Electroencephalogram-Based Motor Imagery Signals Classification Using a Multi-Branch Convolutional Neural Network...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) signals are crucial for brain-computer interface (BCI) applications, particularly in classifying motor imagery (MI).
• Classifying EEG-MI signals presents challenges due to signal weakness, artifacts, low signal-to-noise ratio, and variability influenced by patient state.
• Accurate MI classification is vital for stroke rehabilitation and assistive technologies.

Purpose of the Study:

• To propose a novel, lightweight deep learning model for enhanced classification of EEG-MI signals.
• To incorporate an attention mechanism and fusion techniques to improve classification accuracy and robustness.
• To evaluate the proposed model's performance against state-of-the-art methods.

Main Methods:

• Development of a Multi-Branch EEGNet incorporating a Convolutional Block Attention Module (MBEEGCBAM).
• Application of channel-wise and spatial-wise attention mechanisms within the neural network architecture.
• Utilizing a fusion approach (FMBEEGCBAM) to further enhance classification performance.

Main Results:

• The MBEEGCBAM model achieved 82.85% accuracy on the BCI-IV2a dataset and 95.45% on the high gamma dataset.
• The fusion approach (FMBEEGCBAM) improved accuracy to 83.68% and 95.74% on the respective datasets.
• The proposed model demonstrates superior performance with fewer parameters compared to existing models.

Conclusions:

• The MBEEGCBAM model offers a promising lightweight solution for accurate EEG-MI signal classification.
• Attention mechanisms and fusion techniques significantly enhance the performance of BCI systems.
• This approach holds potential for advancing stroke rehabilitation and other BCI-driven applications.