High capacity computing taking place right here in South Carolina
by Dave Ramsey
SCRA Vice President, Research Services
Readers may already be familiar with some of SCRA's physical assets, including the SCRA Innovation Centers throughout South Carolina that are designed to complement its existing research park system in support of the knowledge economy. The following focuses on some of the digital assets on which SCRA works in conjunction with its partners.
With computers embedded in our cars, phones, appliances, game systems and the ever present laptop, we sometimes forget that computers were originally designed for scientific endeavors where the mathematical problems to be solved were so complex that time and human error made them impossible. E-mail, social networking, and entertainment maybe the attention grabbing applications that everyone recognizes, but there are still large scale science problems being analyzed and some of them are occurring here in South Carolina.
A suite of those applications are being run at SCRA's data center in Charleston. Originally built to manage office services, SCRA now hosts applications, storage and computing for other organizations, one of which is MUSC's Center for Advanced Imaging Research (CAIR). Housed in a small building on Bee Street, CAIR is staffed by researchers at MUSC in multiple disciplines including Psychiatry, Neuro Sciences, Surgery, and Radiology. Its focus has been on studying advanced imaging techniques with a focus on magnet resonance imaging (MRI) of the human brain.
So what does this have to do with computing? MR scanners produce data – not pictures or film. Some of the new techniques produce so much data that a study of just 10 individuals would fill the average PC disk. That's the amount of data generated by an MRI scanner in one or two days! That data has to be moved off of the scanner through high speed networks to large disk arrays at SCRA to begin the computing process.
The data from an MRI scanner represents tiny magnetic signals given off in small – sometimes less than a cubic millimeter – space in the brain. Millions of these cubes, called voxels, make up the magnetic image of the brain and its surroundings. One image can be scanned in seconds and much of the research being conducted is looking at what happens in the brain over time as a stimulus is presented to the person being analyzed. As a result, hundreds of these "pictures" that contain millions of cubes are recorded in something akin to a movie – a 3D movie in this case. But it's not just the sheer volume of data that presents a computing challenge, its searching this data for patterns that represent different tissues, water flow, blood oxygen levels and other factors that make analysis with traditional PC's impossible.
The process often starts with aligning each image taken so that each voxel in each image is of exactly the same spot in the brain. The computer must make millions of small adjustments to compensate for head movements during the scan or differences in how the person was laying during two different scanning sessions. Other issues then must be accounted for. If images are being compared from different scanners or at different times, the entire image must be adjusted to compensate for tiny differences in the magnetic fields. Think of it as trying to compare a picture of a boat taken on a sunny day and a foggy day, to see if an anchor is on deck in each. You have to get rid of the fog or tone down the brightness to see the differences.
But this is all just a prelude. Once the images are prepared, the science begins. Looking for changes in the level of oxygen in the blood at one spot for example, may tell the researcher that the spot is more active than its surroundings. The differences in the signals in these two spots may vary by only a couple of percent and if the analysis is going to be used for surgery or as a target for a stimulator, the location absolutely has to be correct. Another example might be analyzing how water flows throughout the brain may show where the edge of damage from a stroke is. But the computer must analyze each possible direction that water can flow from each voxel and there are dozens of those. Multiply that by millions of voxels and tracing water flow suddenly becomes a mammoth task.
To solve these problems, SCRA employs multiple computers arranged in clusters to solve these analytical problems. Utilizing commercial and academic software along with programs from SCRA itself, these problems are often broken down into smaller tasks each of which can be done in parallel with each being assigned to a separate CPU. Tasks that might take your PC months to compute, can be done in these clusters in hours, thus making the researcher more productive and getting new therapies to market sooner.
High capacity computing, networking, storage, backup all make scientific analysis such as MRI processing possible. Unlike sending an email to a person who you could call, walk to and talk, or send a written note, scientific computing has no alternative. As science advances, computing is no longer helpful, it's a necessity and some of this deep scientific computing is proudly being carried out right here in South Carolina communities.
