When epilepsy can’t be controlled with drugs, neurosurgery is sometimes curative, if the seizures are coming from discrete brain tissue that can be safely removed.

Finding these diseased areas, however, can require invasive surgery to place grids of electrodes on the brain’s surface. That’s followed by long–term, 24–hour EEG monitoring – typically for a week – until a seizure happens.

Neurosurgeons then use this data to map a surgical path. But to actually remove the diseased tissue, a second operation is needed.

That’s enough to deter many families from epilepsy surgery. But what if seizure origins could be mapped without having to actually observe a seizure?

Joseph Madsen, MD, director of Epilepsy Surgery at Boston Children’s Hospital, and Eun–Hyoung Park, PhD, a computational biophysicist in the Department of Neurosurgery, think they have a way to do that – with an algorithm originally used for economic forecasting.

If validated, the algorithm could potentially allow patients with epilepsy to have EEG monitoring right in the operating room. They could then have the seizure–causing tissue removed on the spot, reducing the current two–stage procedure to a single operation.

The algorithm, called Granger causality analysis, is based on a statistical approach developed by Sir Clive Granger, which earned him a shared Nobel Prize in Economics in 2003. It was designed to predict economic relationships, such as those between wealth and consumption, exchange rates and price levels, and short and long–term interest rates.

Publishing in the journal Neurosurgery, Madsen and Park describe how they adapted the method, using it to calculate the probability that activity at one brain location predicts subsequent activity at other brain locations strongly enough to be considered causative.

To test it, they analyzed EEG data from 25 patients who had previously had long–term EEG monitoring at Boston Children’s. Instead of looking at data captured during a seizure, they looked at data recorded between their seizures — specifically, the first 20 seizure–free minutes of monitoring.

“We know that the diseased brain network responsible for the seizures is there all along,” explains Madsen. “So rather than wait for the patient to have a seizure, we set out to find patterns of interaction between various points in the brain that might predict where seizures would eventually start.”

Running the data through the Granger algorithm, Madsen and Park then created maps of the causal relationships in each person’s seizure network. These, in turn, were superimposed over images of the brain.

They then compared their predictions with the actual causative regions identified by long–term EEG monitoring – sometimes many days later – by ten board–certified epileptologists. The regions they predicted and the actual regions correlated strongly.

Madsen and Park have further shown that their calculations can be done quickly enough in the operating room to potentially influence surgical decision–making. They now are investigating how the Granger causality method can best augment readings of EEGs by trained neurophysiologists.

“We still need to validate and refine our approach before it can be used clinically,” notes Madsen. “But we are hopeful that these advanced computer applications can help us treat more children with epilepsy – with less risk and lower cost.”