Klaus H. Fritzsche1, 2, Bram Stieltjes2, Thomas van Bruggen2, Hans-Peter Meinzer2, Carl-Fredrik Westin1, Ofer Pasternak1
1Laboratory of Mathematics in Imaging, Harvard Medical School, Boston, MA, United States; 2German Cancer Research Center, Heidelberg, Germany
Partial volume is a major confounding factor in the analysis of diffusion tensor imaging datasets. The mixing effects of different compartments within each voxel are non-linear, acquisition dependent, and likely to exceed microstructural effects of interest. In this work, we combine two approaches to ameliorate these problems: Free-water elimination that eliminates intra-voxel CSF contamination and partial volume clustering that classifies and probabilistically selects all non-contaminated voxels. We demonstrate the increased sensitivity of this method in a tract specific analysis of the corpus callosum, recognizing abnormalities on a clinical dataset of Alzheimers disease, compared with matched controls.