The magnitude and spatial extent of variance in the structure and function of the human brain across the population is unknown. Traditional atlases have attempted to compensate for this lack of knowledge by linear compressions and expansions of volumes of tissue using a template, typically derived from one, or at best a few, individuals. To determine this variability and to develop an atlas compatible with today's high resolution imaging technologies, a large number of subjects must be examined in vivo at a macroscopic level (1 mm3) and at a microscopic level (approximately 10-100 mm) from postmortem tissue to evaluate cyto- and chemoarchitecture. A probabilistic data structure is also required to manage the descriptive approach to the resultant data set. The product provides a mathematically rigorous and defensible means by which to compare results across individuals, experiments and laboratories as well as the first valid means for exploring detailed structure-function relationships among individuals in the population. Of value not only in basic human neuroscience, such an approach also has diagnostic capabilities in that a quantifiable determination of abnormality or a change in function as the result of conventional or experimental therapies can be determined in an automated, quantifiable and objective means.
<p>
Eight laboratories in seven countries on three continents have now joined the effort to develop the probabilistic atlas and reference system. A goal of 7,000 subjects studied with 1 mm3 structural, multispectral MRI data sets has now reached the point where 2,250 studies have been completed. Each subject has a detailed demographic history including information about race, ethnic origins, habits, diet, education and occupation. Behavioral data is acquired through handedness testing, neurological examinations as well as neuropsychological and psychiatric questionnaires. On a selected subset of individuals, functional imaging with fMRI, PET and event-related potentials are acquired to develop a set of functional landmarks that can be correlated, across the population, with structural anatomy, as well as cyto- and chemoarchitecture. The data set includes 342 twins of which half are mono- and half dizygotic. 5,800 subjects have DNA samples collected for genotype-phenotype-behavioral correlations. Much of the initial work required for this program involved calibrations of instruments since it was the intent of the consortium to identify sites that had different scanners, computer platforms and software algorithms, thereby forcing the consortium to develop the tools for interoperability, ensuring an extensible and generalizable product. This required the use of "smart phantoms;" a set of subjects that would travel from site to site to ensure that methodological differences among sites was both minimized and well characterized. Algorithm development also occurred in a parallel fashion at each site since it was not known in advance which computational strategies would provide the best results. The analytic process including: preprocessing for alignment/registration, intensity normalization, RF field inhomogeneity correction and site-specific scanner aberrations as well as segmentation, tissue classification, regional and local structure parcellation, surface extraction, and linear and nonlinear warping strategies were all developed using this multisite approach. Once a set of algorithms had been developed for each step in the process, they were internally tested within the consortium and then delivered to an outside group for independent beta testing in comparison with other algorithms purported to do the same task. This evaluation was then published and the optimal algorithm selected for use in the ICBM analysis pipeline. Techniques have been developed for the cryopreservation and sectioning of postmortem heads, including modeling of postmortem changes. Selected brain areas have been subjected to cyto- and chemoarchitectural analysis across individuals (typically, N=10) and a set of functional tasks are being piloted for the development of functional landmarks using fMRI, PET and ERP.
|