Recently I explored a fascinating project that utilized machine learning to detect epilepsy in children.  The research, which was a collaborative project between Young Epilepsy, UCL Great Ormond Street Institute of Child Health and the University of Cambridge, focused on Focal Cortical Dysplasia, which is a major cause of epilepsy in children.  It describes the way the brain fails to form normally, and because the abnormalities tend to be small, they tend to be very difficult to pick up on MRI scans.

What would make such work even cooler is if they could predict the onset of a fit before it occurs, raising the potential of then intervening to prevent the seizure.  That’s exactly what a team from Rice University have set out to do.

Predicting seizures

Firstly the team developed an algorithm to predict when seizures might occur.  After several iterations and a large dose of testing, it was able to predict seizures at least two minutes before their onset.

“What our system is trying to do is predict and prevent seizures in real time,” says Sarah Hooper, a senior electrical engineering major at Rice University. “The first stage of our system is to record neural activity from the brain. That activity is then sent to our piece of hardware, which has the algorithm that produces a seizure prediction.  Using the electrical signals that are produced in the brain, we can predict if a seizure is going to occur in the next five minutes or so.”

The system contains a feedback loop whereby the hardware communicates with electrodes implanted in the brain whenever a seizure is about to occur.  The electrodes then apply electrical neurostimulation to try and prevent the seizure from occurring.

“Three years ago, the project was basically an idea,” the team say. “About one-third of the three million epilepsy patients in the United States don’t respond to anti-seizure medications. The only option left for those patients is to undergo surgery to remove the part of the brain that is the issue; we hope to replace that option with something a lot less invasive.”

Next steps

The team hope to further develop the device so that it better interacts with the input electrodes.  They also need to work on the way the output is transmitted to neurostimulators.  This work will build on the extensive development that went into the CPU and AI that underpins the device.

Suffice to say, the work is still at a very early stage, and the team themselves admit that it’s likely to be another three to five years before it can even begin clinical trials, let alone hit the market, but it is nonetheless an interesting glimpse into the kind of work being done.

Check out more about the project via the video below.