Improved Spherical Deconvolution to Solve Fiber Crossing in Diffusion-Weighted MR Imaging

Abstract

An improved spherical deconvolution algorithm to solve fiber crossing in diffusion magnetic resonance imaging is here presented. The introduction of a regularization parameter on the reconstruction of the fibers directions allows to consider the deconvolution as a constrained least squares problem and enforces the normalization of the reconstructed directions. Moreover a new automatic stopping criterion is implemented which allows to push the algorithm to convergence. These modifications improve significantly the performance of the algorithm, decreasing the resolution limit and reconstructing better profiles of the fibers.