An Ensemble Algorithm for Analyzing Raman Spectra of Skin

Skin diseases are widespread all over the world. It is convenient to use optical methods, such as Raman spectroscopy, to diagnose such diseases. However, the Raman scattering spectra of the skin are very difficult to identify because they contain a large amount of information about the chemical composition of the skin. In this study, the multidimensional curve resolution method was used to reduce the amount of data. With his help, the Raman spectra of the skin were divided into thirty components, the relative contribution of which to the total spectrum was further used as classification criteria. An ensemble algorithm based on stacking was used as a method for constructing a classification model. Binary classification models were built for three cases: benign neoplasms versus malignant neoplasms; malignant melanoma versus pigmented nevus and malignant melanoma versus pigmented nevus and seborrheic keratosis. The ROC AUC of the obtained models ranges from 0.78 ± 0.10 to 0.80 ± 0.08. The results of the study confirm the effectiveness of the ensemble algorithm for the task of identifying Raman spectra.