EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Convolutional Neural Networks (CNNs) struggle with long-term dependencies in EEG decoding due to limited perceptual fields.
• Capturing both local and global temporal features is crucial for accurate EEG signal analysis.

Purpose of the Study:

• To propose a unified framework, EEG Conformer, for EEG classification that integrates local and global feature extraction.
• To enhance the interpretability of EEG decoding models through visualization techniques.

Main Methods:

• Utilized a compact Convolutional Transformer architecture.
• Employed 1D temporal and spatial convolution layers for local feature extraction.
• Integrated a self-attention module to capture global correlations within local features.
• Implemented a fully-connected classifier for EEG signal categorization.
• Developed a visualization strategy for projecting class activation mapping onto brain topography.

Main Results:

• Achieved state-of-the-art performance on three public EEG datasets.
• Demonstrated effectiveness in EEG-based motor imagery and emotion recognition tasks.
• Showcased the potential of EEG Conformer as a new baseline for general EEG decoding.

Conclusions:

• EEG Conformer successfully unifies local and global feature extraction for robust EEG decoding.
• The proposed method offers enhanced interpretability and achieves superior performance compared to existing approaches.
• EEG Conformer shows significant promise for advancing the field of brain-computer interfaces and EEG analysis.