EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

EPI-GRAPH: An Adaptive and Explainable Graph Neural Network Framework for Personalized Epileptic Seizure Prediction Using EEG Signals

Benish William, Munazzah Munwer, Rubab Haider, Abdul Rauf, Majid Hussain

The Asian Bulletin of Big Data Management · 31.03.2026 · 6 (1) · 582-597

DOI: 10.62019/grjjx660 ↗

Kurz gefasst

Fachartikel in der Zeitschrift The Asian Bulletin of Big Data Management, online erschienen am 31.03.2026. Verfasst von Benish William und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multichannel EEG seizure prediction is fundamentally a graph-structured problem: electrodes interact, not independently. We model this directly. EPI-GRAPH constructs a functional connectivity graph G = (V, E) over 23 scalp electrodes, where each edge (i,j) exists if the Pearson correlation r_xy exceeds a threshold τ_FC = 0.70, and learns seizure-discriminative representations using a two-layer Graph Attention Network (GAT). The harder problem is not representation, it is personalization. A single global model degrades sharply on unseen patients (baseline F1 = 0.65 on CHB-MIT), yet always defaulting to full retraining is computationally indefensible. We introduce Cost-Adjusted Performance Score: (CAPS = ΔF1 / adaptation time), , which selects the most resource-efficient adaptation strategy subject to a clinical viability threshold of F1 ≥ 0.88. Evaluated under a leave-one-patient-out protocol on CHB-MIT across four strategies fine-tuning, knowledge distillation, sub-graph focusing, and full retraining, our Adaptive Cost-Aware Strategy Selection (ACAS) algorithm achieves a mean F1 of 0.90 while reducing average adaptation time by 68% relative to full retraining. Gradient-based channel attribution identifies electrodes aligned with clinically reported seizure foci. The result is a framework that treats accuracy, efficiency, and interpretability as co-equal constraints rather than a hierarchy.

Zitieren

APA

William, B., Munwer, M., Haider, R., Rauf, A., & Hussain, M. (2026). EPI-GRAPH: An Adaptive and Explainable Graph Neural Network Framework for Personalized Epileptic Seizure Prediction Using EEG Signals. The Asian Bulletin of Big Data Management, 6(1), 582-597. https://doi.org/10.62019/grjjx660

Vancouver

William B, Munwer M, Haider R, Rauf A, Hussain M. EPI-GRAPH: An Adaptive and Explainable Graph Neural Network Framework for Personalized Epileptic Seizure Prediction Using EEG Signals. The Asian Bulletin of Big Data Management. 2026;6(1):582-597. doi:10.62019/grjjx660

BibTeX
@article{William2026,
  author = {William, Benish and Munwer, Munazzah and Haider, Rubab and Rauf, Abdul and Hussain, Majid},
  title = {EPI-GRAPH: An Adaptive and Explainable Graph Neural Network Framework for Personalized Epileptic Seizure Prediction Using EEG Signals},
  journal = {The Asian Bulletin of Big Data Management},
  year = {2026},
  volume = {6},
  number = {1},
  pages = {582--597},
  doi = {10.62019/grjjx660}
}
Kostenlos starten →