Enhanced Graph Attention Network by Integrating Transformer for Epileptic EEG Identification

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalography (EEG) signal classification is critical for diagnosing and monitoring neurological disorders.
• Existing EEG classification methods struggle with complex signal dynamics and patient population generalization.

Purpose of the Study:

• To develop an advanced EEG signal classification method integrating Graph Attention Networks (GAT) and Transformer models.
• To improve the modeling of intricate relationships and context-dependent patterns within EEG data.

Main Methods:

• Integration of GAT and Transformer models for EEG signal classification.
• Leveraging dynamic attention mechanisms to adapt to the variable relevance of brain regions.
• Utilizing the CHB-MIT dataset for evaluating performance on interictal, ictal, and normal EEG patterns.

Main Results:

• The proposed GAT and Transformer integrated approach demonstrated superior performance compared to state-of-the-art algorithms.
• The dynamic attention mechanism effectively captured nuanced EEG patterns across diverse subjects and seizure types.
• The framework successfully distinguished between interictal, ictal, and normal EEG patterns.

Conclusions:

• The combination of GAT and self-attention mechanisms offers a promising avenue for enhancing EEG signal classification accuracy and reliability.
• This approach has the potential to significantly improve EEG-based diagnostics and the management of neurological disorders.