This AI-powered implant chip could offer a deeper, longer look at faulty brain activity
A Northeastern professor is developing a low-power implant that uses machine learning to analyze brain signals in real time. With epilepsy as one test case, the goal is to uncover patterns that brief EEG tests miss.
The brain never clocks out. Whether you’re chatting with a friend, solving an algebra problem, reading a blog post or taking a nap, it’s buzzing with electrical activity. Nerve cells communicate this way, firing up to 1,000 signals every second.
A new device that Northeastern University electrical and computer engineering professor Aatmesh Shrivastava is developing with support from the National Institutes of Health aims to keep tabs on this activity as it unfolds. The chip tracks electrical signals under the scalp and uses AI to analyze the data and pick out meaningful patterns in real time.
Eventually, the device could detect signatures of neurological disorders such as epilepsy, which involves uncontrolled bursts of activity by groups of brain cells firing together. It could also help monitor Alzheimer’s and other conditions involving disrupted brain signaling, and provide insights into ALS, where the brain generates signals that don’t reach their destinations, Shrivastava said.
Scientists detect the brain’s signals with electroencephalogram (EEG) technology developed by German psychiatrist Hans Berger in 1924. It typically works by placing electrodes on the scalp to record waves of activity by large groups of cells.
While EEG provides a useful snapshot, signals overlap and messages get blurred, Shrivastava explained. It’s a bit like multiple people talking over one another on the same phone line.
Measuring under the scalp clears up the confusion by getting closer to the source.
“By going a little bit below, you can actually improve the quality of the signal,” Shrivastava said.
Shrivastava’s device will also record brain activity for much longer periods — days or even months at a time — compared to the typical 20 to 90 minutes that a standard EEG test lasts.
This approach can expose long-term patterns that short recordings miss, Ziv Peremen, chief executive officer of X-trodes Ltd, a company that develops wearable monitoring patches, explained. It can reveal “transitions into and out of abnormal activity, relationships to sleep and behavior, and early signatures that may precede clinical events,” he said.
Beyond capturing individual episodes, continuous monitoring could also provide a more nuanced picture of brain functioning and demonstrate long-term effects of medications, Ivan Gligorijević, chief executive officer and co-founder of mbraintrain.com, a company that develops wearable EEG technology added.
But tracking for that long is practical only if the device uses very little power — after all, you can’t exactly plug a brain implant into the wall every night. Shrivastava’s previous work on low-power and self-powering chips laid the foundation for the new tech, which uses similar principles.
And then there’s the problem of data. Piles of it that someone — or something — must analyze.
Raw EEG output can look like a jumbled mess, Shrivastava said. Alas, the brain doesn’t highlight the useful bits or leave notes to help scientists make sense of it.
That is where machine learning, a form of AI, comes in.
Built directly into the chip, it can break down each recording into attributes — individual characteristics of the signal, such as the height, speed or shape of an EEG wave. Next, it can link these features to any events — such as seizures — that happened during the recording.
Editor’s Picks
The AI will then tag certain combinations of attributes as markers — telltale signatures of seizure activity that vary among people and neurological conditions. A series of unusually sharp, rapid waves, for example, could count as a marker.
“Seizures in different parts of the brain will manifest differently,” Geoff Bobb, executive director of Epilepsy Toronto, a nonprofit, told Northeastern Global News. Ones involving convulsions, unusual sensations or memory problems carry unique markers and require different treatments, he added.
The network within the chip will store the growing list of markers, refining the profile as more data comes in.
This built-in analysis sets the new chip apart from existing continuous sub-scalp systems that act largely as sensors. It can allow physicians to look for potential seizure triggers, Shrivastava said. For example, certain foods, lack of sleep, stress or missed medication might coincide with spikes in activity.
“You can analyze the data, identify disease patterns, identify … aspects of life that you can control,” he explained.
“It is not an intervention,” he clarified, comparing the device to a glucose tracker that regularly updates diabetes patients about their blood sugar.
“Seizures … often follow individual rhythms over days to months,” said Manfred Hartmann, senior research engineer from the Center for Health & Bioresources, a research facility based in Vienna, Austria. Knowing a person’s cycle may make it possible to estimate periods of high seizure risk and warn them ahead of time.”
Detailed recordings can also “show whether someone is truly seizure-free,” he added.
Once the chip works as intended, the prototype will be tested in rats or macaques at the University of Florida.
“This is the kind of innovation our community has been waiting for,” said Tracy Dixon-Salazar, president and chief executive officerof the Lennox-Gastaut Syndrome (LGS) Foundation, a nonprofit dedicated to funding research and spreading awareness of LGS, a rare and severe type of childhood-onset epilepsy.
