Classificação de núcleos com diferentes níveis de displasia em imagens histológicas

Abstract

This work proposes a system capable of evaluating images of different degrees of oral epithelial dysplasia and classifying them into healthy, mild, moderate and severe groups, according to the degree of dysplasia presented in the image. Using the Random Forest classifier and morphological descriptors, Haralick descriptors, structural co-occurrence matrix, Moran's I and entropies, it was possible to obtain a multi-class lesion classifier with an accuracy of 82.5%.