Authors :
Mark Hays, PhD – Neurologic Solutions
Presenting Author: Golnoosh Kamali, PhD – NeuroLogic Solutions
Mary Rose Streett, BS – Johns Hopkins School of Medicine
Ronald Miranda, BS – University of Maryland Medical Center
Guang Ren, BS – University of Maryland Medical Center
Anant Walia, MD – University of Maryland School of Medicine
Sarah Johnson, BS – Thomas Jefferson University
Dale Wyeth, MBA – Thomas Jefferson University
Andrew Gotshalk, BS – NeuroLogic Solutions
Michael R. Sperling, MD – Jefferson Comprehensive Epilepsy Center, Thomas Jefferson University
Jennifer Hopp, MD – University of Maryland School of Medicine
Khalil Husari, MD – Johns Hopkins School of Medicine
Jorge González-Martínez, MD,PhD – University of Pittsburgh Medical School
Sridevi Sarma, PhD – Johns Hopkins University
Rationale:
Epilepsy affects millions worldwide with 5 million new diagnoses each year. Misdiagnosis rates are alarmingly high (20-42%), with potentially severe consequences. A false positive can lead to inappropriate medication and unnecessary employment and social restrictions. Conversely, a false negative increases the risk of seizure recurrence, status epilepticus, and premature death. The current standard for epilepsy diagnosis relies on clinical history, neurological examination, and scalp electroencephalography (EEG). While EEG can confirm epilepsy if abnormalities like interictal epileptiform discharges (IEDs) are visually detected, the sensitivity of EEG in detecting these abnormalities is low (29-55%). As a result, accurate diagnosis often requires multiple visits, taking months to years. There is a critical need for a more accurate method to diagnose epilepsy on the first visit, enabling patients to receive appropriate treatment sooner.
Methods:
We gathered scalp EEG data from the Johns Hopkins Hospital and University of Maryland Medical Center for patients undergoing routine scalp EEG for a first-time seizure-like event. Dynamical network models (DNMs), derived from approximating short windows of EEG as linear time invariant models, were built to characterize the network’s N to N channel interactions across time. Two DNM-derived metrics, neural fragility and source-sink index, have previously been shown to localize the epileptogenic zone (EZ) in intracranial EEG recordings. We hypothesized that these metrics can be applied to interictal scalp EEG to distinguish between epileptic and non-epileptic brains, since the EZ nodes in epileptic brains would have high fragility and behave like strong sinks driven by strong neighboring sources, while these properties would be absent in non-epileptic brains. We trained a logistic regression model with these features to predict epilepsy and non-epilepsy diagnoses in a multicenter cohort of 283 first-time routine EEG patients, including 198 with normal EEG (no IEDs or non-epilepsy-specific abnormalities) that were used in our original retrospective study to validate our patent-protected algorithm known as EpiScalp™ and 85 new patients with abnormal EEG to capture a more general population.
Results:
The EpiScalp™ model was tested on 59 first-time routine EEG patients (36 normal EEG, 23 abnormal EEG) and predicted a definitive diagnosis for 39 routine EEG patients with 95% accuracy (100% sensitivity, 89% specificity). Among the abnormal EEG patients, a population previously unvalidated with EpiScalp™, the model achieved 94% accuracy.
Conclusions:
These findings are significant because they validate that EpiScalp™ can accurately predict the likelihood of epilepsy from a first visit whether IEDs are present or not, enabling clinicians to obtain accurate information about epilepsy risks from inconclusive scalp EEG data to significantly reduce both patient misdiagnoses and the time to diagnosis.Funding:
This study was funded by the Maryland Innovation Initiative and NINDS SBIR Phase 2 grant.