Automated segmentation of cardiac MR images

Two different algorithms were tested for the segmentation of cardiac MR (magnetic resonance) images. The first is the pyramid segmentation scheme, which was finally rejected owing to severe inherent shortcomings. The second is called the cavity detector and is based on classical image processing techniques. It is especially sensitive to compact objects in the image, even when these are not completely surrounded by walls. The second scheme is robust with respect to changes in the (automatically or a priori defined) segmentation parameters (such as threshold value or mask for distance transform) and has been validated by comparison with manual segmentation by a medical expert.<<ETX>>