A non-invasive machine learning approach using urinary extracellular vesicle physical parameters could improve prostate cancer detection, according to a study of 222 participants.
Researchers analyzed urine samples from eligible participants, isolating extracellular vesicles with a commercial kit. Using nanoparticle tracking analysis, they measured concentration and size characteristics. Five machine learning models were then applied to differentiate prostate cancer patients from benign controls.
Patients with prostate cancer had a significantly higher proportion of 30-150 nm vesicles and smaller overall particle size compared to those with benign prostatic hyperplasia. The eXtreme Gradient Boosting model demonstrated the strongest performance, with AUC values of 0.934 in the training cohort and 0.864 in the testing cohort.
The model outperformed conventional markers like PSA and PSA density, offering greater clinical net benefit across a threshold probability range of 0-60%. The study suggests this non-invasive strategy could help clinicians identify patients more accurately while reducing unnecessary prostate biopsies.