Spatiotemporal segmentation of MR image sequence based on hierarchical analysis

In this paper, a technique is described to segment blood volume of the left ventricle on MR image sequences based on graph representation. First, watershed transform is used to build an indexed hierarchy represented by a set of nested partitions of the image. The segmentation is based on two operations, the addition and suppression of classes or regions of the indexed hierarchy. These operations delete or include classes selected automatically and hierarchically according to centroid distance and grayscale criteria. Finally, we show experimental results and a preliminary comparison with the manual segmentation for one sequence of magnetic resonance images. The images are obtained from the informatics division of Institute do Coragao de Sao Paulo (In-Cor).