Fast segmentation of brain magnetic resonance tomograms

We describe a combination of a region growing and a watershed algorithm optimized for the detection of homogeneous structures in magnetic resonance (MR) volume datasets. No prior knowledge is used except a segment model. The adaptation to different data sets is controlled by parameters which can be determined interactively due to the high speed of the algorithm. Results are shown for the segmentation of the basal ganglia and the white matter of the brain