At the University of Delaware, a team of researchers is teaching artificial intelligence to hear the whispers of epilepsy hidden in the brain's electrical noise—picking up warning signals that no human eye has trained itself to see. The breakthrough comes from a simple insight: seizures might be predictable before they strike, if only we could teach machines to recognize the subtle patterns that precede them.
The challenge facing neurologists is stark. Epilepsy is notoriously difficult to diagnose because seizures often refuse to cooperate with a 20-minute routine brain-wave recording. A patient sits in a clinic while doctors attach electrodes to capture an EEG—and then nothing happens. Without a seizure captured during that narrow window, clinicians are forced to hunt for invisible clues in the electrical squiggles, patterns so faint that even trained eyes struggle to spot them. This diagnostic gap leaves families in painful uncertainty, caught in an anticipatory cycle they cannot predict or control.
Austin Brockmeier, an assistant professor in electrical and computer engineering at UD, and Amanda Hernan, an affiliated associate professor of psychological and brain sciences and biomedical engineering who also serves as a senior research scientist at Nemours Children's Health, decided to approach the problem differently. Rather than asking doctors to become better at visual pattern recognition, they asked: what if a machine could learn the brain's "language" of electrical waveforms? Their algorithm works much like a language learner encountering an unfamiliar tongue—it identifies patterns that appear frequently in EEG recordings and builds a dictionary of what those patterns mean in context, catching subtle differences that humans might miss during manual review.
The researchers tested their approach first in mice, the traditional proving ground for neuroscience breakthroughs. Working with a panel of more than 40 mice, both with and without epilepsy-causing variations in the TSC1 gene, they extracted five days of EEG recordings from each animal. The critical twist: those recordings contained no visible seizures. The algorithm had to detect differences in baseline brain activity alone. And it succeeded. It could distinguish between mouse strains and detect the TSC1 gene variation with high accuracy in two of the three genetic backgrounds tested. The findings, published in the Journal of Neural Engineering, proved that measurable electrical signals of neurological differences exist even without a single seizure appearing.
Now the work moves from laboratory mice to children. Brockmeier and Hernan will apply their algorithm to EEG recordings from children being evaluated for epilepsy at Nemours Children's Health. The transition brings new complexity—pediatric EEGs are shorter than the multi-day mouse recordings, and children present with many different epilepsy types. But the researchers are optimistic about what they might find.
"The goal is to identify biomarkers that flag underlying changes in the brain's electrical activity before seizures occur," Hernan said. Earlier detection could mean earlier treatment and, more importantly, an end to the grinding uncertainty that families endure. If a machine can learn to read the brain's whispers before the storm arrives, diagnosis becomes not something that waits for a seizure to happen, but something that can catch epilepsy in its earliest, quietest moments.