Clinically informed intermediate reasoning enables generalizable prostate cancer prognostication through machine...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Machine learning (ML) shows potential in medical image classification but struggles with generalizability.
• Accurate pre-surgical prognostication of prostate cancer is crucial for effective treatment planning.

Purpose of the Study:

• To develop a data-efficient and generalizable machine learning framework for prostate cancer prognostication from biopsy specimens.
• To improve upon the limitations of the current Gleason grading system.

Main Methods:

• Utilized versatile feature extraction from whole-mount histopathology images.
• Incorporated a clinically informed intermediate reasoning step.
• Validated the pipeline across multiple institutions and specimen types to address dual-domain shifts.

Main Results:

• Achieved consistent external validation, demonstrating robust generalizability.
• The developed prognostic approach outperformed the traditional Gleason grading system.
• The framework proved equitable, interpretable, and clinically applicable.

Conclusions:

• The proposed pipeline enables data-efficient, robust, and generalizable pre-surgical prognostication of prostate cancer.
• This interpretable framework supports actionable clinical decisions for prognosis and treatment planning, even with limited data.
• The approach offers a reliable alternative to the Gleason grading system in real-world clinical settings.