Segmentation of 3D MR image sequences

Presents an algorithm for the segmentation of three dimensional MR image sequences. The algorithm integrates edge-based and region-based techniques through the morphological algorithm of watersheds. Segmentation starts with an edge-preserving noise reduction process followed by differentiation with the Gaussian filter. At the next step, an initial segmentation of the image is obtained by detecting the watersheds of the image gradient magnitude. This initial segmentation is improved by a hierarchical region merging process at each step of which the most similar pair of neighboring regions is merged. The segmented image is represented by a Region Adjacency Graph (RAG) and merging is implemented using a priority queue (heap) that stores the edges of the RAG. Merging is remarkably accelerated by the use of an additional graph, the Most Similar Neighbor Graph. Results obtained with three dimensional cardiac Magnetic Resonance image sequences are presented.