Illinois researchers, in collaboration with the Mayo Clinic, have developed a machine-learning-based approach that uses alpha-rhythm-related features to determine the potential for epilepsy and identify the seizure-generating side of the patient's brain.
Epilepsy is a neurological disease that causes unprovoked seizures and affects approximately 1% of the world's population. Chronic epilepsy could lead to permanent neuronal damages and inhibit brain function. Currently, clinicians visually interpret scalp electroencephalograms (EEGs) and MRI scans to diagnose epilepsy. These EEG recordings typically take between 20-60 minutes, which may not be enough time to record any epileptiform activity or subtle abnormalities in the brain. Patients with epilepsy will commonly produce "normal" EEG data which may delay their treatment.
Illinois researchers, in collaboration with the Mayo Clinic, have developed a machine-learning-based approach that uses alpha-rhythm-related features to determine the potential for epilepsy and identify the seizure-generating side of the patient's brain. The results were reported in a paper titled, "Electrophysiological correlates of brain health help diagnose epilepsy and lateralize seizure focus". Yogatheesan Varatharajah, the first author of this paper, was named a finalist in the 2020 Engineering in Medicine and Biology Society Student Paper Competition, chosen from over 1000 submissions. Varatharajah is a graduate of electrical and computer engineering at Illinois and will be joining the department of bioengineering as a research faculty this summer.
“The recent research by Yoga is an important advance that uses machine learning to uncover subtle abnormalities in scalp EEG previously reported as normal that identify patients with focal epilepsy,” said Dr. Gregory Worrell of the Mayo Clinic.
The research team evaluated alpha-rhythm in the EEG as a potential biomarker for brain health. Although alpha-rhythm abnormalities are recorded in an EEG, they are not currently a part of the diagnostic criteria. Researchers compared EEG data of healthy individuals with patients with drug-resistant focal epilepsy. The team first extracted the alpha rhythm features from the EEG data of healthy individuals and compared them with those of the epilepsy population. Next, they trained a machine learning model with the alpha features to classify whether the data is from healthy individuals or those with epilepsy and whether seizure foci was on the left or right hemisphere of the brain.
"We find that, when the seizure generating side of the brain is the patient’s dominant hemisphere (left hemisphere for right-handed patients), the alpha abnormalities are worsened globally (everywhere in the brain), and particularly in the regions generating seizures" said Varatharajah. "So by knowing the dominant brain hemisphere of the individual and how severe the abnormalities are, we can predict which side of the patient’s brain is generating seizures."
Through identifying subtle abnormalities using signal processing methods and machine learning, this approach may achieve a more timely epilepsy diagnosis and ultimately provide better care for patients.
“Working closely with our Mayo collaborators, Yoga has made several contributions to improving the treatments for patients suffering from epilepsy” said his Ph.D. advisor Professor Ravishankar Iyer. “This work highlights a combination of signal processing and machine learning methods that can help diagnose epilepsy even when experienced clinicians are doubtful. Yoga is the first to demonstrate this result, and his collaborators at the Mayo Clinic are excited about this new possibility. Through our partnerships with medical providers like Mayo Clinic, there is a great potential for societally impactful research like this.”
This research was supported through the Mayo Clinic & Illinois Alliance, the National Science Foundation, the National Institute of Health and an IBM faculty award.