Does one size fit all? Reconstructing crossing fibers in diffusion MRI using spherical deconvolution with local response functions

We present Local Spherical Deconvolution (LSD), a diffusion MRI deconvolution method to reconstruct crossing fiber directions in the brain. In contrast to previous approaches which assumed a single deconvolution kernel for the entire brain, LSD utilizes information theory to identify an optimal kernel in each image voxel. Using a high-resolution post-mortem chimpanzee brain, we show that fibers are reconstructed with LSD with increased precision and a reduced number of false peaks compared to conventional methods. A test-retest analysis supports the stability and accuracy of LSD