Local Region with Optimized Boundary Driven Level Set Based Segmentation of Myocardial Ischemic Cardiac MR Images

摘要

In this work, an attempt is made to segment endocardium and epicardium of left ventricle in normal and myocardial ischemic cardiac magnetic resonance (CMR) images using local region with optimized boundary driven level set. Myocardial ischemia (MI) is a cardiac disorder that results in deprivation of oxygen supply to myocardium and can be analyzed by study of abnormal anatomical changes in CMR. This study is carried out on short-axis view CMR images from Medical Image Computing and Computer-Assisted Intervention (MICCAI) database. The edges are computed by simple Laplacian and Laplacian of Gaussian (LOG) operator. LOG is optimized to obtain enhanced edges of endocardium and epicardium. The quality of edge is validated with edge preservation index (EPI) and gradient magnitude similarity deviation (GMSD) measure. Local region with optimized boundary (LROB) driven level set is utilized for simultaneous segmentation of endocardium and epicardium of left ventricle in CMR images. The results are compared with local region (LR) driven and LR with LOG-driven level set. Further, the efficacy of the segmentation is validated with different similarity measures. The optimized LOG image visually shows better endocardium and epicardium contours. Optimized LOG with a higher EPI and lower GMSD provides better enhanced edges compared to Laplacian and LOG functions. The computed similarity measures for LR with LOG-driven level set are significantly higher compared to LR-based level set for segmentation of endocardium and epicardium. Further, LROB-driven level set shows higher similarity measures than LR with LOG-driven level set. Thus, LROB-driven level set provides better segmentation accuracy for epicardium and endocardium of left ventricle than LR-based level set and LR with LOG-driven level set. The efficiently segmented endocardium and epicardium could aid the diagnosis of myocardial ischemia with their ability to quantify anatomical changes in LV.