Segmentation of 2D stress echocardiography sequences using rest-based patient-specific prior information

In stress echocardiography, the heart is imaged at rest and again when stressed to observe the change in function between these two states; the idea being that abnormalities will be exaggerated and therefore easier to identify in stress, but importantly this is referenced to the rest state. Despite the development of segmentation and tracking techniques for the heart at rest, there is little literature on the same for the stressed heart. First we propose a patient-specific segmentation technique that gives a prediction of stress dataset segmentation given rest dataset segmentation for a healthy heart through the use of a global motion model based on Canonical Correlation Analysis (CCA). Secondly, we refine this prior segmentation using texture measures from the rest dataset as reference parameters for maximum likelihood estimation of the boundary in the stress dataset. Results show that for 52 out of 78 datasets, our model gives better results than using the technique described in.