Mapping the influence of genes on brain structure using tensor based morphometry (TBM) and diffusion tensor imaging (DTI)

Genetic studies have been growing in popularity as a subfield of computational neuroimaging. Improvements in computational techniques for genetic studies may greatly further our understanding of the contribution of genes to brain structure and function. Here we use Tensor-based morphometry (TBM) to analyze both structural MRI (sMRI) and diffusion tensor images (DTI) to understand group differences in anatomy in populations with congenital disorders (Fragile X, Velo-cardio facial syndrome (VCFS), congenital blindness) compared to healthy control populations. We also look at the influence of genes on normal brain anatomy using sMRI and DTI images from healthy young adult twins.;TBM consists of registering all subjects from two groups to be compared to a common space, and analyzing the Jacobian matrices induced by the deformation to obtain group differences in local brain size and/or shape. Here we first show results of a standard TBM analysis on Fragile X and VCFS datasets. We extend TBM to DTI by performing group statistics on the diffusion tensors instead of the Jacobian matrices. We also improve on the more common univariate analyses by looking at several multivariate measures. Both uni- and multi-variate DT-derived measures are analyzed, including: the 6-dimensional full DT and the 3 eigenvalues, standard univariate measures such as the fractional anisotropy (FA) and the mean diffusivity (MD), and the more geometrically correct geodesic anisotropy. The method is applied to obtain anatomical differences from innately blind subjects, as well as to extract genetic information from the twins. Effect sizes are stronger with the multivariate statistics, compared to the more standard scalars ones.;Standard scalar twin statistics include the intraclass correlation and Falconer's heritability. More recently, structural equation models were used in twin statistics (ACE model) to better distinguish between the various genetic and environmental factors that may contribute to phenotypes. We extend and apply all of these statistical methods from univariate to DT-derived multivariate measures. Genetic analysis of the full DT leads to better-fitting statistical models with higher power to detect effects of gene and shared-environments. The multivariate intraclass correlation can be improved by using the restricted maximum-likelihood method.