Statistical Learning for Epilepsy

Overseen By Taylor J Abel, MD

Age: < 65

Sex: Any

Trial Phase: Academic

Sponsor: University of Pittsburgh

Disqualifiers: Intellectual disabilities, Abnormal epileptiform activity, others

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: 2

What You Need to Know Before You Apply

What is the purpose of this trial?

This trial explores how the brain processes speech, focusing on the supratemporal plane, which plays a role in hearing and understanding sounds. By using special monitoring methods during necessary medical procedures (sEEG), researchers aim to learn how the brain adapts to different speech sounds and listening situations through Dimension-Based Statistical Learning. The study seeks English-speaking individuals aged 15-25 who are already undergoing sEEG for epilepsy or language mapping and have normal hearing. As an unphased trial, it offers participants the chance to contribute to groundbreaking research on brain function and speech processing.

Will I have to stop taking my current medications?

The trial information does not specify whether you need to stop taking your current medications.

What prior data suggests that this protocol is safe for patients undergoing sEEG?

Research has shown that Dimension-Based Statistical Learning, particularly when combined with machine learning for epilepsy, is generally safe. For example, studies using similar machine learning methods to detect epileptic seizures have not revealed major safety issues. These studies primarily focus on model effectiveness rather than negative effects on patients.

This study also employs sEEG, a method already used in hospitals to monitor brain activity, indicating its safety in regular medical practice. Although specific safety data for Dimension-Based Statistical Learning in humans might not be detailed, the absence of reported problems in similar studies suggests confidence in its safety. ^{1 2 3 4 5}

Why are researchers excited about this trial?

Researchers are excited about the Dimension-Based Statistical Learning technique for epilepsy because it offers a fresh approach to understanding and managing seizures. Unlike traditional treatments that focus on medication or surgical interventions, this method uses advanced data analysis to interpret brain signals, potentially identifying unique patterns associated with seizure activity. By leveraging statistical learning, this technique could lead to more personalized and effective management strategies for epilepsy, offering hope for patients who haven't responded well to existing treatments.

What evidence suggests that Dimension-Based Statistical Learning might be an effective treatment for epilepsy?

Research shows that Dimension-Based Statistical Learning, which participants in this trial will experience, can enhance understanding and prediction of seizures. Studies have found that analyzing various aspects of EEG tests, which measure brain activity, provides a good estimate of seizure risk. Advanced computer techniques, such as CNN-SVM and DNN-SVM models, have successfully identified epilepsy with high accuracy. One study demonstrated that using EEG signals predicted seizures with over 99% reliability. These findings suggest that this approach could effectively aid in understanding and managing epilepsy. ^{4 6 7 8 9}

Who Is on the Research Team?

Taylor J Abel, MD

Principal Investigator

University of Pittsburgh

Are You a Good Fit for This Trial?

This trial is for individuals aged 15-25 with epilepsy who are undergoing sEEG in the supratemporal plane and have normal hearing, vision, cognitive, and speech-language skills. They must be fluent English speakers without a history of autism or ADHD.

Inclusion Criteria

You have normal or corrected-to-normal vision.

My hearing is normal in both ears.

I am between 15 and 25 years old.

Exclusion Criteria

You do not speak or understand English well.

You have serious problems with understanding language or hearing.

You have been diagnosed with autism or ADHD in the past.

Timeline for a Trial Participant

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

sEEG-EEG Recording

Neural activity is measured via simultaneous EEG-sEEG monitoring in the supratemporal plane and other cortical regions as participants listen to acoustic stimuli with manipulated acoustic dimensions and in different listening contexts.

Up to 3 hours per session

1 session

Behavioral Response Collection

Behavioral responses are collected as participants provide category judgments based on perceived phonemes during acoustic stimuli presentation.

Up to 3 hours per session

1 session

Follow-up

Participants are monitored for any adverse effects or changes in neural response post-recording sessions.

4 weeks

What Are the Treatments Tested in This Trial?

Interventions

• Dimension-Based Statistical Learning

Trial Overview

The study explores how the brain processes speech by examining responses to voice onset time (VOT) and fundamental frequency (F0) during routine clinical sEEG monitoring in patients with epilepsy.

How Is the Trial Designed?

