P2A-2 Three-Dimensional Cardiac Image Segmentation Using Adaptive Filtering and 3D Deformable Simplex Meshes

Abstract

Semi-automatic segmentation of the myocardium in three-dimensional (3D) echographic images may substantially support clinical diagnosis of (congenital) heart disease. It can facilitate visualization of abnormal cardiac anatomy and may serve as an important preprocessing step for automated cardiac strain imaging. Echocardiographic image sequences of the left ventricle of two healthy subjects and one piglet were obtained in radiofrequency (RF) format, directly after beamforming, in 3D live and in Full Volume mode. To optimize the distinction between blood and myocardium, 3D Adaptive Mean Squares (AMS) filtering was performed on the demodulated rf-data. Earlier work on 2D data revealed that this filter reduces speckle noise, while preserving the sharpness of edges between various structures. In this study a 3D deformable model based on a simplex mesh was then used to segment the endocardial surface. The model deforms under influence of internal (regularization) and external (data) forces and is initialized by placing a spherical surface model in the left ventricle. A gradient and a speed force were included in the external force of the model. Weighting factors of internal, gradient and speed forces were interactively set to balance data fitting and mesh regularity. Initial results show that segmentation of the endocardial surface using 3D deformable simplex meshes in combination with adaptive filtering is feasible. The speed force led to improved segmentation in all datasets as the deformable model was less dependent on initialization. The method is promising for application to nonstandard heart geometries without having to impose strong shape constraints. To prevent the model from leaking into the left atrium or crossing areas with weak boundary information, the use of attractor forces and weak shape constraints could be helpful.