Decoding muscle activity via CNN-LSTM from 3D spatiotemporal EEG.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Non-invasive brain-computer interfaces (BCIs) aim to reconstruct muscle activity from electroencephalogram (EEG) signals.
• Extracting muscle-related signals from complex EEG data presents significant challenges due to signal mixing from various cortical regions.

Purpose of the Study:

• To introduce and evaluate a novel deep learning method for estimating muscle activity from non-invasive EEG signals during the grasp and lift (GAL) task.
• To compare the performance of the proposed deep learning model against traditional decoding methods.

Main Methods:

• EEG features were extracted from delta, theta, alpha, beta, and gamma frequency bands, computed as envelopes similar to electromyogram (EMG) envelopes.
• A deep learning model combining convolutional neural networks (CNN) for spatial and long short-term memory (LSTM) for temporal information was employed.
• The CNN-LSTM model was benchmarked against multivariate linear regression (mLR) and multilayer perceptron (MLP) decoding methods.

Main Results:

• The CNN-LSTM model achieved an average normalized root mean square error (nRMSE) of 0.21 ± 0.05 and a coefficient of determination (R²) of 0.54 ± 0.17.
• The model demonstrated a significant improvement over mLR and MLP (p < 0.016), with higher EEG frequencies showing greater decoding effectiveness.
• Correlation coefficients (CC) between estimated and actual muscle activity averaged 0.76 ± 0.10.

Conclusions:

• The proposed deep learning approach effectively captures nonlinear brain-muscle activity relationships.
• This method holds significant potential for enhancing the accuracy and reliability of non-invasive BCIs.