Improved skin xerosis detection by combining extracted features from Raman spectra

In this paper, we studied skin roughness, so called skin Xerosis, with Raman Spectroscopy technique. The main goal of this study was to determine the features from the Raman spectral data associated with skin xerosis and finally detecting Xerosis from normal skin by using an optimal classification method. The Raman spectral dataset was constructed from two classes of spectral data, 70 spectra of normal intact skin and 70 spectra of irritated rough skin. Roughness irritation was done by sodium dodecyl sulfate(SDS) non-ionic surfactant applied once a day on rat skin within a week. The spectra were obtained from two rats from legs and dorsal anatomical regions. Some features related to specific bond vibrations of water, lipid and protein structures were extracted from the spectra. T-test statistical analysis was then utilized to determine if the specified feature could discriminate the two classes of spectral data. The reported efficient features from t-test stage were then applied to well-known classification methods such as LDA and KNN for classification. Classification performance was calculated using k-fold cross validation method for selecting the optimum classifier and features. The statistical analysis of water content and lipid structures between two classes showed a significant difference by p-value<<0.1, whereas alterations in features related to proteins were not remarkable between two classes of data. These results suggest that water content and lipid structures are the proper features for skin roughness detection. Classifiers performance results propose that water content feature is the most altered feature amongst the two classes.