Comparative Performance of Machine Learning Models in Reducing Unnecessary Targeted Prostate Biopsies

Area of Science:

• Urology
• Oncology
• Medical Imaging

Background:

• Conventional prostate cancer diagnosis via core needle biopsy presents challenges, including diagnostic uncertainty and potential complications.
• There is a growing need for advanced risk assessment methods that integrate clinical and imaging data.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in predicting clinically significant prostate cancer (csPCa).
• To determine if ML models can reduce the rate of unnecessary prostate biopsies.

Main Methods:

• A retrospective analysis of 1884 patients who underwent prostate MRI and biopsy was performed.
• Twelve ML models were developed and validated using clinical data (age, PSA, imaging scores, volumes) to predict csPCa (Gleason grade group ≥2).
• Model performance was assessed using area under the receiver operating characteristic curve and decision curve analysis.

Main Results:

• The top-performing ML model demonstrated a potential reduction in biopsies by 13.07% with a 1.91% false-negative rate.
• Model performance remained consistent across different academic centers.
• The study's limitations include a small number of participating centers and lack of detailed clinical data.

Conclusions:

• ML-enhanced clinical models effectively predict csPCa using standard clinical data, offering a generalizable approach.
• These models facilitate personalized risk assessment, support clinical decision-making, and enhance workflow efficiency in prostate cancer diagnosis.
• The integration of ML can lead to more tailored patient management and improved healthcare outcomes.