Reconhecimento de tumores cerebrais utilizando redes neurais convolucionais

Abstract

Recently, with the advancement of technology, new forms of detection and, hence, disease prevention have been employed. Thus, computer-aided diagnosis systems have been increasingly used and therefore new opportunities have been opened for medical computing professionals as well as a variety of research options in the academic field. In the present work, a Convolutional Neural Network (CNN) is implemented and tested with promising results for the early detection of brain tumors using 512 ◊ 512 size images in gray scales. The algorithm has the objective of, through successive training of the network, to pre-classify the test images as having a brain tumor (when applicable). For this, when presenting a certain test image to the network, it should classify it as having a brain tumor or not. This is very significant, since the early detection of brain tumors is not a simple task to perform. The algorithm obtained an index of 87% coverage for disease-related images and 66% for healthy images, thus presenting satisfactory results for pre-detection of the disease. The network in question still obtained about 94% accuracy to classify healthy images and 62% for images.