PCa.
• To develop and evaluate a deep learning model combining clinicopathologic data with histopathology images for PCa risk assessment.

Main Methods:

• A deep learning model was developed using hematoxylin and eosin- and Ki-67-stained histopathology images and clinicopathologic data.
• The model was trained and tested on a dataset of 502 treatment-naïve PCa patients using five-fold cross-validation.
• Performance was evaluated using the concordance index, comparing it against Gleason grading and CAPRA-S models.

Main Results:

• The ML-based convolutional neural network model demonstrated superior performance in risk stratification compared to Gleason grading and CAPRA-S.
• The model correctly reclassified 3.9% of low-risk patients to high-risk and 21.3% of high-risk patients to low-risk.
• This indicates improved accuracy in identifying patient risk categories.

Conclusions:

• ML-driven histopathology image analysis serves as an objective tool for PCa risk stratification.
• The proposed digital pathology risk classification can aid in guiding adjuvant therapy decisions, such as radiotherapy, after radical prostatectomy.
• Further validation on larger cohorts is recommended to solidify its clinical utility.