GCEENet: A Global Context Enhancement and Exploitation for Medical Image Segmentation

Abstract

Despite advancements in deep learning and computer vision, medical image segmentation is still a challenging problem. A major challenge for many segmentation models is the inherent complexity and inter-connectivity of pixels in medical images. These characteristics require modeling not only local features but also a global understanding of image semantics. In this paper, we propose a deep convolutional neural network called GCEENet to effectively address the above challenges. GCEENet features a combination of global context encoders and local distribution modules, working in conjunction to preserve the global image context. Our experiments on several medical image segmentation datasets show that GCEENet outperforms current state-of-the-art models in all measured metrics.