Segmentation Of Medical Images Using Deep Learning

Abstract

It offers computerized delineation of particular anatomical structures of interest and also facilitates numerous subsequent tasks like shape analysis and volume measurement. Specifically, the fast innovation of deep learning methods in the past few years has significantly contributed to improving the performance of segmentation algorithms by making effective use of large amounts of labelled data in order to improve complex models. Nevertheless, obtaining manual labels to train can become a significant burden for the applicability of learning-based approaches in medical images. Scientists have investigated numerous unsupervised and semi-supervised learning techniques to lessen the labelling limitation in an effort to address this problem.  This chapter presents the fundamental ideas of deep learning-based segmentation as well as some of the most advanced techniques now in use, arranged according to the level of supervision. Our aim is to give the reader some potential solutions for model choice, training strategy, and data manipulation based on a certain segmentation task and dataset.