The objective of this research is to develop automated ventricular border detection algorithms to be used with three-dimensional image data. This would be used to obtain cross-sectional areas, chamber volume, left ventricular chamber and wall shapes, and regional wall motion. The investigators have been working on algorithms for segmentation of three-dimensional images based on stacks of two-dimensional images. The objective is to extend these algorithms directly to three-dimensional data obtained from a volumetric ultrasound imaging system. The main algorithm developed by the investigators is based on fitting a smooth surface to a rough estimate of the surface. Smoothness is obtained by minimizing a cost function subject to several smoothness constraints which prohibit the fitted surface from changing more than a certain amount from one voxel to the next. The optimum fit is obtained using a graph theoretic approach. The goal is to provide accurate quantitative data that are not presently available in order to enable more accurate investigations of global and regional cardiac dysfunction.
