i-EPIC: A Data-Driven Time-Frequency Approach for Identifying intracranial EEG Biomarkers of Seizure Generating Tissue
Abstract number :
1.177
Submission category :
2. Translational Research / 2C. Biomarkers
Year :
2025
Submission ID :
788
Source :
www.aesnet.org
Presentation date :
12/6/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Blanca Romero Milà, MS – University of California Irvine
Marco Pinto-Orellana, PhD – University of California Irvine
Atsuro Daida, MD, PhD – Saitama Children's Medical Center
Sotaro Kanai, MD, PhD – Division of Pediatric Neurology, Department of Pediatrics, David Geffen School of Medicine at the University of California, Los Angeles, California, USA
Naoto Kuroda, MD, PhD – Wayne State University
Daniel Shrey, MD – Children’s Hospital of Orange County
Eishi Asano, MD, PhD – Wayne State University
Hiroki Nariai, MD, PhD, MS – Department of Pediatrics, Division of Pediatric Neurology, David Geffen School of Medicine at the University of California, Los Angeles, California, USA
Beth A. Lopour, PhD – Department of Biomedical Engineering, University of California, Irvine, Irvine, CA, United States
Rationale: Identifying seizure-generating tissue is crucial for successful epilepsy surgery. Current methods are time-consuming, expert-dependent, and yield seizure freedom rates of only about 50%. High-frequency oscillations (HFO) and interictal epileptiform discharges (IEDs) have a strong association with epileptogenic tissue, but low specificity has limited their use. Moreover, HFOs and IEDs lack precise, quantifiable definitions; visual EEG interpretation remains the gold standard, which is subject to bias and low interrater reliability. Therefore, we developed an automated, data-driven algorithm to identify novel intracranial EEG (iEEG) biomarkers in the time-frequency domain for accurate delineation of seizure-generating tissue.
Methods: Our method, intracranial EEG Pattern Identification and Classification (i-EPIC), detects Events of Interest (EOIs) across 1-500 Hz and characterizes them using eight time-frequency features: frequency of peak power, maximum power, area, duration, height, density, density of simultaneous events, and density of surrounding events. We applied i-EPIC to human iEEG from 14 seizure-free post-surgical UCLA patients, using cross-validation to identify EOI categories capable of differentiating between resected and non-resected channels. These categories were then combined using machine learning to increase accuracy and generalizability across subjects. We validated i-EPIC on 87 seizure-free Detroit Medical Center (DMC) subjects, comparing its performance to a spike detector, a spike-ripple detector, and four published HFO detectors. Finally, we applied i-EPIC to 8 non-seizure-free UCLA subjects and 40 DMC subjects to evaluate the method’s performance in suboptimal surgical outcomes.
Results: We identified 10 categories of EOIs that occurred significantly more frequently in resected than non-resected tissue. Using a support vector machine to combine all 10 categories, we achieved classification accuracy with median AUROC = 0.75, outperforming six established algorithms (Figure 1A-C; median AUROC = [0.47, 0.52, 0.58, 0.61, 0.52, 0.61]). Validation in the DMC dataset confirmed i-EPIC’s ability to generalize, as the median AUROC was significantly higher than 5 out of 6 existing algorithms (Figure 1D-F). Moreover, i-EPIC successfully differentiated between UCLA seizure-free and non-seizure-free subjects, demonstrating significantly different distributions for AUROC, AUPRC, and PPV, a distinction that the other biomarkers failed to achieve.
Conclusions: Our novel, data-driven approach identified promising iEEG biomarkers for seizure-generating tissue, enabling classification accuracy that rivals or exceeds traditional biomarkers. Notably, this approach demonstrated robust performance under the most stringent validation conditions – application to an independent dataset without any modifications to the algorithm. i-EPIC’s ability to generalize to new, unseen data suggests high potential to improve epileptogenic tissue delineation and surgical outcomes in drug-resistant epilepsy.
Funding: This work was supported by the National Institute of Neurological Disorders and Stroke of the NIH under Award Number R01NS116273.
Translational Research