DISTRIBUTED GRAPHICS SUPERCOMPUTING FOR NEUROINFORMATICS

University Of California Los Angeles, Los Angeles CA

Investigators

Abstract

Rapid advances in the field of neuroinformatics have resulted in the generation of massive amounts of image and statistical data concerning the human brain. These advances have resulted in a need for scientific visualization of diverse kinds of dense data sets. From the widespread demand for the visualization of simple scalar fields, to the growing demand for novel techniques to render static and dynamic vector and tensor fields, the ability to answer these challenges has relied on corresponding advances in computational techniques. The Laboratory of Neuro Imaging has been a frontrunner in the adoption of cutting edge technology to understand the developmental and degenerative changes of the human brain. Our group has gained worldwide recognition for the development of high-order mathematical approaches to the investigation of brain registration, the analysis of variance between and within populations. and the visualization of these data. The techniques used, however, now must accommodate four dimensions, as research underway at LONI and elsewhere has begun to produce complex time- varying and multidimensional statistical fields. The dynamic visualization of these new data opens a new frontier in understanding the dynamic brain, enabling investigators to interact with the statistics of structural and functional change. In response to these computational challenges, a group of neuro- and biomedical scientists with common interests and computational needs have come together to seek funding to enhance the graphics and computational capabilities of a shared supercomputer. This group has a critical need for potent visualization hardware, extremely fast, parallel CPU architectures and considerable amounts of data storage and transfer capabilities to satisfy the demands placed on equipment by four dimensional volume viewing, intricate surface extraction, nonlinear warping and the volumetric blending of complex polygonal objects with both raw data and the results of statistical analyses. The requested package of graphics, computational and networking equipment would be available on a shared basis for interactive use locally and offline processing for both CPU and graphics intensive jobs remotely. The equipment package selected is compatible with and fully leverages an existing SGI Onyx2 graphics supercomputer. Simulations and benchmarking confirmed the ability of this instrumentation to fulfill the needs of the participants. An administrative plan is already in place by which the equipment can be managed equitably. Software that would greatly enhance the research projects exists to take advantage of this new equipment. Programming experts in this architecture are already members of the staff. Technical and management personnel also are part of the funded group of participants. The existing collaborations of the participants and the common algorithmic requirements will enable sharing of computer code, analytic procedures, computational strategies and infrastructural capabilities. The provision of such an instrumentation package would enhance the productivity of ongoing funded research and foster the development of leading edge technology and applications for all participants.