Order statistic based cardiac boundary detection in 3D+t echocardiograms

Abstract

We propose a boundary detector for echocardiographic images to be used in conjunction with deformable models. It is well suited to detect endocardial and epicardial boundaries in both 2D and 3D images. We demonstrate its capabilities on an example of Active Shape Models, where it is used as a force driving the mesh towards the cardiac walls. Although the proposed approach is mostly specific to echocardiography, it does not require any training to learn the image appearance (since construction of a training set of echocardiograms is very difficult and error prone). The detector is based on computing the medians of a series of neighborhoods and analyzing the change in their values to look for the evidence of an edge. The proposed algorithm was tested on thirty 3D echocardiographic sequences (corresponding to 10 healthy and 10 dyssinchronous hearts, the latter imaged at two stages of cardiac resynchronization therapy: before and at twelve month followup). © 2011 Springer-Verlag Berlin Heidelberg.