PROSTATE cancer diagnosis may be improved through a non-invasive machine learning approach using urinary extracellular vesicle physical parameters, according to new research evaluating five predictive algorithms in 222 participants.
Urinary Extracellular Vesicles and Prostate Cancer Diagnosis
Researchers investigated whether the physical characteristics of urinary extracellular vesicles could support earlier and more accurate prostate cancer diagnosis while reducing reliance on invasive biopsy procedures. Although extracellular vesicle biomarkers have shown promise in recent years, many detection methods remain expensive and difficult to implement routinely in clinical practice.
Urine samples were collected from 222 eligible participants, and urinary extracellular vesicles were isolated using a commercial kit. Concentration and size characteristics were then measured using nanoparticle tracking analysis. These physical parameters formed the basis of five machine learning models designed to distinguish patients with prostate cancer from benign controls.
Analysis revealed clear differences in urinary extracellular vesicle profiles between groups. Patients with prostate cancer showed a significantly greater proportion of 30–150 nm vesicles and a smaller overall particle size compared with individuals with benign prostatic hyperplasia (p<0.001). These findings suggested that extracellular vesicle size distribution may hold predictive value for prostate cancer diagnosis.
Machine Learning Model Demonstrates Strong Performance
Among the evaluated algorithms, the eXtreme Gradient Boosting model achieved the strongest overall diagnostic performance. In distinguishing prostate cancer from benign prostatic hyperplasia, the model produced area under the receiver operating characteristic curve values of 0.934 in the training cohort; 0.864 in the testing cohort.
The investigators also evaluated model performance using learning curves, calibration curves, and decision curve analysis. Decision curve analysis showed that the model generated greater clinical net benefit across a threshold probability range of 0–60% when compared with routine clinical measures.
In addition, SHapley Additive exPlanations analysis was used to visualise the contribution of individual predictors within the model, improving interpretability of the machine learning framework.
Clinical Utility in Prostate Cancer Diagnosis
The study demonstrated that the machine learning model outperformed conventional diagnostic indicators, including prostate specific antigen and prostate specific antigen density, about both diagnostic efficacy and clinical utility.
Researchers concluded that combining urinary extracellular vesicle physical parameters with machine learning may provide a practical adjunctive tool for prostate cancer diagnosis. The non-invasive strategy could help clinicians identify patients more accurately while potentially reducing unnecessary prostate biopsies.
Reference
Jiang K et al. Development and validation of a noninvasive machine learning model using urinary extracellular vesicle physical parameters for prostate cancer diagnosis. Sci Rep. 2026;DOI: https://doi.org/10.1038/s41598-026-54920-3.
Featured image: auremar on Adobe Stock
