A Review of Automatic Cardiac Segmentation using Deep Learning and Deformable Models

In this chapter, semi- or fully-automatic approaches for the segmentation of cardiovascular images (especially approaches focused on the segmentation of the LV) that use deep learning, deformable models, or a combination of those methods are reviewed. This chapter is organized as follows: Section 2 discusses the medical background, Section 3 reviews the deep learning-based segmentation of the LV, Section 4 explores deformable model-based segmentation and hybrid methods combining deep learning and deformable models, Section 5 includes the conclusions. This chapter discusses the semi- or fully-automatic approaches for the segmentation of cardiovascular images that use deep learning, deformable models, or a combination of those methods. Deformable models will add some feature-engineering to the deep learning approaches, resulting in the functioning of segmentation in the presence of a low amount of labeled data and without any user interaction. Deformable models will then generate the final refined segmentation contours based on the initial contour and shape priors. Deformable model-based, deep learning-based, and hybrid methods combining deep learning along with deformable models for the automatic segmentation of the cardiac images are discussed in this review paper. Hybrid methods, including the combination of deep learning and deformable models are offered to overcome the shortcomings of separate approaches. The right ventricle has a thinner wall and more complex geometry compared to the circular shape of the left ventricle.