IBM researcher Stefan Harrer and a team at the University of Melbourne are building a system with a neural network, or computer software that mirrors the web of neurons in the human brain, Wired reports. The network, which works similarly to those used by Facebook to identify photos and Androids to recognize commands, reads brainwaves to learn how to flag epilepsy. “We’re trying to extract all the meaningful information from all the background noise. We want to be able to detect a specific seizure for a specific patient,” Harrer told the magazine.
Harrer and the team run the neural network on an experimental IBM chip called TrueNorth. Unlike other neural networks, which usually run across many machines in large computer data centers, TrueNorth is flexible. It doesn’t require much power and could eventually run on a laptop, tablet or phone.
Eventually, the chip could lead to a device that works with a brain implant to monitor for seizures 24/7 and notify patients before they happen. “That’s the only way this technology will have an impact beyond cool research papers,” Harrer said, as quoted by Wired.
The system grew out of research from a previous study at the University of Melbourne. Scientists there collected data from a less complex implant that took EEG readings from epilepsy patients over three years. Now, Harrer and his team are using the data to train their neural network. Even though a seizure warning system is still years in the making, Harrer and University of Melbourne researchers are optimistic about its prospects. In the future, the system could even prevent seizures entirely by detecting them at onset and sending out electrical impulses to stop them. “Our aim is to replace broken neural systems with machines–machines that can interact with the brain in a very natural way,” Dean Robert Freestone, a senior research fellow at the University of Melbourne, told Wired.
REFERENCE: Fierce Medical Devices; 11 APR 2016; Emily Wasserman