Automated heart localization for the segmentation of the ventricular cavities on cine magnetic resonance images

In this paper, a previously developed segmentation algorithm was improved by adding a fully automated definition of a region of interest around the left ventricle. The method, dedicated to cine images, proceeds in two major steps. First, heart localization was achieved by exploiting the heart's beating motion. Then, a region of interest was created around the left ventricle by exploiting its circularity. Segmentation of the left ventricle was also performed with the previously proposed approach. The algorithm was tested on the cine MR images of the 2009 MICCAI Challenge database, including 45 subjects. The regions of interest and resulting segmentation were visually graded slice by slice. Heart localization was successful in all cases, and the restricted region of interest was acceptable in 2/3 of all cases. Finally the segmentation using the fully automated approach was satisfying for about 50% of the slices.