Estimating the number of fiber orientations in diffusion MRI voxels: a constrained spherical deconvolution study

Introduction: It has long been recognized that the Gaussian diffusion tensor model is inappropriate for voxels with complex fiber architecture [1, 2]. Many groups have tried to classify voxels in terms of their diffusion complexity. The earliest studies distinguished between voxels with isotropic, single-fiber anisotropic, and multifiber anisotropic complexity and have reported clustered and symmetric regions of increased complexity, supporting genuine effects consistent with anatomical knowledge [3, 4]. However, they were not able to report the number of fiber orientations found in each voxel. Recent advances of high angular resolution diffusion imaging allow the extraction of multiple fiber orientations [5] and have spawned an interest for classification of voxels by the number of fiber orientations. Recently, a Bayesian Automatic Relevance Determination method was proposed to infer the number of fiber orientations in a multi-compartment model [6]. The study was able to find voxels with up to two fiber orientations and estimated the proportion of white matter fiber voxels that contain complex fiber architecture at approximately one third. In this work, we estimated the number of fiber orientations within each voxel using the constrained spherical deconvolution (CSD) method [7] with the residual bootstrap approach [8]. We showed that multiple-fiber profiles arise consistently in various regions of the human brain where complex tissue structure is known to exist. Moreover, we detect voxels with more than two fiber orientations and detect a much higher proportion of multi-fiber voxels than previously reported.