Paraffin-Embedded Prostate Cancer Tissue Grading Using Terahertz Spectroscopy and Machine Learning

Abstract

The automatic classification of paraffin-embedded prostate cancer tissue biopsy samples in terms of the Gleason scale is proposed using terahertz (THz) spectroscopy and machine learning. The samples with normal tissues (N=80) and prostate cancer tissues corresponded to the Gleason 4 (N=10), and 8 (N=13) scores, were analyzed. Absorption spectra of paraffin-embedded prostate cancer and healthy tissues were measured in the 0.2–1.5 THz range. The principal component analysis, support vector machine (SVM), and “majority vote” classification were applied to analyze experimental data. The original algorithm of spatial regions of interest selecting was developed to reduce the influence of the plastic base of a paraffin block on the results of a sample classification. The predictive model of the experimental spectral data of the paraffin blocks in the THz range was created using a set of the “One-Vs-One” binary SVM classifiers. We used multiple random splitting of the spectral data on the training and testing sets in 60%:40% proportion to teach the SVM classifiers. Validation of the predictive model showed 100% accuracy of the classification of the samples from the testing set.