Classificação Automática para Auxílio no Diagnóstico de Lesões de Pele Usando Deep Convolucional Neural Network com Modelos de Transfer Learning

Abstract

In Brazil, skin cancer has become the most frequent neoplasmamong patients. This type of cancer, if detected early, increases the chances of cure. However, with the lack of health professionals qual-ified to perform this procedure, for example, in distant regions large urban centers, difficulties in early diagnosis process are recurring. Thus, a possible solution to this problem is the development of mod-els that allow classification for the diagnosis skin based on Deep Learning. Therefore, this work presents a model that can contributeto the diagnosis of types of skin devices in the HAM10000 dataset.The proposed model achieved a balanced accuracy of 74.50% witha set.