MR Morphometry by Means of Deformation Field Analysis

Brain morphometry techniques which are based on structural MR tomography enable one to examine the anatomy of human brains in vivo. These techniques can also be used to search for structural characteristics being related to mental or neurological diseases. Such characteristics may help to understand the mechanisms of these diseases, and may be relevant for their therapy or diagnosis. Considering that the resolution of routine MR techniques for in vivo scans of the whole brain is limited to approximately 1mm, MR morphometry usually examines the macroscopic structure of human brains. We have employed a new technique which is based on the registration of the MR datasets of individual brains (sources) to a target brain: The registration algorithm aims to map anatomically corresponding regions of the source brains and the target brain onto each other. The registration process consists of a global, affine transformation and a following local, elastic warping. The latter defines, for each source dataset, a deformation field: This is a three-dimensional vector field, which assigns to each voxel of the target brain a deformation vector, pointing to the corresponding position in the source data set. Thus, the deformation fields relate the anatomical structure of the source brains to the target brain. From these deformation fields, we have derived voxel-wise maps of the local volume changes. The maps of different subjects are overlaid with each other, thus enabling a voxel-wise statistical analysis of local structural differences. In a next step, we compared the locations of structural differences with maps of cytoarchitectonical areas, which were obtained by microscopic analysis of post-mortem brains. As the position of cytoarchitectonical areas relative to macroscopic landmarks is not fixed, these areas are represented by probabilistic maps. The probabilistic maps had been warped to the same target brain, which was used for the morphometric analysis. We have applied this technique for a comparison of the MR datasets of 12 schizophrenia patients, 12 unaffected siblings, and 24 matched normal control subjects. Main regions of differences between patients and controls were the amygdala (left), insula (left), Heschl's gyrus (left), gyrus parietalis inferior (left, right), and gyrus frontalis medius (left, right). Supported by the Deutsche Forschungsgemeinschaft (Schn362/13–1,-2) and by the National Institute of Health (MH52176–10).