Prostate cancer is the most prevalent malignancy among men, ranking first in incidence in the US and UK and second in Korea. Given its high occurrence and the limited curative options at advanced stages, early and accurate diagnosis remains essential. However, current approaches largely rely on invasive biopsy with physical examinations. As a simple and rapid diagnostic tool, nuclear magnetic resonance (NMR) and near-infrared (NIR) spectroscopy have been explored in this study. Especially, a hyperspectral system enabling of high-throughput analysis was employed to acquire NIR spectra of multiple dried urines samples on a Teflon.
Absolute amounts (concentrations) of metabolites were different from patient to patient, so overall intensities of NMR peaks of the samples accordingly varied. To compensate for the variation, Probabilistic Quotient Normalization (PQN) was applied to each of the NMR spectra. PQN enabled the identification of correlations between the prostate cancer and altered levels of creatinine and phosphocholine. Then, ratios of these peaks between the prostate cancer and the non-cancer groups were examined and distribution of the samples after nonlinear dimensionality reduction was investigated to identify prostate cancer. Although, the NIR peaks were less molecular-specific compared to those of NMR, overall pattern of NIR spectra representing whole composition of urine metabolites would be alternatively valuable for the identification. Differences in the NIR peaks between the prostate and non-cancer groups were then searched, and a multivariate classifier was employed to identify the prostate cancer samples. Finally, the identification accuracies based on the NMR and NIR data were compared, and the pros and cons of both methods were discussed.
