Archives
Helping survivors of shaken-baby syndrome
Researchers use Dell computer clusters to model what happens inside babys skull
By ELLEN GOLDBAUM
Contributing Editor
In an emergency room, where tragedy is a daily occurrence, the diagnosis of shaken-baby syndrome still jolts even the most seasoned health-care workers. Prevention efforts have shown success, but an estimated 600-1,400 cases still occur annually in the U.S. The ability to best care for the small victims may depend, in part, on work now under way with UB's Dell computer clusters.
The Applied Computational Mathematics and Mechanics Research team led by Abani Patra, associate professor of mechanical and aerospace engineering, develops computational techniques to use supercomputers to study complex, phenomena, such as volcanic eruptions and the function of car safety seats for children during crashes.
"Engineering analysis boils down to creating good mathematical models to represent the physics," Patra explains. "The Dell cluster enables us to construct more accurate numerical approximations and to examine more modeling assumptions."
While modeling simple, one-time injuries has many things in common with injuries caused by shaken-baby syndrome, Patra says modeling the latter is more complex because of the type and mechanisms of tissue damage that result from the repeated shaking of the brain.
By revealing more about the biomechanics and how such injuries are sustained, Patra's work, along with that of collaborators at Pennsylvania State University, is contributing to the development of better diagnoses of the syndrome and one day may lead to improved treatments for survivors, about half of whom experience significant problems, including blindness, seizures, developmental delays and paralysis.
Before the arrival of the Dell cluster on campus, Patra and his UB colleagues had to make what he describes as "gross simplifications" about the many complex phenomena that occur inside the skull while pediatric brain injury is being sustained.
"We've had to grossly simplify details of brain geometry, like what happens to the connections between blood vessels and brain tissue," he says. "We've had to neglect the effect of the complex interactions with the cerebro-spinal fluid and had to simplify how all these different components interact. "Our hope is that now that we have access to a much faster computer that the number and severity of our assumptions will be sharply reduced," he adds.
Patra explains that "for distributed memory computing, we �cut up' the problem into discrete pieces on which each processor can work simultaneously, yet independently."
Once each of 600 Pentium-4 processors in the Dell cluster available to all UB researchers has finished its piece of the problem, the scientists use the cluster's Myrinet network to integrate all of the individual solutions, eventually producing a much more accurate representation of what happens inside the baby's head.
Patra says that loosely coupled phenomena, such as the calculation of the odds that a flipped coin will land on its head, can be done on computers with loose interconnections between processors, since each spin of the coin is an independent action and information does not have to be exchanged between individual processors.
By contrast, pediatric brain injury, he explains, is an example of physical phenomena that can be studied in sufficient detail only on machines that have very fast connections between individual processors, which is a key function of parallel computing and the new Dell cluster.
"The cluster," he notes, "immediately allows us to work on the computational scale where the cutting-edge work is being done in terms of methods we can develop to more efficiently create accurate simulations."