Group I: Patient Participants Experimental Treatment 1 Intervention

This single-group study will recruit patients through the PI's clinical practice who are undergoing invasive neurophysiological monitoring (sEEG) with clinically necessary placement of electrodes in the supratemporal plane. All participants will complete the same behavioral response paradigms.

Who Is Running the Clinical Trial?

University of Pittsburgh

Lead Sponsor

Trials

1,820

Recruited

16,360,000+

National Institutes of Health (NIH)

Collaborator

Trials

2,896

Recruited

8,053,000+

Carnegie Mellon University

Collaborator

Trials

80

Recruited

540,000+

National Institute on Deafness and Other Communication Disorders (NIDCD)

Collaborator

Trials

377

Recruited

190,000+

Published Research Related to This Trial

A machine learning model using multi-modality features from 103 children with tuberous sclerosis complex (TSC) achieved a strong predictive accuracy (AUC of 0.812) for epilepsy drug treatment outcomes, highlighting the potential of advanced analytics in clinical settings.

Key factors influencing treatment outcomes included specific EEG characteristics, age of onset, and MRI lesion types, indicating that detailed clinical and imaging data can significantly inform treatment strategies for epilepsy in TSC patients.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Machine learning and statistic analysis to predict drug treatment outcome in pediatric epilepsy patients with tuberous sclerosis complex. Zhao, X., Jiang, D., Hu, Z., et al.[2023]

The study introduces new methods to reduce high-dimensional recordings of epileptic activity into low-dimensional (2D or 3D) representations, making it easier to visualize and analyze seizure prediction data.

Clustering algorithms were evaluated to automatically identify patterns within these low-dimensional descriptors, and the methods were adapted to effectively handle long-term datasets that can span several days or weeks.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Uncovering low-dimensional structure in high-dimensional representations of long-term recordings in people with epilepsy. Rapela, J., Proix, T., Todorov, D., et al.[2021]

The improved manifold adaptive Farahmand-Szepesvári-Audibert (FSA) dimension estimator provides a more accurate measure of intrinsic dimensionality in data analysis, outperforming previous methods like the mode and mean.

Using the corrected median-FSA estimator, researchers were able to identify brain areas with lower-dimensional dynamics that may be causal sources for epileptic seizures, highlighting its potential application in understanding neural dynamics during seizures.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Manifold-adaptive dimension estimation revisited. Benkő, Z., Stippinger, M., Rehus, R., et al.[2022]

The EpiBioS4Rx project aims to identify biomarkers for epileptogenesis after traumatic brain injury (TBI) by analyzing extensive data from both human subjects and animal models, which is crucial for developing antiepileptogenic therapies.

Using advanced data analysis techniques, the project seeks to standardize and interpret complex data from multiple sources, enabling the design of effective clinical trials to prevent or cure epilepsy.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Analytic Tools for Post-traumatic Epileptogenesis Biomarker Search in Multimodal Dataset of an Animal Model and Human Patients. Duncan, D., Barisano, G., Cabeen, R., et al.[2020]

A predictive model developed from MRI scans of 53 subjects (12 left TLE, 21 right TLE, and 20 healthy controls) identified that longer durations of temporal lobe epilepsy (TLE) are linked to reduced functional connectivity and structural atrophy in the ipsilateral temporal lobe, particularly the hippocampus and pallidum.

The study found that significant changes in brain structure and function occur over 10 to 35 years of TLE, indicating a gradual functional isolation of the affected temporal lobe and some impact on the contralateral hemisphere.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Increasing structural atrophy and functional isolation of the temporal lobe with duration of disease in temporal lobe epilepsy. Morgan, VL., Conrad, BN., Abou-Khalil, B., et al.[2022]

The patient-specific algorithm for predicting epileptic seizures, tested on 216 patients with a total of 16,729.80 hours of EEG data, achieved a sensitivity of 38.47% and a low false positive rate of 0.20 per hour.

The method showed statistically significant performance in only 11% of the patients, highlighting the need for further refinement in feature selection to improve the distinction between pre-ictal and nonpre-ictal states.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

A Realistic Seizure Prediction Study Based on Multiclass SVM. Direito, B., Teixeira, CA., Sales, F., et al.[2019]

The developed algorithm effectively detects seizure onset and identifies seizure-generating brain regions using only one or two seizure examples from intracranial electroencephalography (iEEG) recordings of 16 drug-resistant epilepsy patients, demonstrating a high generalization rate of 100% for most patients.

