This talk will describe the preprocessing steps involved in the technique of voxel-based morphometry. The procedure begins by segmenting anatomical images of a number of subjects into their different tissue classes, using a
generative modelling approach. The model combines tissue classification, bias correction and nonlinear registration of pre-generated tissue
probability maps. The segmentation results are used to generate approximately rigidly aligned gray and white matter images for each subject.
More precise inter-subject registration is then performed by repeatedly aligning the grey and white matter images to their own mean, using a
high-dimensional warping approach that preserves the one-to-one mapping among brains. Warped versions of the gray matter images are generated, which are locally scaled in order to compensate for expansion/contraction during warping. These data are spatially smoothed, prior to performing voxel-wise statistical tests within the framework of the General Linear Model.