A Multivariate Group-wise Genetic Analysis of White Matter Integrity using Orientation Distribution Functions

Diffusion magnetic resonance imaging has become an important tool for comparing brain white matter fiber structure between groups of subjects. While voxel-wise statistical comparison are typically performed on scalar values derived from the diffusion tensors (DT ), several authors have advocated applying multivariate statistics to better exploit the information contained in the tensors, as they show significant improvements over their univariate counterparts. The DTs are good approximations to the fiber orientation in regions with no fiber crossings or partial volumed voxels; however, fiber crossings are ubiquitous in the brain, and the tensor approximation fails throughout a significant portion of the image. Consequently, here we treat this issue by analyzing the raw diffusion data directly and by building orientation distribution functions (ODF s), using the modified spherical harmonic decomposition from [7]. More precisely, we first perform linear and nonlinear registrations to transform these diffusion-weighted images to a common space, and a subsequent local rotation to reorient the diffusion signal and thus align it to the anatomy. We then build the ODF s. Our dataset consists of 25 identical and 25 fraternal twin pairs (100 healthy adult twins, age: 24.6± 1.8 SD years), for whom we aimed to study the genetic contribution to white matter integrity. A multivariate intraclass correlation value (ICC) is obtained from the coefficients of the spherical harmonics at each voxel, resulting in a map that represents similarity within twin pairs in each of the twin groups. From there, the heritability can be computed. These statistics obtained from the multivariate ICC are compared to those from a more commonly used scalar measure: the generalized fractional anisotropy (GFA). As shown here, accounting for directionality increases the signal detection.