A Novel Machine Learning-based Predictive Model of Clinically Significant Prostate Cancer and Online Risk Calculator.

Area of Science:

• Urology
• Oncology
• Medical Informatics

Background:

• Prostate cancer detection relies on accurate risk stratification.
• The Prostate Imaging-Reporting and Data System (PI-RADS) score and PSA density (PSAD) are key metrics.
• Predictive models can enhance diagnostic accuracy for clinically significant prostate cancer (csPCa).

Purpose of the Study:

• To develop and validate a machine learning model for predicting csPCa.
• The model integrates PI-RADS scores, PSAD, and clinical variables.
• To assess the model's performance against individual predictors.

Main Methods:

• A multi-national cohort of 1272 patients undergoing prostate biopsy was analyzed.
• Data included age, BMI, PSA, prostate volume, PI-RADS, and biopsy history.
• Lasso, XGBoost, and LightGBM models were trained and validated internally and externally.

Main Results:

• All models achieved high ROC-AUC values (0.830-0.851).
• LightGBM demonstrated superior performance with ROCs of 0.851 (test set) and 0.818 (external dataset).
• PI-RADS score, PSAD, and prior biopsy history were the most influential variables.

Conclusions:

• A robust machine learning model for csPCa detection was developed.
• The integrated model significantly outperformed individual predictors.
• The model showed strong internal and external validation with good calibration.