Spatial normalization is an important preprocessing step used to reduce intersubject anatomical variability in human brain mapping studies. The most basic form of spatial normalization uses landmarks and a 9-parameter affine transform to adjust position, orientation, and size of an individual brain to match a reference brain.
For many years, the 1988 Talairach Atlas brain has served as the ad hoc standard for reporting locations of activation foci in functional brain mapping studies in part to its detailed anatomical labeling:
Talairach J, Tournoux P (1988). Co-planar stereotaxic atlas of the human brain. Thieme, New York. [Amazon]
The anatomical region labels were electronically derived from axial sectional images in the 1988 Talairach Atlas. While 3-D coordinates are precise, quantitative descriptions of locations (and location variances), more traditional (but less quantitative) anatomical descriptions, such as surface anatomy (sulci/gyri) and architectonic areas, are needed to allow comparisons between coordinate-based data and those not using standardized coordinates.
The Talairach Atlas has been digitized and manually traced into a volume-occupant hierarchy of anatomical regions. Hemispheres, lobes, lobules, gyri and nuclei have been outlined and labeled. Gray matter, white matter and CSF regions will also be defined. For cerebral cortex, all Brodmann areas have been traced and expanded into 3-D volumes. The Talairach Atlas includes Brodmann area (BA) labels, but lacks explicit boundaries between BAs and has several inconsistencies. Explicit boundaries were rule-based and primarily justified by using Brodmann's 1909 monograph.
Brodmann area, gyrus, lobe and hemisphere region labels are available in the Talairach Labels database. Both a flat list of every label and a hierarchical list segregated by level are available. The TD will improve the accuracy and consistency with which coordinate-based studies make reference to traditional anatomical nomenclature. Its electronic format facilitates rapid updating if discrepancies are found.
Currently there are nine SP Maps within the database: Caudate, Putamen, Thalamus, Insula, Frontal lobe, Temporal lobe, Parietal lobe, Occipital lobe, and Cerebellum. Probabilities can be queried with the Talairach Daemon. Fifty-percent probability overlays can be displayed in the applet to view the locations most likely associated with the SP Maps. SP Map data were provided by Alan Evans and his research team at the Montreal Neurologic Institute. They were calculated using 50 or more MRI brain volumes that were automatically labeled using the non-linear image-matching ANIMAL algorithm (Collins et al.,1995). This algorithm allows for a large number of degrees of freedom, in principle as many as one per voxel.
The structure probability maps were generated as follows. Using ANIMAL, a single-template MRI brain image that has been labeled (i.e., each voxel is identified anatomically) was 3D-warped to fit MRI brain images from normal subjects. The spatial mapping between the template and subjects can then be applied to the template labels so that each subject brain is now labeled in its original orientation. As a second, and quite separate, step, labeled brains can be mapped into a standardized stereotaxic space. This is done by using a simple 9-parameter transformation to match each labeled brain image against a target stereotaxic image. For this step, the target image is not a single MRI image but the average of 305 MRI images, each mapped into stereotaxic space in an order-independent manner. This average MRI, ICBM305, defines the stereotaxic space in which all the probability maps are expressed. Since the nine parameters do not account for non-linear neuroanatomical variability, the labels for each structure from all brains are not precisely aligned, giving rise to a fuzzy boundary which represents the variability. A probability map for any structure or SP Map is calculated at each x-y-z location where the structure is present. These maps are a measure of the percent incidence of the structure at each of these locations in the defined stereotaxic space.
The probabilities reported by the Talairach Daemon using SP Maps provide insight into anatomical variability. Methods to measure anatomical variability depend on the reference space (in this case, ICBM305) and the transformation used to put each brain in that space (in this case, nine parameters). They are also dependent on the method of mapping labels from a standard to the subject brain. The nine-parameter stereotaxic transformation removes differences in global position, orientation and scale along the three cardinal axes. This provides a standard level of spatial normalization where non-global differences can be visualized and evaluated. Different spaces and transformations will give slightly different probability maps, which should be taken into account when analyses based on these maps are performed.