Chantal M.W. Tax1, Ben Jeurissen2, Max A. Viergever1, Alexander Leemans1
1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands; 2Department of Physics, University of Antwerp, iMinds-Vision Lab, Antwerp, Belgium
Spherical deconvolution techniques characterize complex fiber configurations within a voxel. Currently, the response function (RF) needed for deconvolution is derived from voxels with the highest fractional anisotropy (FA). Poor accuracy of FA in the high b-value and low SNR regime and the ad-hoc nature of selecting these voxels complicate RF estimation, which may lead to the detection of false positive peaks and altered peak magnitudes. In this work, the computation of the RF is optimized by excluding "crossing fibers" voxels in a recursive framework, which does not rely on FA. Feasibility is demonstrated on simulated and real diffusion MRI data.