Investigação de normalização de corantes para classificação de imagens da cavidade oral

Abstract

Cancer is one of the diseases with the highest mortality rate in the world. Dysplasia is a pre-cancerous lesion that is difficult to diagnose, performed from histological samples, which often do not have a good relationship of Hematoxylin and Eosin (H&E) dyes, which makes the diagnosis made by specialists in histology even more difficult. In this work, a method of normalization of dyes in histological images was investigated. This method employs a generative neural network based on a U-net for image generation and a PatchGAN architecture for information discrimination. The normalized histological images were evaluated in relation to the classification of the lesions. In these experiments, CNN models based on ResNet and AlexNet architectures were used, as well as hybrid models that use a junction of CNNs with machine learning algorithms. The use of ResNet architecture and the Random Forest algorithm provided results with an accuracy rate around 97% for the images normalized with the investigated method.