The influence of the registration on voxel-based statistics of fractional anisotropy images: using detected white matter degenerations associated with juvenile myoclonic epilepsy as a gold standard

Introduction: Recent developments have enabled automated voxel-based statistical (VBS) analyses of DTI data, so that fractional anisotropy (FA) images, for example, may be quantitatively compared between groups of subjects without manual investigator dependency. In contrast to conventional VBM of T1w images, spatial normalization for VBS of FA images (FA-VBS) may be based on different image contrasts such as structural T1w images, T2w images measured without diffusion gradient (i.e. with b=0 s/m2, here denoted as the b0 image) or FA images. In another study, we developed a SPM normalization toolbox for FA-VBS that enables the construction of singleand multicontrast registrations based on FA and b0 images [1]. In particular, each registration can be employed iteratively to improve the registration accuracy. Yet, the questions, which registration contrast is best for FA-VBS and if iterative registration is necessary, could not be sufficiently answered, due to the lack of a gold standard. In this study, we apply our toolbox to investigate the influence of the spatial registration on FA differences detected by voxel-based statistics. To assess the influence of the registration on the FA-VBS results, we analyze the white matter of juvenile myoclonic epilepsy (JME) patients with a-priori known damage that correlates with the frequency of generalized tonic-clonic seizures (GTCS) [2]. Objective: (i) To assess the registration accuracy when using iterative registration. (ii) To investigate the correlation between FA reduction and GTCS increases when the registration accuracy is improved. Methods: Sixty-six healthy subjects and ten subjects with JME participated in this study. Diffusion tensor imaging (DTI) was performed by using echo planar imaging at 3T (20xb=1000 s/mm; 3xb=0 (b0 images)). DTI data were pre-processed and FA values were generated using in-house developed software. Analysis I: To assess the registration accuracy, we calculated in a first step the F-map by the quotient of the FA variance between single-step registration