Evaluation of White Matter Fiber Clustering Methods for Diffusion Tensor Imaging
Moberts B, Wijk J, Buijs J, Pul C, Roos F, Vilanova A
Eindhoven University of Technology, Eindhoven University of Technology
Fiber tracking generates individual fibers from Diffusion TensorImaging data. In literature, we can find several methods forclustering of white matter fibers. However, it is not clear whichmethod produces the best results. Many combinations exist between:fibers similarity measures, clustering techniques and parameterdefinitions. We present a comparison of some clustering methodsusing a quality measure that was calibrated and validated for thistask. From the tested methods, the results show that the mean of the closest pointsdistance gives the best results as similarity measure, and that thehierarchical single link gives best results concerning clusteringand robustness.