summary: Using machine learning, researchers were able to identify neurological warning signs hidden in the brain’s baseline electrical rhythms, bypassing the need to capture active seizures to diagnose epilepsy. This study demonstrated that advanced pattern recognition algorithms can detect subtle electroencephalogram (EEG) abnormalities associated with genetic epilepsy with high accuracy.
This computational framework builds a customized “dictionary” of waveforms to reveal underlying brain changes and establish a clear path to early intervention and non-invasive precision medicine in children.
important facts
- Diagnostic window bottleneck: Neurologists rely heavily on EEG to diagnose epilepsy, but a standard clinical session provides only a 20-minute snapshot of brain activity, making manual detection extremely difficult if a seizure does not occur spontaneously during recording.
- Building a waveform dictionary: Rather than tracking overt seizures, AI algorithms treat baseline brain wave measurements like an unfamiliar language, identifying frequently repeated electrical patterns and learning their structural meaning in context to zero in on abnormalities that human reviewers miss.
- Seizure-free assay: To test the system, the researchers collected EEG recordings over several days from a panel of more than 40 mice. These included mice with brain wave fluctuations that can cause epilepsy. TSC1 gene. The algorithm analyzed baseline segments with zero seizure activity.
- High precision gene detection: Machine learning approaches successfully distinguish between different genetic backgrounds and TSC1 Highly accurate detection of mutations across 2 of 3 mouse strains purely from baseline EEG.
- Pediatric clinical stages: With support from the Delaware Clinical Translational Research ACCEL program, the team is moving this method into the clinic, analyzing short electroencephalographic recordings of children being evaluated for epilepsy at Nemours Children’s Health.
- reduce family anxiety: Epileptic seizures follow a natural and unpredictable cycle. Early identification of objective biomarkers can eliminate the high cognitive burden and severe anxiety that families experience while waiting for disease onset.
- Possibility of precision treatmentLead researchers Dr. Austin Brockmeyer and Dr. Amanda Hernan note that advanced brainwave typing could prevent doctors from misinterpreting drug efficacy during spontaneous seizure lulls, while also laying the foundation for continuous tracking of related conditions such as autism and ADHD via wearables.
sauce: University of Delaware
Diagnosing epilepsy is not always easy. Seizures often do not occur during routine electroencephalography (EEG), leaving doctors unable to make the direct observations needed to make a definitive diagnosis. Researchers and collaborators at the University of Delaware are working to close that gap by using artificial intelligence to detect early warning signs hidden in the brain’s electrical rhythms.
In a proof-of-concept study in mice, the team showed that their approach can identify subtle EEG differences associated with genetic forms of epilepsy, even in the absence of visible seizures. The findings, published in the Journal of Neural Engineering, set the stage for the next phase of research, which will test the method using EEGs of children being evaluated for epilepsy at Nemours Children’s Health.
EEG dictionary
Neurologists often use electroencephalography to diagnose epilepsy, but routine recordings only provide a snapshot of brain activity over about 20 minutes. If a seizure is not captured during that period, clinicians must look for much more subtle cues that are difficult to detect visually.
That’s where AI comes in. UD researchers’ algorithms are much like language learners encountering an unfamiliar language. First, identify patterns that appear frequently in EEG recordings, learn what those patterns mean in context, and effectively build a dictionary of electrical patterns.
“Our machine learning approach allows algorithms to learn the ‘language’ of brain waveforms and find subtle patterns that humans might miss during manual review,” said Austin Brockmeier, assistant professor of electrical and computer engineering and computer and information science.
Start small with mouse models
When Brockmeier, a faculty advisor in UD’s Interdisciplinary Graduate School of Neuroscience (ING) program, presented his research in computational neuroscience at an ING seminar, it caught the attention of Amanda Hernan, UD associate professor of psychology and brain sciences and biomedical engineering and senior research fellow in Nemours Children’s Health. Hernan, who is also an ING faculty mentor, studies how changes in brain activity affect the thinking and learning of children with epilepsy.
The pair decided to test machine learning using the brain waves of mice that carry epilepsy-causing mutations in the TSC1 gene. The researchers used a panel of more than 40 mice, including animals with and without the genetic mutation, spanning three different genetic backgrounds or strains. They extracted EEG segments from each mouse’s 5-day recording for analysis.
Since the EEG segments do not contain seizure activity, the algorithm needed to detect differences only in the brain’s baseline activity. We were able to distinguish between mouse strains and detect TSC1 gene mutations with high accuracy in two of the three strains.
“These results show that even in the absence of visible seizures, brain wave patterns contain measurable signals of neurological differences,” Hernan said.
Take it to the clinic
The team is now taking the method from the lab to the clinical setting. With funding from the Delaware Clinical Translational Research ACCEL program, Brockmeier and Hernan will next apply their approach to electroencephalographic recordings of children being evaluated for epilepsy at Nemours Children’s Health.
