Image processing was performed on a computer workstation (Silicon Graphics Inc, Mountain View, Calif) using the locally developed BRAINS (Brain Research: Analysis of Images, Networks, and Systems) software package. Detailed descriptions of image analysis methods are provided elsewhere.34- 36 In brief, the T1-weighted images were spatially normalized and resampled so that the anterior-posterior axis of the brain was realigned parallel to the anterior-posterior commissure line, and the interhemispheric fissure was aligned on the other 2 axes. The T2- and proton density–weighted images were aligned to the spatially normalized T1-weighted image using an automated image registration program.37 These images were then warped into standardized stereotaxic Talairach atlas space38 to generate automated measurements of the frontal, temporal, parietal, and occipital lobes; cerebellum; and subcortical regions.39 To further classify tissue volumes into gray matter (GM), white matter (WM), and CSF, we used a discriminant analysis method of tissue segmentation based on automated training class selection that used data from the T1, T2, and proton density sequences.40 This method allowed us to identify the range of values that characterized GM, WM, and CSF in our multispectral data (10-70 for CSF, 70-190 for GM, and 190-250 for WM). Each voxel was given an intensity value that was based on the weights assigned by the discriminant function and that reflected the relative combination of GM, WM, and CSF in a given voxel that allowed us to correct for partial volume.40 Intraclass correlations for this automated tissue segmentation analysis ranged from 0.97 to 0.98. Regions of interest (ROIs) examined in the patient-control comparison included total brain tissue volume; lateral ventricle volume; cortical sulcal CSF volume; tissue and CSF volumes for the frontal, temporal, and parietal lobes; and cerebellar tissue volume.