Image processing was conducted using the Voxel-based morphometry toolbox (VBM5.1) (http://dbm.neuro.uni-jena.de/vbm/) implemented in SPM5 running on Matlab 7.5. VBM in SPM5 combines tissue segmentation, bias correction, and spatial normalization into a unified model (19). Hidden Markov Random Fields were applied to improve accuracy of tissue segmentation (medium HMRF 0.3). Otherwise, default parameters were used. Individual brains were normalized to tissue probability maps provided by International Consortium for Brain Mapping (ICBM). A 12 mm FWHM Gaussian kernel resulted in a final smoothing of 14 × 15 × 14 mm3. At the second level, whole brain data were modeled across the groups using analysis of covariance (ANCOVA) with total GM volume and age as covariates. The effects of total GM volume were removed to allow inferences about regional differences in GM volume. An absolute threshold mask of .1 was used. Statistical maps were set at a cluster-level threshold of p<.05, corrected for multiple comparisons using family-wise error (FWE), and a voxel-level threshold of p<.005. To ensure the validity of cluster-level statistics, a non-isotropic smoothness correction was applied (20).
