...on data
We have chosen to work within two different application domains, fluid dynamics, specifically in the study of turbulence, and neuroscience. This will ensure that we develop general algorithms and software tools that may be applied to a variety of visualization problems.
Turbulence
Experimental Data
Recent advances in digital-imaging technology have permitted a new experimental attack on turbulence. In particular, high-resolution, relatively high framing-rate 10242-pixel CCD's now allow the full dynamic range of scales to be recorded. The research group of Dr. Paul Dimotakis plans to construct a laser-scanning imaging facility that will be capable of acquiring 1024 times 1024 times 1024 samples of turbulent flow 1000 times a second. This facility will produce a terabyte of data per second. Figure A presents an isosurface of a 512 times 512 times 256 turbulent flow dataset acquired from his current laser-scanning facility. Three-dimensional analysis and visualization limitations associated with such data remain a serious impediment. Current visualization implementations force subsampling of the full-resolution data to reduce the data size handled in the computer at one time, defeating the purpose and insight that full-resolution data can provide. Similar considerations apply to turbulence data generated computationally.
Simulation Data
The Rayleigh-Taylor instability is an unsteady flow that occurs when a lighter fluid is accelerated into a heavier one. To study it, scientists have performed computer simulations that encompass uniform block grids describing the state of the data in terms of density and flow momentum. This simulation studies the reaction in a period of a few seconds, with a thousandth of a second intervals. However, the actual computation required several days of CPU-time performed on the Pacific-Blue computer at Lawrence Livermore National Laboratory by Dr. Andrew Cook.
The Rayleigh-Taylor instability data set was computed on a grid of 256x256x512, with several scalar and vector values produced at each grid point. Thus, each simulation time step outputs a dataset approximately one-half gigabyte in size. Multiplying this number by the thousands of time intervals computed for this one simulation the total size for this one dataset exceeds one terabyte. Likewise, as newer runs are performed, the grid resolutions will be increased, producing even larger datasets. Beyond the initial data produced by the simulation, scientists are interested in computing additional scalar, vector, and tensor information from the raw simulation results. Through its involvement in the ASCI program, the Caltech team has access to this data, and has been actively working to develop visualization methods to view, explore, and analyze it. Numerous other ASCI datasets of this magnitude and larger will be available to the Caltech team in the future.
Neuroscience
We are working with two neuroscience groups, one at the Caltech Biological Imaging Center, and the other at the National Center for Microscopy and Imaging Research at UC San Diego.
Multi-Echo MRI of Developing Brains
The mission of the Caltech Biological Imaging Center (BIC) is to develop new technologies for imaging biological structure and function. The first step towards understanding the relationship between structure and function in developing brains is to produce a series of detailed models of the shape of the brain as it develops over time. Towards this end, the BIC is collecting numerous MR datasets of mouse embryos ranging in age from gestation day 7 to neonates at approximately half-day intervals. Similar efforts involving live quail have just begun and the recent establishment of a breeding colony of mouse lemur at Caltech will allow in vivo experiments with these small primates. Currently, the datasets are manually traced, classified, and annotated to produce the structural information needed for the atlases. This is a labor-intensive, time-consuming process which impedes the dissemination of this valuable data, and inhibits the collection of additional datasets.
The MR data is multi-echo, with typically 3 echos, and in some cases utilizing several recycle times, along with gradient echo images. This process produces multi-dimensional volume datasets, where 3 to 10 values may be associated with each voxel. These datasets may be as large as 512 times 512 times 256 and each value may be stored as a floating point number. A dataset with these characteristics is well over 2 Gigabytes in size. To date most of the datasets from the BIC have not been this large, due in part to the lack of tools to process and analyze this amount of data. Provided with the proper analysis and visualization tools the BIC could easily acquire over 100 Gigabytes of data in the next 3 years.
Electron Tomography of Neuronal Structures
The National Center for Microscopy and Imaging Research (NCMIR) is currently utilizing high voltage electron microscopy (HVEM) to view the fine features of neuronal spiny dendrites and organelles such as the Golgi apparatus and endoplasmic reticulum. The problem with such data has always been in analyzing and understanding complex 3D geometries and extracting quantitative information. These problems have been overcome recently by the development of computational methods for deriving 3D data from HVEM datasets using electron tomography. In electron tomography, as is the case with the more familiar tomographic methods used in medical imaging, different views of the specimen are used to create a 3D volume reconstruction which is amenable to quantitative analysis using various software packages. In the past, the production of these tomographic volumes was prohibitively time-consuming. With the increase in computing power and the development of digital cameras, an entire tilt series can be acquired and aligned in under 3 hours. Through the use of parallelized reconstruction algorithms, a 400 megabyte volume can be computed in less than 30 minutes. Given these advances in electron tomography it is possible to acquire and generate 10 Gigabytes of data per week with the electron microscope at NCMIR.