This algorithm outperforms existing methods, including deep learning approaches, by achieving higher specificity (94.84%) and accuracy (95.42%) while maintaining a significantly lower memory footprint, making it a promising tool for clinical decision-making in epilepsy management.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Hyperdimensional Computing With Local Binary Patterns: One-Shot Learning of Seizure Onset and Identification of Ictogenic Brain Regions Using Short-Time iEEG Recordings. Burrello, A., Schindler, K., Benini, L., et al.[2020]

A study of 462 epilepsy patients identified three distinct clusters based on psychosocial health, revealing that those with poor psychosocial health had significant predictors such as failure to achieve seizure freedom and social determinants like income support needs.

The findings suggest that targeted interventions focusing on social support and medication management could improve health outcomes for patients in the 'poor psychosocial health' cluster, highlighting the importance of addressing both medical and social factors in epilepsy care.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Psychosocial profiles and their predictors in epilepsy using patient-reported outcomes and machine learning. Josephson, CB., Engbers, JDT., Wang, M., et al.[2021]

In a study of 91 patients undergoing epilepsy monitoring, 61.5% experienced seizure clustering, which is defined as three or more seizures within 24 hours, with significant risk factors including a history of clustering at home and the presence of mesial temporal sclerosis (MTS) on MRI.

Seizure clustering did not affect the localization of the epileptogenic zone, as the mean interseizure intervals were similar for both clustered and non-clustered seizures, indicating that both types of seizures are equally important for determining the source of epilepsy.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Seizure clustering during epilepsy monitoring. Haut, SR., Swick, C., Freeman, K., et al.[2023]

Citations

1.

pmc.ncbi.nlm.nih.gov

pmc.ncbi.nlm.nih.gov/articles/PMC12353893/

EEG signal dimension is an index of seizure propensity ...

Dimension provided a statistically robust (inverse) estimate of seizure susceptibility of mice, including mice with low seizure frequency or no ...

2.

nature.com

nature.com/articles/s41598-025-95831-z

EEG-based epilepsy detection using CNN-SVM and DNN ...

This study focuses on epilepsy detection using hybrid CNN-SVM and DNN-SVM models, combined with feature dimensionality reduction through PCA.

3.

sciencedirect.com

sciencedirect.com/science/article/abs/pii/S1746809423001805

Novel seizure detection algorithm based on multi- ...

The result shows that the AC, SP and seizure-based SE were 99.36%, 82.98% and 99.41% respectively. In order to solve the problem of unbalanced seizure data ...

4.

frontiersin.org

frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1417748/full

On the performance of seizure prediction machine learning ...

This study presents a patient-specific seizure prediction algorithm applied to diverse databases (EPILEPSIAE, CHB-MIT, AES, and Epilepsy Ecosystem).

5.

onlinelibrary.wiley.com

onlinelibrary.wiley.com/doi/full/10.1002/eng2.12918

Deep learning‐based seizure prediction using EEG signals ...

In this study, we present a method for predicting epileptic seizures using electroencephalogram (EEG) signals.

6.

pmc.ncbi.nlm.nih.gov

pmc.ncbi.nlm.nih.gov/articles/PMC12190198/

Epileptic Seizure Detection Using Machine Learning

This meta-analysis reviews the performance of ML models in seizure detection and analyzes factors such as the model type (deep learning vs. traditional ML), ...

7.

sciencedirect.com

sciencedirect.com/science/article/pii/S0306452225003550

Entropy-driven deep learning framework for epilepsy ...

This paper introduces a novel approach for the automatic identification of epilepsy in EEG signals by incorporating advanced entropy-based measures with modern ...

8.

nature.com

nature.com/articles/s41598-025-18479-9

Epileptic seizure detection from electroencephalogram ...

A review paper has examined various state-of-the-art machine-learning techniques for predicting seizures using EEG signals. To enhance the ...

9.

researchgate.net

researchgate.net/publication/396141455_Interpretable_Machine_Learning_for_Epileptic_Seizure_Detection_on_the_BEED_Using_LIME_with_an_Ensemble_Network

(PDF) Interpretable Machine Learning for Epileptic Seizure ...

This study aims to identify seizures in four different stages among epileptic patients, utilizing the Bangalore Epilepsy Dataset (BEED). This ...