Unsupervised classification of skeletal fibers using diffusion maps

In this paper, we propose an application of diffusion maps to fiber tract clustering in the human skeletal muscle. To this end, we define a metric between fiber tracts that encompasses both diffusion and localization information. This metric is incorporated in the diffusion maps framework and clustering is done in the embedding space using k-means. Experimental validation of the method is performed over a dataset of diffusion tensor images of the calf muscle of thirty subjects and comparison is done with respect to ground-truth segmentation provided by an expert.