Abstract
This study investigates the potential of Raman spectroscopy for liquid biopsy in prostate cancer using serum samples. We evaluated four machine learning models and Principal Component Analysis (PCA) to classify prostate cancer based on Raman data. Support Vector Machine (SVM) demonstrated the best performance, achieving high accuracy, sensitivity, and F1 scores, with the highest overall. Random Forest (RF) also showed strong results, with accuracy of 0.87. The analysis identified two key spectral bands: 1306 cm(-1) and 2929 cm(-1), as potential biomarkers for prostate cancer. The PCA revealed that particular it is possible to differentiate serum collected from control and prostate cancer patients. Correlation analysis showed that the 2929 cm(-1) band was significantly associated with PSA levels, MRI PIRADS, and lymph node metastasis (pN+), while the 1306 cm(-1) band showed strong correlation with PSA and MRI PIRADS. These findings suggest that Raman spectroscopy, particularly the 2929 cm(-1) band, holds promise as a reliable method for prostate cancer detection in liquid biopsy applications.