A New Approach to Predicting Epileptic Seizures
Torrents of data produced by implanted microelectrodes could finally yield a prediction system
Photo courtesy sonya hearn
Sonya Hearn’s brain waves could help predict seizures.
In July 2006, after suffering from epilepsy for more than 30 years, 41-year-old Sonya Hearn arrived at an unusually comfortable corner room on the eighth floor of Columbia University Medical Center, in New York City. During her 20-day stay there, she had several epileptic seizures while doctors recorded the electrical activity of her brain through electrodes leading out of an 8-centimeter hole in her head.
Such observation is standard for epilepsy patients, because it allows doctors to pinpoint the part of a patient’s brain where the seizures originate. But the data that neurologists gleaned from Hearn’s brain was anything but standard. While at Columbia, Hearn was the first to have a new kind of brain-wave recording device implanted, a device that neurologists hope will lead to a way to predict seizures—and someday, a way to prevent them.
Anticonvulsant drugs fail to work for about 25 percent of people with epilepsy, roughly 10 million people worldwide. For this group, a dozen or so research labs are exhaustively mining brain-wave data for patterns that reliably predict an oncoming seizure.
Since the 1970s, neuroscientists have tried—unsuccessfully—to find predictive patterns in the data, which come from a set of standard 4-millimeter-wide electrodes that sit on the surface of the brain. But for Hearn and six others treated at Columbia so far, the measurements also came from an additional array of 96 closely packed 3-micrometer microelectrodes that actually penetrated the cortex.
Researchers can collect more useful information from the smaller electrodes, according to Columbia neurophysiologist Catherine Schevon. ”We’re finding that there’s a lot of activity going on at this very tiny resolution area that we had no idea about before,” she says.
The microelectrodes sample data at extremely high rates—30 000 times per second, compared with 500 times per second for standard macroelectrodes. Faster sampling means that the microelectrodes can pick up higher-frequency brain waves (1000 hertz or more), called fast ripples, which may play a role in seizures.
About 50 blocks south of the hospital, computer scientist David Waltz’s team at Columbia’s Center for Computational Learning Systems is parsing the data from two of Schevon’s seven patients—all 8 terabytes’ worth.
To begin processing the data, Waltz’s team had to first decide where in that continuous stream of brain waves to look for patterns before the seizure. ”We don’t even have a good definition of what ’before’ is,” says Philip Gross, senior staff associate at the center. ”It’d be great to find a 10-minute precursor, but maybe those things don’t exist. Maybe it’s 10 seconds or 24 hours. No one knows.”
And they’re going to need a lot more data to find out. ”It’s probably too much to hope that any single technique could work across all known forms of epilepsy. There’s incredible variety,” says Gross.
Schevon’s team is one of two research groups studying seizure data from implanted microelectrodes. The other group, led by bioengineer Bradley Greger and neurosurgeon Paul House, at the University of Utah, in Salt Lake City, has recorded data from three patients so far. In April, Greger began an official data-sharing collaboration with the Columbia group. ”This is absolutely necessary, because no single group has enough information to look at all types [of epilepsy],” he says.
The researchers’ long-term goal is to design machine-learning interfaces that could learn what brain-wave features predict seizures in individual patients. Hypothetically, the researchers say, this system could eventually take the form of an implanted ”brain pacemaker,” stimulating the brain to prevent the seizure from happening in the first place.