Within-group nonlinear registration improves between-group Voxel-Based Morphometry

Accurate registration is imperative for VBM methodologies. Typically, linear and/or nonlinear registration is used to conform all brains to the same shape, orientation and size. We suggest disease-specific reference targets to reduce normal inter-subject anatomical variability while at the same time enhance pathologically-induced differences between groups. These targets can be created by averaging subjects following linear registration, and we propose to use nonlinear registration of subjects to their respective group target prior to performing VBM analyses. Our results demonstrate that: (A) this methodology increases the number, extent and statistical significance of clusters; (B) a smaller smoothing kernel can be used to improve spatial localization; and (C) there is no detectable increase in the false-positive rate. Within-group nonlinear registration therefore improves the detection ability in between-group voxel-based morphometry. Abbreviations used: CDA (community dwelling adults), GLM (generalized linear model), GM (grey matter), HC (hippocampus), HA (HC atrophy), IGK (isotropic gaussian kernel), MRI (magnetic resonance imaging), SPM (statistical parametric mapping), TLE (temporal lobe epilepsy), VBM (voxel-based morphometry), WM (white matter)