Urine-based Raman markers for prostate cancer diagnosis: A machine learning approach using fingerprint and lipid spectral region

Abstract

This study investigates the potential of Raman spectroscopy in distinguishing between healthy individuals and prostate cancer patients using urine samples. The Boruta algorithm was applied to Raman spectral data in two distinct wavenumber regions: 800–1800 cm⁻¹ (fingerprint region) and 2800–3000 cm⁻¹ (lipid region). The algorithm identified important spectral features from both regions that were used to construct decision trees for classification. Key wavenumbers in the fingerprint region (1009 cm⁻¹) and high-wavenumber region (2937 cm⁻¹) were found to be significant markers for prostate cancer detection. Principal Component Analysis (PCA) revealed that the intensity of these markers effectively separated healthy and cancerous samples, with the 1009 cm⁻¹ marker showing higher discriminative power. Furthermore, four classification models: Decision Tree (DT), k-Nearest Neighbors (kNN), Random Forest (RF), and Support Vector Machine (SVM) were evaluated for their performance in classifying urine samples based on Raman spectral features. The RF and kNN models exhibited the best overall performance, with high accuracy and sensitivity, particularly in the 800–1800 cm⁻¹ region. The study also explored the correlation between Raman markers and clinical parameters, finding that the 2937 cm⁻¹ marker had strong correlations with critical clinical variables like Gleason scores and MRI PIRADS scores, suggesting its utility for prostate cancer diagnosis and staging. These findings highlight the potential of Raman spectroscopy as a non-invasive tool for prostate cancer detection and monitoring.