Machine learning-assisted serum Raman spectroscopy for risk stratification in men with suspected prostate cancer: a retrospective proof-of-concept study

Background Serum prostate-specific antigen (PSA) remains the principal laboratory investigation for men with suspected prostate cancer but has limited specificity, particularly in patients with PSA concentrations between 4 and 10 ng/mL. Additional blood-based tests capable of improving risk stratification before imaging or biopsy could reduce unnecessary investigations while maintaining diagnostic sensitivity. Raman spectroscopy provides a biochemical fingerprint of biological samples and, when combined with machine-learning methods, may offer a novel approach to serum-based cancer detection. Methods Archived serum samples obtained from men with histologically confirmed prostate cancer and symptomatic control subjects were analysed using dried-serum Raman spectroscopy. Raman spectra were incorporated into supervised machine-learning classification models to generate predicted probabilities of prostate cancer. Diagnostic performance was evaluated in the complete cohort and separately in patients with PSA concentrations between 4 and 10 ng/mL. Results The study included 119 men with prostate cancer and 60 symptomatic controls. Predicted probabilities of prostate cancer were significantly higher in cancer cases than controls. The classifier demonstrated moderate discrimination in the complete cohort (AUC approximately 0.77) and maintained similar performance in patients with PSA concentrations between 4 and 10 ng/mL (54 prostate cancer and 41 symptomatic controls, AUC approximately 0.78). Conclusions Machine learning-assisted analysis of dried-serum Raman spectra shows promise as an adjunct to PSA testing. The maintenance of diagnostic performance within the PSA grey zone suggests a potential role as a second-line risk-stratification test before MRI or biopsy. Prospective validation within contemporary diagnostic pathways is now required.