A Multi-Branch Network for Integrating Spatial, Spectral, and Temporal Features in Motor Imagery EEG Classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Accurate decoding of motor imagery (MI) electroencephalogram (EEG) signals is crucial for effective brain-computer interface (BCI) systems.
• Extracting discriminative features from complex, nonlinear EEG signals across spatial, spectral, and temporal dimensions presents a significant challenge.
• Improving MI-EEG decoding performance hinges on addressing these multidimensional feature extraction complexities.

Purpose of the Study:

• To develop a deep neural network capable of jointly modeling spatial, spectral, and temporal features in MI-EEG signals.
• To enhance the classification performance of MI-EEG decoding by capturing complex, multidimensional signal characteristics.
• To improve the practical utility and precision of brain-computer interface (BCI) systems through advanced signal processing.

Main Methods:

• A multi-branch deep neural network was proposed, integrating four complementary feature extraction branches.
• The network processes both 3D power spectral density tensors and 2D time-domain EEG signals for unified multidimensional modeling.
• Gradient-weighted class activation mapping (Grad-CAM) was utilized for visualizing model-prioritized spatial and spectral features, aiding interpretability.

Main Results:

• The proposed model achieved 86.34% accuracy and a 0.829 kappa coefficient on the EEGMMIDB dataset (five-class task).
• On the BCI Competition IV Dataset 2a (BCIIV2A), the model attained 83.43% accuracy and a 0.779 kappa coefficient (four-class task).
• Performance demonstrated superiority over existing state-of-the-art methods in MI-EEG classification.

Conclusions:

• The developed multi-branch deep neural network effectively decodes motor imagery (MI) EEG signals, outperforming current methods.
• The model's ability to capture multidimensional features enhances brain-computer interface (BCI) performance.
• Grad-CAM visualizations confirmed the model's neurophysiological interpretability by highlighting key spatial channels and frequency bands.