Redes neurais convolucionais para segmentação de tumores cerebrais em imagens de ressonância magnética.

Abstract

Cerebral cancer is one of the cancer types with bigger mortality rates in the present days. The most common form of brain cancer is the glioma, which is easily visualized in magnetic resonance images. However, the measurement and evaluation of gliomas is complex and time-consuming due to the great variety and number of medical image data the medical professional has to deal with day-to-day. This increases intra- and inter-operator error variability, making it necessary to develop computer assisted diagnosis (CAD) tools to overcome such problems. In this work, these concepts are approached thoroughly in order to propose and present a method for segmentation of brain tumours on MRI images. The proposed method achieved a Dice score of 88% for anomaly segmentation, evaluated on the BRATS 2016 database.