Probability propagation approach to left ventricular volume estimation from three-dimensional echocardiography

This paper presents an algorithm, based on a framework of probability propagation in factor graphs, that estimates left ventricular (LV) volume by tracking boundary points through a three dimensional echocardiographic data set. The tracking is made robust to frame-to-frame variations through the use of soft "edge likelihood" information. The only operator input to the algorithm is marking some highly visible landmark points along the LV boundary in a few frames. The points establish search regions for spatially intermediate frames, within which boundary points are found by maximizing edge likelihood. A similar procedure, employing forward and backward tracking, is used for temporal boundary tracking at each spatial location. LV volume is calculated from the set of estimated boundaries using planimetry. The results are validated by comparing ejection fractions to those computed from nuclear volumes and by cardiologists using commercial software.