A learnable EEG channel selection method for MI-BCI using efficient channel attention.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs) for motor imagery (MI) often utilize numerous electrodes, demanding significant computational resources.
• Effective channel selection is critical for optimizing performance and reducing computational load in MI-BCI systems.

Purpose of the Study:

• To propose and evaluate an automated channel selection method for EEG-based MI-BCIs.
• To enhance classification accuracy while reducing the number of channels used.

Main Methods:

• Integration of an efficient channel attention (ECA) module with a convolutional neural network (CNN).
• The ECA module automatically assigns channel weights based on their importance for BCI classification accuracy.
• Channel subsets are formed based on the established ranking of EEG channel importance.

Main Results:

• The proposed method achieved 75.76% accuracy with 22 channels and 69.52% accuracy with 8 channels in a four-class classification task.
• Outperformed existing state-of-the-art EEG channel selection methods on the BCI Competition IV dataset 2a.

Conclusions:

• The proposed ECA-CNN method offers an effective approach for channel selection in EEG-based MI-BCIs.
• This method successfully reduces the number of channels without compromising classification accuracy, making BCIs more computationally efficient.