Classification Algorithm for Electroencephalogram-based Motor Imagery Using Hybrid Neural Network with...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery (MI) is a key brain-computer interface (BCI) technology.
• Existing BCI decoding methods are mature for two-class motor tasks.
• Decoding four-class motor imagery tasks requires further research.

Purpose of the Study:

• To develop an effective hybrid neural network for four-class motor imagery classification.
• To improve the accuracy and performance of EEG-based BCIs.

Main Methods:

• Designed a hybrid neural network combining spatiotemporal convolution and attention mechanisms.
• Utilized spatiotemporal convolution for feature extraction.
• Employed a Multi-branch Convolution block and Transformer encoder with self-attention for classification.

Main Results:

• Achieved 83.3% average classification accuracy and a 0.78 kappa value on the BCI Competition IV 2a dataset.
• Demonstrated superior performance compared to most existing methods.
• Validated on well-known MI datasets (BCI Competition IV 2a and 2b).

Conclusions:

• The proposed hybrid neural network effectively decodes four-class motor imagery tasks.
• This approach offers a significant advancement for EEG-based BCIs.
• The method shows strong potential for enhancing BCI applications.