Distance regularized level sets for segmentation of the left and right ventricles

In this chapter, we present a novel active contour method, called the distance regularized two-layer level set (DR2LS) algorithm to automatically segment both the left and right ventricles. The DR2LS approach fuses the strengths from two of our previous works, including the region-scalable fitting (RSF) model in Li et al. [1] and the distance regularized level set evolution (DRLSE) method in Li et al. [2], both proposed by Li et al. We first review the frameworks of the RSF and DRLSE models, and then we describe how to combine them efficiently into DR2LS model and the application of this combined model toward the left and right ventricle segmentation. In the combined level set algorithm, named DR2LS, we use the 0-level and k-level contours of a level set function to mathematically represent the endocardial and epicardial surfaces of both ventricles. In order to combine the strengths from our previous works, we update the formulate of level set accordingly, with the data term from the RSF model adopted for the detection of endocardium and epicardium, and the distance regularization term in the DRLSE framework to regularize the distance between the 0-level and k-level contours, and a commonly used arc-length term to regularize the 0-level and klevel contours. Lastly, we provide some preliminary segmentation results of the left and right ventricles by initializing the boundaries with the standard edge-based DRLSE model and then implementing the DR2LS model to extract both endocardium and epicardium. Experimental results have demonstrated the effectiveness of the proposed two-step level set approach to segment cardiac left and right ventricles simultaneously.