Urine-Based Machine Learning Assay Detects Prostate Cancer

Background/Objectives: Prostate cancer testing relies on prostate-specific antigen testing and digital rectal examination, which have limited specificity and face cultural or geographic barriers to access. We developed a non-invasive urine-based liquid biopsy assay using engineered hydrogel arrays and machine learning to detect disease-specific biochemical profiles. Methods: We collected voided urine samples from 283 participants at 26 U.S. urology practices prior to prostate biopsy. Random forest classifiers trained on 184 biopsy-confirmed cancer cases and 75 controls analyzed colorimetric signatures. Results: Across all Gleason grades (6-10), the assay achieved 97.8% sensitivity and 53.3% specificity. Performance varied by grade: high-grade cancers showed 97.3% specificity, while low-to-intermediate grades demonstrated 94.0% sensitivity. Conclusions: This accessible, culturally-appropriate platform could expand prostate cancer detection in diverse populations while reducing unnecessary invasive biopsies.