Comparative analysis of methods for ascertainment the similarity between reflected spectra obtained from skin lesions

摘要

The presented investigation demonstrates the results of tests of abilities for classification of reflected spectra from human cutaneous lesions. Typical peculiarity for these spectra is big deviation for collections obtained from healthy skin areas and from lesions with same diagnoses. Two different approaches are used to ascertain similarity: calculation of the distance to the point of full identity in area of chromaticity and neural network identification. The test is provided on data from patients with different histologically fixed diagnoses. Each diagnose is presented by etalon distribution. The test samples are preliminary processed transforming the distributions to normalized spectra, which are ratio between spectra measured from healthy and damaged skin. The etalon distributions are extracted by averaging the ratio-spectra. Methods are compared by criterion of number coincidence of histological and classified diagnoses.