Segmentierung der weißen Hirnsubstanz auf der Grundlage von MR-DTI-Vorzugsrichtungen

MR diffusion tensor imaging (DTI) allows the visualization of location and course of brain fiber bundles. To obtain these results, however, special evaluation techniques are necessary in addition to image acquisition and reconstruction. These include first the calculation of a preferential diffusional direction of water molecules in each voxel, and then the tracking of brain fibers or segmentation of regions with similar fiber directions. In both cases, the procedures available thus far require the interactive definition of seed points. In this paper, we propose a method to segment voxel groups of connected data points without the need of setting seed points. This method is based first on the identification of all voxels of a brain volume with a sufficiently unique preferential diffusional direction and with interconnection. For each selected voxel, neighboring voxels are then identified that have a small deviation from the chosen preferential direction and can therefore be grouped with this point. Finally, the largest partial volumes determined in this way are marked and color-coded to present them as three-dimensional structures. The present method was applied to a DTI data set of a healthy female volunteer, resulting in a largely automatic subdivision of the white matter in the brain in a number of bilateral partial volumes.