Generative and recognition models for neuroanatomy.

As with previous critiques of voxel-based morphometry (e.g., see Bookstein, 2001, Mehta et al., 2003), Christos Davatzikos (2004) reprises issues that have engaged the functional neuroimaging community for many years. In this instance, the issue is the distinction between multivariate and mass-univariate analyses of imaging data. Put simply, Davatzikos is pointing out that pathology can be expressed, anatomically, in a distributed and complicated fashion over the brain. Critically, its expression in one part of the brain may depend on its expression elsewhere. Characterizing these interregional dependencies requires a multivariate model (e.g., see Ashburner et al., 1998; Bookstein, 1984) of how pathology causes anatomical changes. In the former example, canonical variates analysis (CVA) was used to assess gender differences using deformation-based morphometry. Deformation-based morphometry represents an analysis of the deformation fields that spatially normalize images. However, differences in brain anatomy may not be completely encoded by these deformations; local structural differences may persist after spatial normalization. Voxelbased morphometry (VBM) was introduced to characterize these differences.