CCS Seminar
Friday - October 12, 2007
12:00 noon
Physics Research Building - Room 595
Professor Eric Schwartz - Cognitive & Neural Systems - Boston University
Professor of Cognitive & Neural Systems, Electrical & Computer Engineering and Neurobiology & Anatomy

"Computational Issues in Brain Imaging"

Magnetic resonance based human brain imaging studies require the experimental measurement, mathematical representation and digital manipulation of data that is embedded in highly convoluted cortical surfaces. In this talk, an overview of the computational issues associated with the measurement of functional architecture in the brain will be reviewed. Functional architecture studies are illustrated with the example of quasiconformal map complexes. These are physiological representations of the surface of the retina, relayed to the cortex in the form of multiple copies, or "maps" with shared boundary conditions, of a strongly non-linear, spatially warped retinal visual pattern. The two-dimensional dipole pattern, familiar from electrostatics, has provided a conjecture for the basic structure of these maps. Recently, this conjecture has been verified in both monkey and human brain in the form of the "wedge-dipole" model, using a variety of brain imaging methodologies. These experimental studies are dependent on access to maximally accurate, near-isometric surface flattening methods. Critical to these results are current methods for brain flattening, based on the computation of exact minimal geodesic paths on polyhedral surfaces, together with metric multi-dimensional scaling. Metric distortion in the range of 5-10% is achievable by full distance matrix flattening, with computation times (16 Gbyte 2Ghz Opteron) of roughly ten hours for (10k polygon) cortical surfaces spanning V1, V2 and V3, i.e. most of the occipital pole. These methods, which would greatly benefit from super-computer acceleration, have demonstrated that the detailed topographic structure of human and macaque visual cortex (in areas V1, V2 and V3) is very similar both across the two species, and across individuals in both species.

Recent publications describing this work can be found at http://eslab.bu.edu/publications/publications.php (articles and abstracts)

Supported by NIH/NIBIB EB1550