Analysis of Stokes shift spectra for distinguishing human prostate cancerous and normal tissues using machine learning methods

Early detection of prostate cancer is critical for the success of cancer therapy. It is believed that the biochemical changes that cause the optical spectra changes would appear earlier than the histological aberration. The aim of this ex vivo study was to evaluate the ability of Stokes Shift Spectra (S3) to identify human prostate cancerous tissues from the normal. Fifteen (15) pairs of with pathologically confirmed human prostate cancerous and normal tissues underwent Stokes Shift Spectra measurements with selective wavelength interval of 40 nm. The spectra were then analyzed using machine learning (ML) algorithms to classify the two types of tissues. The ML algorithms including principal component analysis (PCA) and nonnegative matrix factorization (NMF) were used for dimension reduction and feature detection. The characteristic component spectra were used to identify the key fluorophores related to carcinogenesis. The results show that these key fluorophores within tissue, e.g., tryptophan, collagen, and NADH, have different relative concentrations between cancerous and normal tissues. A multi-class classification was performed using support vector machines (SVMs). A leave-one- out cross validation was used to evaluate the performance of the classification with the gold standard histopathological results as the ground truth. The results with high sensitivity and specificity indicate that the S3 method is effective for detecting changes of fluorophore composition in human prostate tissues due to the development of cancer.