Research of the Feature Informativeness in the Classification of Raman Spectra for the Identification of Malignant Melanoma of the Skin

The aim of the research is to analyze the informativeness of spectral features of in vivo Raman spectra of skin neoplasms for the identification of malignant melanoma. Raman spectra of skin neoplasms were studied in the range of 300-1800 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup>. Spectral samples and complex features obtained by decomposing the Raman spectra using the multivariate curve resolution method were selected as features. Classification models of malignant melanoma and benign pigmented neoplasms, malignant melanoma and seborrheic keratosis were built. Logistic regression, discriminant analysis with projection onto latent structures and gradient boosting were selected as classification methods. The area under the receiver operating characteristic curve of the models takes values from 0.63 to 0.87. It is shown that complex features contain sufficient information about the spectral features of a skin area, and adding Raman spectra samples to the feature space does not affect the efficiency of malignant melanoma identification.