Pediatric EEGs are shorter than the multi-day recordings used in mouse studies, and children exhibit different types of epilepsy. But researchers are optimistic.
“The goal is to identify biomarkers that indicate fundamental changes in the brain’s electrical activity before a seizure occurs,” Hernan said. Early detection can lead to early treatment and reduce uncertainty for families.
That uncertainty is taking its toll, Hernan said. “Seizures occur according to a natural cycle, but if you don’t have a way to know where you are in that cycle, that anticipation can be incredibly anxiety-provoking,” she explained.
Improved pattern recognition could also improve treatment decisions. For example, if a new drug is introduced during a natural lull in seizure activity, its effect may be overestimated.
Looking further ahead, researchers envision a future in which wearable EEG devices can continuously monitor people at high risk of seizures in real time. A similar approach could eventually be applied to other neurological conditions such as autism and ADHD.
“This is a step toward precision medicine,” Brockmeyer said. “Electroencephalography typing may help identify which interventions are most effective for a particular patient.”
Such precision can make a huge difference for families navigating the daily anxiety of epilepsy.
Answers to key questions:
a: By learning the brain’s unique “language” of background rhythms. The University of Delaware’s algorithm builds a custom dictionary of frequently occurring waveforms, allowing researchers to accurately discover microscopic patterns and genetic abnormalities hidden in normal baseline brain activity that are completely invisible to the human eye.
a: The types of epilepsy in children are very diverse, and clinical pediatric EEG provides a much shorter range of data than controlled multi-day laboratory records. Despite these factors, researchers are very optimistic that AI will be successful in isolating early tracking biomarkers.
a: Prevents misinterpreting natural lulls in activity as signs of successful treatment. Because seizures progress through a natural hidden cycle, an objective EEG mapping system can tell doctors exactly where a patient is in that cycle, ensuring that the true benefits of a drug are never overestimated.
Editorial note:
- This article was edited by the editors of Neuroscience News.
- Journal articles were reviewed in full text.
- Additional context added by staff.
About this epilepsy and AI research news
author: Marina Jones
sauce: University of Delaware
contact: Marina Jones – University of Delaware
image: Image credited to Neuroscience News
Original research: Closed access.
“Interpretable EEG Biomarkers in Mouse Neurological Disease Models Using Bag-of-Waves Classifiers” by Maria Isabel Cano Achuri, Montana Kay Lara, Khalil Abed Rabbo, Benjamin T. Wilson, Austin Meek, J. Matthew Mahoney, Amanda E. Hernan, and Austin J. Brockmeier. Neurotechnology Journal
DOI:10.1088/1741-2552/ae4d8c
abstract
Interpretable EEG biomarkers in mouse neurological disease models using bag-of-waves classifiers
Objective.
Electroencephalography (EEG) is a time-series recording of electrical potentials from collective neural activity in the brain. EEG waveform patterns (rhythmic, irregular oscillations, and temporal patterns of sharp waves or spikes) are potential phenotypic biomarkers that reflect genotype-specific neural activity. This is particularly relevant when diagnosing epilepsy without direct seizure observation. This is common in clinical practice and in animal models that often have subtle neurological phenotypes in the absence of overt epilepsy. Here, we investigate genotype prediction from long-term EEG signals in freely moving mice belonging to six groups defined by the presence or absence of neurological disease genotypes (TSC1 gene knockout) in three different inbred lines with different genetic backgrounds.
approach.
We propose a machine learning approach that predicts the genotype of individual mice from the number of waveform occurrences that approximate a short window of EEG. In other words, the dictionary of waveforms is optimized to approximate the window for each genotype, and the vector of waveform occurrences becomes the feature for predicting the genotype by the logistic regression model.
Main results.
Through two cross-validations of waveform dictionary learning and one-individual-out genotype prediction, we found that waveform counts pooled over multiple time segments allowed reliable prediction of mouse lineage with an accuracy of 70% (95% CI 62 to 78) compared to a chance rate of 38%. For two of the three strains, DBA2 and C57B6, the epilepsy genotype was determined reliably by the strain-specific classifier (TSC1 (gene knockout) accuracy was 86% (95% CI 70-101) and 67% (95% 55-79), respectively. There was no evidence of overt seizures or EEG-based seizure detection in these strains of mice. In comparison, a state-of-the-art time series classification approach (Hydra) allows for a comparable higher distortion classification of 98%. TSC1– Genotype prediction for two strains (86% and 71%, respectively), but this method is not interpretable.
significance.
This methodology and results demonstrate the potential of EEG waveforms as interpretable phenotypes and bag-of-waves as feature representations for identifying epilepsy genotypes.
