Medical Image Segmentation Using Deep Learning

Abstract

This chapter aims at providing an introduction to deep learning-based medical image segmentation. First, the reader is guided through the inherent challenges of medical image segmentation, for which actual approaches to overcome those limitations are discussed. Secondly, supervised and semi-supervised architectures are described, where encoder-decoder type networks are the most widely employed ones. Nonetheless, generative adversarial network-based semi-supervised approaches have recently gained the attention of the scientific community. The shift from traditional 2D to 3D architectures is also discussed, as well as the most common loss functions to improve the performance of medical image segmentation approaches. Finally, some future trends and conclusion are described.