Anisotropic Kernel Smoothing of DTI Data

r ( Kt ⊗ and the convolution in practical computation requires a limited t for a prescribed discrete window size, if a larger smoothing kernel is desired, the kernel convolution may be iterated to increase the effective t . Their mathematical relationship is: t t t t t n K ... K K K K ⊗ ⊗ ⊗ = (n times) Methods Data and gold standard images: A single-shot spin echo EPI sequence with diffusion-tensor encoding (12 directions, b=1000s/mm 2 ) , was used to get 9 sets (identical slice locations, voxels = 0.93x0.93x3.0mm interpolated to 0.9375~1mm quasi isotropic, 34 slices, 24 cm FOV) of DTI data from a single subject. Image misregistraion between DW images and data sets was corrected using affine image registration software(AIR; bishopw.loni.ucla.edu/AIR5/). "Gold standard" FA and principal eigen vector images were obtained by averaging all 9 sets of diffusion weighted images. Isotropic versus Anisotropic Kernel Smoothing: Both anisotropic kernel smoothing and isotropic Gaussian kernel, which was provided