A BREAKTHROUGH in prostate cancer diagnosis has emerged with the development of an interpretable machine learning model that reduces the need for invasive biopsies in patients with PSA levels below 20 ng/ml. The study highlights the potential of the XGBoost algorithm to provide highly accurate, non-invasive diagnostics.
Researchers analysed data from 655 patients who underwent transperineal prostate biopsies at the First Affiliated Hospital of Wannan Medical College between 2021 and 2023. Using advanced feature selection techniques, including LASSO regression, the team identified four key variables for model construction: age, prostate-specific antigen mass ratio (PSAMR), Prostate Imaging–Reporting and Data System (PI-RADS) score, and prostate volume.
Among the machine learning models tested, XGBoost outperformed others, achieving a receiver operating characteristic (ROC) score of 0.93 in the validation set, compared to logistic regression (0.89) and AdaBoost (0.90). When applied to test data, XGBoost maintained strong performance with an ROC score of 0.92, demonstrating its robustness and reliability.
The study also utilised SHapley Additive exPlanations (SHAP) to ensure the model’s interpretability, offering insights into the contributions of each variable to the diagnostic decision. Decision curve analysis revealed that patients with threshold probabilities ranging from 10% to 95% could benefit significantly from the model, reducing unnecessary biopsies and associated discomfort.
“This XGBoost-based model provides a non-invasive, high-accuracy tool for diagnosing prostate cancer in patients with low PSA levels, potentially transforming clinical workflows and improving patient outcomes,” the researchers noted.
The findings pave the way for broader adoption of AI-driven diagnostics in urology, combining accuracy, efficiency, and patient comfort.
Aleksandra Zurowska, EMJ
Reference
Li D et al. Using XBGoost, an interpretable machine learning model, for diagnosing prostate cancer in patients with PSA < 20 ng/ml based on the PSAMR indicator. Sci Rep. 2025;DOI: 10.1038/s41598-025-85963-7.
