Authors :
Presenting Author: Derek Hu, PhD – California State University, Long Beach
Blanca Romero Milà, MS – University of California Irvine
Mandeep Rana, MD – Carilion Clinic
David Adams, MD – Children's Hospital of Orange County
Daniel Shrey, MD – Children’s Hospital of Orange County
Shaun A. Hussain, MD, MS – Division of Pediatric Neurology, Department of Pediatrics, UCLA Mattel Children's Hospital, David Geffen School of Medicine
Beth Lopour, PhD – UC Irvine
Rationale:
The development of EEG biomarkers for the diagnosis and treatment of epilepsy is heavily reliant on visual review by electroencephalographers. In the case of Infantile Epileptic Spasms Syndrome (IESS), the classic EEG biomarker known as hypsarrhythmia is often confounded by low interrater reliability, contributing to diagnostic delay and hampering efforts to predict response to treatment. Here, we use scalp EEG Pattern Identification and Categorization (s-EPIC), an unsupervised time-frequency image analysis previously used for EEG biomarker discovery in Lennox-Gastaut syndrome (Hu et al. Epilepsia 2024), to identify and classify abnormal EEG waveforms associated with IESS.
Methods:
We retrospectively identified 20 subjects with IESS and 20 approximately age-matched healthy controls and collected 10-minutes of clean interictal scalp EEG during NREM sleep for each subject. EEG were analyzed in the time-frequency domain to identify events of interest (EoIs) containing continuous time periods with high power. Each EoI was characterized based on its height (range of significant frequencies), length (duration in seconds), spread across electrodes, density in time-frequency space, and mean power in the delta, theta, alpha, sigma, beta, and gamma frequency bands. EoIs were sorted into 96 feature categories, with each category defined by the frequency band with the highest power and either high or low height, length, spread, and density, relative to the associated median value across all EoIs.
Results:
We identified 16,445 EoIs across 40 subjects, with 50.7% of EoIs originating from IESS subjects. Ten of the 96 categories contained significantly more IESS EoIs than control EoIs with a large effect size (p< 0.05, permutation test, Bonferroni corrected, Cliff’s δ > 0.5). In these 10 IESS categories, 94.8% of EoIs had a high height ( >22 Hz), 60.1% had high length ( >355 milliseconds), and 100% had high spread ( >25.9%) and density ( >9.5%). Beta and gamma band EoIs had high potential as IESS biomarkers, as 96.4% of all beta events and 92.1% of all gamma events in the ten categories belonged to IESS patients. A support vector machine was used to classify IESS from controls based on each subject’s number of EoIs in the ten categories. The algorithm was evaluated using a five-fold cross validation, yielding a 92.5% accuracy in differentiating IESS from control populations.
Conclusions:
The identification of time-frequency EoIs, coupled with feature categorization, revealed significant EEG waveforms associated with IESS that could be used to discriminate IESS subjects from healthy controls. This approach to biomarker discovery could be applied to IESS diagnosis, as well as evaluation and prediction of treatment response and the prediction of epileptogenesis.
Funding:
This work was supported in part by the John C. Hench Foundation