Autocorrelation-Based Technique for Automated Segmentation of Endocardial Border in 3D Echocardiograms

Left-ventricular (LV) segmentation is essential in the early detection of heart disease, where left-ventricular wall motion is being tracked in order to detect ischemia. In this paper, a new method for automated segmentation of the left-ventricular chamber is described. An autocorrelation-based technique isolates the LV cavity from the myocardial wall on 2-D slices of 3D short-axis echocardiograms. A morphological closing function and median filtering are used to generate a uniform border. The proposed segmentation technique is designed to be used in identifying the endocardial border and estimating the motion of the endocardial wall over a cardiac cycle. To this purpose, the proposed technique is particularly successful in border delineation by tracing around structures like papillary muscles and the mitral valve, which constitute the typical obstacle in LV segmentation techniques. The results using this new technique are compared to the manual detection results in short-axis views obtained at the papillary muscle level from 3D datasets in human and canine experiments in vivo. Qualitatively, the automatically-detected borders are highly comparable to the manually-detected borders enclosing regions in the left-ventricular cavity with a relative error within the range of 4.2%-6%. Thus, the new technique constitutes a robust segmentation method for automated segmentation of endocardial borders and is suitable for wall motion tracking for automated detection of ischemia in vivo.