Nonlinear Registration of Diffusion Tensor Images Using Directional Information

G. K. Rohde1,2, S. Pajevic3, C. Pierpaoli1 STBB/LIMB/NICHD, National Institutes of Health, Bethesda, Maryland, United States, Dept. of Mathematics, University of Maryland, College Park, Maryland, United States, MSCL/CIT, National Institutes of Health, Bethesda, Maryland, United States Introduction In Diffusion Tensor (DT) MRI [1], local diffusion properties are described via a 3x3 symmetric diffusion tensor. From the DT and the T2-weighted a mplitude parameters several diffusion related quantities can be derived. E.g.: eigenvectors and eigenvalues of the DT, trace, diffusion anisotropy, etc. Thus, when registering (spatially aligning) DT images one has several potential choices of parameters to use to measure image similarity during registration. Previous work [2] has shown the advantages of using multiple DT parameters simultaneously during image registration, though the parameters used in that work were rotationally invariant. In this work we compare the use of channel configurations that include rotationally invariant scalar quantities derived from the DT model against channel configurations that include directional information, such as the DT elements. Methods The multi-channel image registration problem is defined as an optimization problem where the goal is to find spatial transfor mation f : x → R3 that maximizes some multi-variate image similarity measure I (⋅,⋅) between a source multi-channel image S(x) and a target image T(x) . Image similarity is measured using the multivariate measure defined in equation (1), where