M3T-attention: a multi-level multi-scale temporal attention transformer for EEG hand movement trajectory decoding.

Area of Science:

• Neural Engineering
• Human-Computer Interaction
• Biomedical Signal Processing

Background:

• Brain-computer interface (BCI) technology shows promise for neural engineering and human-computer interaction.
• Decoding upper-limb movements from electroencephalography (EEG) signals is a key research area.
• Predicting continuous 3D movement trajectories from EEG faces challenges like low signal-to-noise ratio, inter-subject variability, and decoding complexity.

Purpose of the Study:

• To enhance the accuracy of decoding continuous 3D hand movement trajectories from EEG signals.
• To address the limitations of existing methods in capturing complex motor control patterns.
• To propose a novel deep learning framework for improved BCI performance.

Main Methods:

• Development of a Multi-level Multi-scale Temporal Attention Transformer framework (M3T-Attention).
• Extraction of temporal features across multiple time scales from EEG signals.
• Integration of features using cross-scale attention mechanisms for nonlinear mapping to 3D kinematic parameters (position, velocity, acceleration).
• Training and validation using the WAY-EEG-GAL dataset.

Main Results:

• The M3T-Attention model achieved high prediction accuracy, with Pearson correlation coefficients (PCCs) of 0.8816 (X-axis), 0.8841 (Y-axis), and 0.8711 (Z-axis).
• The proposed method demonstrated robust performance across all subjects, outperforming existing state-of-the-art approaches.
• Comparative experiments, statistical significance analysis, and ablation studies validated the model's ability to capture neural coding patterns.

Conclusions:

• The M3T-Attention framework significantly enhances the decoding performance of movement trajectories from EEG signals.
• The study offers a novel approach for BCI applications in complex motor control scenarios.
• The source code for the M3T-Attention model is publicly available, facilitating further research and development.