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

Conventional core needle biopsy for prostate cancer diagnosis can lead to diagnostic uncertainty and complications, prompting exploration of alternative risk assessment approaches that use clinical and imaging features. Our aim was to evaluate the effectiveness of machine learning (ML) models in reducing unnecessary biopsies.

We conducted a retrospective analysis of data for 1884 patients across two academic centers who underwent prostate magnetic resonance imaging and biopsy between 2016 and 2020 or 2004 and 2011. Twelve ML models were assessed for prediction of clinically significant prostate cancer (csPCa; Gleason grade group ≥2) using combinations of clinical features, including patient age, prostate-specific antigen level and density, Prostate Imaging-Reporting and Data System/Likert score, lesion volume, and gland volume. The models were trained and validated using a tenfold split for intrasite, intersite, and combined-site data sets. Model effectiveness was evaluated using the area under the receiver operating characteristic curve and decision curve analysis.

The best-performing ML model would reduce the number of biopsies by 13.07% at a false-negative rate of 1.91%. Performance was consistent across sites, although the study is limited by the small number of centers and the absence of specific clinical data.

ML-enhanced clinical models provide an effective and generalizable approach for prediction of csPCa using standard clinical data. These models allow personalized risk assessment and follow-up, support clinical decision-making, and improve workflow efficiency.

Models that are enhanced by machine learning can predict the severity of prostate cancer and help doctors in tailoring treatments for individual patients. This approach can simplify health care decisions and improve clinical efficiency.