According to the World Health Organization, approximately 50 million people worldwide have epilepsy and 30 percent of epilepsy cases do not respond well to treatment. In addition, the National Society for Epilepsy has recorded more than 40 different epilepsy syndromes, including those caused by congenital defects and neural-tissue trauma. Sudden and periodic seizures are a common symptom of Epilepsy.
The neural networks being modeled at Argonne, located just outside of Chicago, IL, will provide a means of mapping the barely understood connections that serve as fertile ground for epileptiform activity - the convulsive electrical discharges that may signal a seizure.
Rick Stevens, Associate Laboratory Director for Computing, Environment and Life Sciences at Argonne, compares neural network models to models of other physical phenomena, including those of earthquakes, and likens epileptiform activity to the foreshocks that precede an earthquake.
Pinpointing what drives epileptiform activity could lead to “significant pharmaceutical outcomes” and broad implications in surgery, and the development of a more accurate treatment device, worn or implanted, which acts as a “brain pacemaker.” Such devices already exist and monitor anomalous brain activity, but the trick is finding out how to “budge it back to a normal state,” says Stevens.
One the most challenging aspects of the project is finding out how complex and numbered the neural connections should be in order to impart noteworthy results. Computer scientist Mark Hereld is one of the researchers working with Stevens to determine the optimal scale and computational level of each model used in their epilepsy simulations.
Argonne employs a network of super computers, dubbed JAZZ, which includes 380 individual server machines. Also on the premises is IBM’s Blue Gene/P, which has 160,000 processors and is one of the most advanced super computers in the field.
The human brain is thought to contain at least 100 billion neurons. Test models at Argonne are capable of replicating about a half-million brain cells, but scientific results are better using 10,000.
“We take our simulations and we cut it up into little pieces within the network,” says Hereld. “The pieces are 2-D, like a piece of paper, and they replicate a slice of your brain. The separate computers across the network are then synchronized.”
The precision and versatility of modeling far exceeds methods used by other disciplines to document brain activity, like electroencephalography (EEG), which tracks brainwaves but provides “a funky average,” and high-resolution microscopy, which focuses on too finite of an area, Hereld says.
Hereld also works in tandem with Wim van Drongelen, who leads the partner research at the University of Chicago Hospitals’ epilepsy center. It’s here that they try to recreate the simulated findings with resected tissues and the various excitatory and inhibitory biochemical agents that trigger and propagate seizures, particularly those in pediatric patients.
“The folks in the lab as well as the clinic have an array of these constituents they can work with,” says Hereld.
The project is already testing conventional wisdom. It is a commonly held theory that seizures are brought on by a high level of activity in the brain, but the research at Argonne points toward the opposite. Epileptiform activity is shown to be more associated with low levels or “strength” of biochemical agents.
“At each of our stopping points we observed the behavior and discovered somewhat surprising results by turning down the strength,” says Hereld.
Funding for the project comes from U.S. Department of Energy and the National Science Foundation. Other neural modeling projects include IBM’s Blue Brain Project in Lausanne, Switzerland and a joint venture into artificial brains at IBM’s Almaden Research Center in San Jose, CA.
Chartered in 1946, the Argonne National Laboratory was named the country’s first national laboratory. It is currently one of the U.S. Department of Energy’s largest research centers.