Multi-Tensor Field spectral segmentation for white matter fiber bundle classification

We present an algorithm for segmenting the white-matter axon fiber bundles from HARDI images. We formulate the segmentation problem as a Multi-Tensor Field segmentation problem in which the compartments of each Multi-Tensor may belong to different classes, allowing the algorithm to handle crossing fiber tracts. Experimental results on two publicly available synthetic datasets and the fiber-cup phantom show that fiber crossings can be effectively separated by using this approach, and preliminary results on real data show that the segmentation obtained is consistent with known anatomical structures.