Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GEFM: Graph-Enhanced EEG Foundation Model

Limin Wang, Toyotaro Suzumura, Hiroki Kanezashi

TL;DR

GEFM tackles the data-label scarcity of EEG by combining a graph-based inter-channel module with a masked autoencoder foundation model. By inserting a two-layer GNN before the BENDR encoder and standardizing sequence lengths across datasets, GEFM captures both spatial and temporal EEG structure; experiments across three downstream tasks show that GCN with edge weights, especially with a linear sequence-length adjuster, yields consistent improvements over baseline. This work establishes a robust EEG foundation model that generalizes across tasks and base configurations and suggests future exploration of explainability and broader task deployment. It also provides a practical blueprint for integrating graph structures into EEG foundation modeling and highlights the relative strengths of graph-enhanced architectures in this domain.

Abstract

Electroencephalography (EEG) signals provide critical insights for applications in disease diagnosis and healthcare. However, the scarcity of labeled EEG data poses a significant challenge. Foundation models offer a promising solution by leveraging large-scale unlabeled data through pre-training, enabling strong performance across diverse tasks. While both temporal dynamics and inter-channel relationships are vital for understanding EEG signals, existing EEG foundation models primarily focus on the former, overlooking the latter. To address this limitation, we propose Graph-Enhanced EEG Foundation Model (GEFM), a novel foundation model for EEG that integrates both temporal and inter-channel information. Our architecture combines Graph Neural Networks (GNNs), which effectively capture relational structures, with a masked autoencoder to enable efficient pre-training. We evaluated our approach using three downstream tasks and experimented with various GNN architectures. The results demonstrate that our proposed model, particularly when employing the GCN architecture with optimized configurations, consistently outperformed baseline methods across all tasks. These findings suggest that our model serves as a robust foundation model for EEG analysis.

GEFM: Graph-Enhanced EEG Foundation Model

TL;DR

GEFM tackles the data-label scarcity of EEG by combining a graph-based inter-channel module with a masked autoencoder foundation model. By inserting a two-layer GNN before the BENDR encoder and standardizing sequence lengths across datasets, GEFM captures both spatial and temporal EEG structure; experiments across three downstream tasks show that GCN with edge weights, especially with a linear sequence-length adjuster, yields consistent improvements over baseline. This work establishes a robust EEG foundation model that generalizes across tasks and base configurations and suggests future exploration of explainability and broader task deployment. It also provides a practical blueprint for integrating graph structures into EEG foundation modeling and highlights the relative strengths of graph-enhanced architectures in this domain.

Abstract

Electroencephalography (EEG) signals provide critical insights for applications in disease diagnosis and healthcare. However, the scarcity of labeled EEG data poses a significant challenge. Foundation models offer a promising solution by leveraging large-scale unlabeled data through pre-training, enabling strong performance across diverse tasks. While both temporal dynamics and inter-channel relationships are vital for understanding EEG signals, existing EEG foundation models primarily focus on the former, overlooking the latter. To address this limitation, we propose Graph-Enhanced EEG Foundation Model (GEFM), a novel foundation model for EEG that integrates both temporal and inter-channel information. Our architecture combines Graph Neural Networks (GNNs), which effectively capture relational structures, with a masked autoencoder to enable efficient pre-training. We evaluated our approach using three downstream tasks and experimented with various GNN architectures. The results demonstrate that our proposed model, particularly when employing the GCN architecture with optimized configurations, consistently outperformed baseline methods across all tasks. These findings suggest that our model serves as a robust foundation model for EEG analysis.

Paper Structure

This paper contains 26 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Foundation Models
  4. EEG Foundation Models
  5. EEG Models with Graphs
  6. Graph-Enhanced Models in Other Domains
  7. Methods
  8. BENDR
  9. Our Proposal: GEFM
  10. Alternative Architectural Variants
  11. Incorporating Graphs Into Other EEG Foundation Models
  12. Experiments
  13. Dataset
  14. Pre-training Dataset
  15. Downstream Datasets
  16. ...and 11 more sections

Figures (1)

  • Figure 1: Comparison between two model architectures during pre-training and fine tuning. Figures (a), (c) and (e) represent BENDR kostas2021bendr, while (b), (d) and (f) represent our proposed model, GEFM.