Comparison of a physical model and principal component analysis for the diagnosis of epithelial neoplasias in vivo using diffuse reflectance spectroscopy

We explored the use of diffuse reflectance spectroscopy in the ultraviolet-visible (UV-VIS) spectrum for the diagnosis of epithelial precancers and cancers in vivo. A physical model (Monte Carlo inverse model) and an empirical model (principal component analysis, (PCA)) based approach were compared for extracting diagnostic features from diffuse reflectance spectra measured in vivo from the dimethylbenz[alpha]anthracene-treated hamster cheek pouch model of oral carcinogenesis. These diagnostic features were input into a support vector machine algorithm to classify each tissue sample as normal (n=10) or neoplastic (dysplasia to carcinoma, n=10) and cross-validated using a leave one out method. There was a statistically significant decrease in the absorption and reduced scattering coefficient at 460 nm in neoplastic compared to normal tissues, and these two features provided 90% classification accuracy. The first two principal components extracted from PCA provided a classification accuracy of 95%. The first principal component was highly correlated with the wavelength-averaged reduced scattering coefficient. Although both methods show similar classification accuracy, the physical model provides insight into the physiological and structural features that discriminate between normal and neoplastic tissues and does not require a priori, a representative set of spectral data from which to derive the principal components.