Generalized Diffusion Tractography Based on Directional Data Clustering

Abstract

A new methodology to reduce uncertainty in estimating the orientation of neuronal pathways in diffusion magnetic resonance imaging is proposed. The methodology relies on three main features. First, an optimized high angular resolution diffusion imaging reconstruction technique is adopted. For each voxel, the orientation distribution function (ODF) on the unit sphere is reconstructed to extract the principal diffusion directions. Second, directional statistics are used to estimate the principal ODF profile directions from data distributed on the unit sphere. For this purpose, a mixture-model approach to clustering directional data based on von Mises-Fisher distributions is adopted. Third, a modified streamline algorithm able to accommodate multiple fiber tracts and multiple orientations per voxel is used, to exploit the directional information gathered from estimated ODF profiles. The methodology has been tested on synthetic data simulations of crossing fibers and on a real data set.