About 2.2 million people in the United States suffer from epilepsy. Due to varying factors contributing to the disorder, the Epilepsy Foundation estimates that for around half the people diagnosed the cause is unknown. Though there are some warning signs of an imminent seizure, the events are largely unpredictable.

IBM and University of Melbourne researchers are looking to change that. According to Wired, the researchers are training a neural network to track brain wave patterns with the hope that it may one day predict an oncoming seizure.

The researchers are utilizing IBM’s TrueNorth chip, “a novel neuromorphic computing platform consisting of (1) million neurons that operate with ultra-low power consumption,” according to the researchers. They recently posted a proof of concept paper online regarding the chip’s ability to analyze neurophysiological signals such as electroencephalogram (EEG) and local field potentials (LFP).   

In the study, participants underwent an experiment where they were asked to clench their left and right hands. While performing the tasks, the participants were hooked up to an EEG that monitored and recorded their brain waves.

The dataset was then used to train the deep learning system. Then, the researchers tested out its prediction powers.

“The maximum classification accuracy we currently achieve via the TrueNorth–compatible network is 76 percent,” wrote the researchers. Ongoing “work is focused on further improving the classification performance and to demonstrate real-time operation.”

Stefan Harrer, one of the researchers behind the work, told Wired that they want to implement the chip in a wearable device that works with a brain implant, and monitors the wearer’s brainwaves around the clock. This may provide researchers with enough brainwave data to predict seizure events.

The paper, according to Wired, will be presented in May at the ACM Computing Frontiers conference. 

This article originally appeared in R&D.