Graph-cut based edge detection for kalman filter based left ventricle tracking in 3D+T echocardiography

摘要

Consistent endocardial border segmentation in 3D echocardiography is a challenging task. One of the major difficulties rises due to the fact that the trabeculated structure of the endocardium causes the endocardial intensity profile characteristics to change over a cardiac cycle. In this paper, we present a hybrid edge detection approach using both max flow/min cut (MFMC) and step criterion (STEP) edge detectors, and its integration into a Kalman filter based left ventricle (LV) tracking framework. We treat the endocardial edge detection problem as a graph partitioning problem where the graph is defined by using the intensity profiles, and propose a max flow/min cut based solution. For the end-systole, the step criterion edge detector is a more suitable option. Accordingly, we introduce the weighted combination of these techniques called the hybrid edge detector (Hybrid) where the weight factor is determined by the size of the tracked endocardial mesh. Surface and volumetric measurement comparisons between the STEP, MFMC and Hybrid shows that the Hybrid handles the specific problem of time-dependent intensity profiles better than the other approaches.