For people with epilepsy, simply knowing when a seizure will occur could make a huge difference for their quality of life and reduce the risk of bodily harm. Although some people have warning signs, seizures are generally unpredictable and potentially life-threatening. Epilepsy affects around 2.9 million adults and 465,000 children nationwide and costs billions in healthcare expenses, according to the U.S. Centers for Disease Control and Prevention.

"About a third of the people with epilepsy experience recurrent seizures despite treatment, making post-diagnostic seizure monitoring extremely important," said V John Mathews, professor of electrical and computer engineering.

Mathews teamed up with Epitel Inc., a Salt Lake City-based health solutions company, to improve the algorithms for the company’s wearable device to detect and monitor seizures caused by epilepsy. The Information Processing Group Mathews leads specializes in signal processing and machine learning.

"Seizure detection uses electroencephalogram data, and that data is nothing but signals," Mathews said. "Engineers can do a lot to help with diagnosis, prognosis, and monitoring of diseases using biological signals including EEGs, electrocardiograms, and electromyograms."

A device for continuous seizure monitoring

Current EEG systems are bulky and impractical for daily use. So, Epitel created a single-channel EEG system that is capable of continuously recording EEGs while a person continues their daily activities. The company’s first system had four sensors. Mathews’ team worked on a new system that has just one.

The device, called Epilog, is being developed for two purposes. One is to warn people with epilepsy of an oncoming seizure so they can take actions that will help prevent harm. A warning system would also contact caregivers or family members in the case of a seizure. A second purpose is to support epileptologists to more efficiently review long-term data, which could help to identify triggers of seizures.

"Working with companies enables us to translate our research into commercial products," Mathews said. "In this case, the company was already on a path to commercialization. So, knowing that what we do could be used in practice is very satisfying."

Building better algorithms

Mathews and his team improved the company’s seizure detection by developing three algorithms. Performance evaluation of the new algorithms showed both higher sensitivity and lower false alert rates than competing algorithms.

Their contributions included a two-stage algorithm for seizure detection that focused on electrographically focal seizures; an algorithm for determining the seizure types from EEG data; and an algorithm to improve seizure detection decisions through post-processing the preliminary outputs of the system.

To train and test the algorithms, Epitel provided data from over 700 people. Data for each person typically spanned 3 to 6 days of continuous recordings.

The team attributed the improved performance to three key enhancements. First, their machine learning model included memory at the input that analyzed EEG features from adjacent segments. Secondly, the iterative learning system utilized the decisions made for prior segments. Finally, adding the second stage further analyzed segments in the region where the seizure starts. The combination of all three together provided the best results.

Beyond the results

"This project provided a great opportunity for the graduate students and postdoctoral researcher to get experience working with a company," Mathews said.

The team included Shini Renjith, who finished her postdoctoral work at Oregon State, and Karthik Gopalakrishnan and Nazifa Tabassum, doctoral students in electrical and computer engineering who are continuing work on the project.

This research is focused on seizure detection only, but the team has already started on the next step, which is seizure prediction. Mathews says the goal is to give people enough warning prior to an oncoming seizure so they can take precautionary measures.

"The efforts not only address a critical gap in epilepsy management but enable the translation of seizure alerting and seizure forecasting into inpatient and outpatient settings," said Mark Lehmkuhle, founder and CTO of Epitel.