Abstract
Scalar diffusion tensor imaging (DTI) measures, such as fractional anisotropy (FA) and mean diffusivity (MD), are increasingly being used to evaluate longitudinal changes in brain tissue microstructure. In this study, we aimed at optimizing the normalization approach of longitudinal DTI data in humans to improve registration in gray matter and reduce artifacts associated with multisession registrations. For this purpose, we examined the impact of different normalization features on the across-session test-retest reproducibility error of FA and MD maps from multiple scanning sessions. Diffusion data were pre-processed, fit to a tensor model to obtain FA and MD scalar maps and registered to standard stereotaxic space using different approaches that only differed in the features used in the normalization process, namely: 1) registration algorithm (FSL vs ANTs), 2) target image template (FMRIB58 FA vs MNI152 T1), 3) moving image (FA, MD, b0), and 4) normalization strategy (direct vs using an intermediate template). We found that a normalization approach using ANTs as the registration algorithm, MNI152 T1 template as the target image, FA as the moving image, and an intermediate FA template yielded the highest test-retest reproducibility in registering longitudinal DTI maps for both gray matter and white matter. Our optimized normalization pipeline opens a window to quantify longitudinal changes in microstructure at the cortical level.
