Development and validation of a noninvasive machine learning model using urinary extracellular vesicle physical parameters for prostate cancer diagnosis

Urinary extracellular vesicles (uEVs) are promising biomarkers for prostate cancer (PCa). Although novel uEV-based biomarkers have advanced early diagnosis, their detection processes remain cumbersome and cost-prohibitive. The physical parameters of uEVs offer untapped predictive potential for PCa, providing a more convenient, rapid, and cost-effective diagnostic approach. This study aims to construct a predictive model integrating uEV physical parameters with machine learning (ML) algorithms to enhance the early diagnosis of PCa. Urine samples were collected from 222 eligible participants. uEVs were isolated using a commercial kit, and their concentration and size parameters were quantified via nanoparticle tracking analysis (NTA). Utilizing these physical parameters, five ML algorithms were trained and evaluated to identify the optimal diagnostic model for PCa. Model performance was systematically assessed using the area under the receiver operating characteristic curve (AUC), learning curves, calibration curves, and decision curve analysis (DCA). Additionally, SHapley Additive exPlanations (SHAP) were employed to visualize the contributions of key predictors. Analysis of uEV physical parameters revealed that, compared to benign controls, PCa patients exhibited a significantly higher proportion of 30–150 nm uEVs and a smaller overall particle size (p < 0.001). Among the evaluated algorithms, the eXtreme Gradient Boosting (XGBoost) model demonstrated superior performance. For discriminating PCa from benign prostatic hyperplasia (BPH), the XGBoost model achieved AUC values of 0.934 and 0.864 in the training and testing cohorts, respectively. DCA demonstrated that the XGBoost model yielded a higher clinical net benefit across a threshold probability range of 0–60%. Overall, the diagnostic efficacy and clinical utility of the XGBoost model significantly outperformed routine clinical parameters, including prostate-specific antigen (PSA) and prostate-specific antigen density (PSAD). This study successfully developed and validated a noninvasive ML-based diagnostic model utilizing the physical parameters of uEVs. This approach serves as a preliminary adjunctive tool to assist clinicians in accurately identifying prostate cancer, thereby potentially reducing the incidence of unnecessary prostate biopsies.