Optimal Design of DT-MRI Experiments Using a New Tensor-variate Gaussian Distribution

P. J. Basser, S. Pajevic NICHD / NIH, Bethesda, MD, United States, CIT / NIH, Bethesda, MD, United States SYNOPSIS: We propose that a necessary condition for an optimally designed DT-MRI experiment is that a new normal distribution for the estimated diffusion tensor, p(D) ∝ exp(1/2 D:A:D), be characterized by an isotropic 4-order precision tensor, A. This requirement ensures that statistical properties of D (and quantities derived from it) are all rotationally invariant. We propose a measure of anisotropy of A, which can be used to assess the degree of rotational invariance of any DT-MRI experimental design. We apply this measure to evaluate various DT-MRI experimental designs.