Diagnosing Epilepsy from Scalp EEG with and Without IEDS: A Network-based Analysis
Abstract number :
3.123
Submission category :
2. Translational Research / 2C. Biomarkers
Year :
2022
Submission ID :
2204503
Source :
www.aesnet.org
Presentation date :
12/5/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:25 AM
Authors :
Patrick Myers, MSE – Johns Hopkins University; Kristín Gunnarsdóttir, PhD – Johns Hopkins University; Adam Li, PhD – Computer Science – Columbia University; Jacob Feitelberg, BS – Johns Hopkins University; Vladislav Razskazovskiy, BS – University of Pittsburgh Medical Center; Alana Tillery, BS – Johns Hopkins University; Satvik Saripalli, GED – Pacific Ridge High School; Dale Wyeth, REEGT – Thomas Jefferson University Hospital; Edmund Wyeth, REEGT – Thomas Jefferson University Hospital; Khalil Husari, MD – Johns Hopkins Hospital; Nirav Barot, MD, MPH – University of Pittsburgh Medical Center; Michael Sperling, MD – Thomas Jefferson University Hospital; Joon-Yi Kang, MD – Johns Hopkins Hospital; Jorge Gonzalez-Martinez, MD, PhD – University of Pittsburgh Medical Center; Sridevi Sarma, PhD – Johns Hopkins University
Rationale: While routine scalp EEG is standard of care for characterizing seizure-like events as epileptic or not, its usefulness for diagnosing epilepsy is highly debated. The EEG reader visually searches for specific events in the record, primarily interictal epileptiform discharges (IEDs). However, the reliability of these events is questionable. The presence of IEDs on epileptic scalp EEG varies from 29-55% and it has been widely reported that EEG artifacts are often misinterpreted as IEDs. The overinterpretation and/or overreliance on visual inspection of scalp EEG is a major contributor to the 30% misdiagnosis rate of epilepsy. Therefore, a more objective analysis of scalp EEG is needed.
Methods: We developed a novel algorithm that processes a single 20–30-minute acute interictal EEG snapshot to predict whether a patient has epilepsy or not. The algorithm extracts network-based features that characterize the underlying dynamics of the EEG signals, which are then fed into a logistic regression model. This approach focuses on the dynamic interactions between brain regions, specifically the "source-sink" or inhibitory surround properties of the epileptogenic zone, rather than the gold-standard abnormal events in single EEG channels (IEDs). Patient-specific Dynamic Network Models (DNMs) are estimated based on artifact-removed EEG data, from which unique source-sink features that capture the unique characteristics of the epileptic brain at rest are characterized._x000D_
_x000D_
Our novel network-based features are combined with more standard frequency-based features to compute an epilepsy risk score from the model. We provide a suggested threshold on this risk score to classify patients as having epilepsy.
Results: We conducted a retrospective analysis consisting of 291 patients from Jefferson Hospital, Johns Hopkins Hospital, and University of Pittsburgh Medical Center, to test our predictive model. Each patient’s final diagnosis was confirmed with an admission to the EMU before being added to the study. The EEG record for the patient’s first visit to the corresponding epilepsy center was included in the study. Based on this record, patients were categorized into the groups: not epilepsy (n=116), epilepsy with no IEDs present (n=64), or epilepsy with IEDs present (n=111). We hypothesized that features extracted from our DNMs could distinguish the epilepsy patients from the non-epilepsy patients, regardless of whether the EEG had IEDs or not. To test our hypothesis, we combined our novel network-based features with more standard frequency-based features in a simple Logistic Regression model. Based on the suggested threshold demonstrated by the red horizontal line in Figure 1, we achieve a predictive accuracy of 0.802, a sensitivity of 0.829 and a specificity of 0.764.
Conclusions: Our approach could significantly change current clinical practice by improving the usefulness of Scalp EEG in the diagnostic process. Given Scalp EEGs relative convenience and inexpensive nature, this paradigm shift would be favorable.
Funding: Alvin and Fanny B. Thalheimer Foundation; National Science Foundation SBIR Phase I
Translational Research