EEG-EMG spatiotemporal cross-attention fusion network for functional upper-limb movement classification.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Technology

Background:

• Unimodal electroencephalography (EEG) and electromyography (EMG) for motor intention decoding have limitations in robustness and information representation.
• Hybrid EEG-EMG brain-computer interfaces (BCI) offer potential but face challenges in signal alignment, fusion modeling, and clinical generalization.

Purpose of the Study:

• To develop and evaluate STCAFusion, a spatiotemporal cross-attention framework for integrating EEG and EMG signals.
• To enhance the accuracy and reliability of hybrid BCI for upper-limb motor control.

Main Methods:

• Proposed STCAFusion, a framework utilizing multi-band dual-branch convolutional encoders and parallel temporal and spatial cross-attention modules.
• Integrated EEG and EMG signals to model inter-modal correlations across time and space.
• Evaluated on a new dataset of synchronous EEG-EMG recordings from 12 subjects during Reaching and Lifting tasks.

Main Results:

• STCAFusion achieved average accuracies of 84.15% (Reaching) and 95.22% (Lifting).
• Outperformed existing EEG-EMG fusion baselines by 3.4% and 1.8% in the respective paradigms.
• Visualized attention weights revealed meaningful spatiotemporal EEG-EMG coupling patterns.

Conclusions:

• Cross-attention-based multimodal physiological signal fusion shows significant potential for reliable hybrid BCI.
• The developed framework can advance the design of wearable devices for upper-limb control and rehabilitation.