Label-free surface-enhanced Raman spectroscopy detection of prostate cancer combined with multivariate statistical algorithm

The purpose of this study was to detect human plasma for screening prostate cancer (PCa) and benign prostatic hyperplasia (BPH) based on surface-enhanced Raman spectroscopy (SERS) and multivariate statistical algorithms. The test was to detect 106 plasma samples, which originated from 39 normal subjects, 26 patients with PCa and 41 patients with BPH. Significant differences in peak intensity at 495, 636, 1135, 1205, and 1675 cm(-1) can be observed from the difference spectrogram, which contributes to initially distinguish the cancer group from the normal group. Then, the multivariate statistical techniques, including principal component analysis (PCA) and linear discriminant analysis (LDA) diagnostic algorithms, as well as recursive weighted partial least square (PLS) method and support vector machine (SVM) algorithm, were used to analyze the spectral data. For PCa versus normal group and BPH versus normal group, the classification accuracy of PCA-LDA was 96.80% and 97.50%, respectively, and the classification accuracy of PLS-SVM was 100.00% and 100.00%, respectively. In the diagnosis of PCa and BPH, the sensitivity, specificity and accuracy of PCA-LDA were 65.40%, 75.60%, and 71.06% respectively, and the area under the curve (AUC) value of the receiver operating characteristic (ROC) curve was 0.788, while the sensitivity, specificity and accuracy of PLS-SVM were 88.46%, 87.80%, and 88.06%, respectively, and the AUC value was 0.881. The diagnostic results of PLS-SVM are better than PCA-LDA, which supported that PLS-SVM algorithm has greater potential than PCA-LDA algorithm in the pre-diagnosis and screening of PCa.