Learning Motor Cues in Brain-Muscle Modulation

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Current brain-muscle modulation studies offer incomplete insights due to analyzing isolated electrophysiological signal properties.
• A comprehensive understanding requires methods that bridge the gap between brain activity (EEG) and muscle responses (EMG).

Purpose of the Study:

• To propose a cross-modal generative model for converting electroencephalography (EEG) signals to electromyography (EMG) signals.
• To investigate the role of motor cues in brain-muscle system interactions.
• To provide a data-driven approach for analyzing brain-muscular modulation.

Main Methods:

• A two-stage generative model was developed to translate EEG to EMG signals.
• Contrastive learning was used to extract shared movement-related information between EEG and EMG.
• Generative adversarial networks (GANs) were employed for the EMG generation stage, conditioned on extracted representations.

Main Results:

• The proposed model demonstrated superior performance in cross-modal EMG generation compared to existing time series methods.
• The model's inference process provided insights into the brain's muscle control strategies during movement.
• Successful conversion of EEG to EMG signals was achieved, validating the model's efficacy.

Conclusions:

• The developed model offers a comprehensive perspective on brain-muscular modulation by linking EEG and EMG signals.
• This research advances the understanding of neural control mechanisms and provides a valuable tool for the neuroscience community.
• The findings highlight the potential of generative models in decoding complex biological signals.