Semantic segmentation of medical images using deep learning

Abstract

Semantic segmentation is a computer vision task that consider the objects of the same class as one entity. it has a large domain of applications such as autonomous driving, automatic identification of pathological tissues and more. . . . We will focus on brain tumor segmentation (BTS) problem which is considered as one of the most difficult segmentation problems and time consuming tasks in the medical domain, where we see that an automatic brain tumor segmentation system might be able to deal with some of these difficulties. In this study, we strive to propose an automatic BTS system based on Magnetic resonance imaging (MRI). where we rely on convolutional neural networks (CNN) in building our systems. We proposed three different approaches for BraTS20 and two approaches for BraTS17. These models were evaluated using dice score and yielded encouraging results.