CTNet: a convolutional transformer network for EEG-based motor imagery classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interface (BCI) technology facilitates direct communication between the brain and machines.
• Electroencephalography (EEG)-based motor imagery (MI) is crucial for BCI systems but faces decoding limitations.
• Existing BCI methods struggle with the complexity and variability of EEG signals.

Purpose of the Study:

• To introduce a novel convolutional Transformer network (CTNet) for enhanced EEG-based MI classification.
• To improve the decoding accuracy and robustness of BCI systems.
• To establish a new benchmark for EEG signal decoding in BCI applications.

Main Methods:

• Developed CTNet, integrating convolutional modules for local feature extraction and Transformer encoders for global dependency analysis.
• Employed a multi-head attention mechanism within the Transformer encoder to capture high-level EEG features.
• Utilized fully connected layers for final EEG signal classification.

Main Results:

• CTNet achieved high subject-specific decoding accuracies: 82.52% (BCI IV-2a) and 88.49% (BCI IV-2b).
• CTNet demonstrated strong cross-subject performance: 58.64% (BCI IV-2a) and 76.27% (BCI IV-2b).
• CTNet outperformed state-of-the-art methods in both subject-specific and cross-subject evaluations.

Conclusions:

• CTNet significantly enhances EEG decoding accuracy for motor imagery classification.
• The proposed architecture shows great potential for advancing BCI applications in human-machine interaction and rehabilitation.
• CTNet sets a new standard for EEG decoding, addressing current limitations in BCI technology.