EEGScaler: A Deep Learning Network to Scale EEG Electrode and Samples for Hand Motor Imagery Speed Decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI)-based Brain-Computer Interface (MI-BCI) systems aid stroke rehabilitation by decoding neural signals.
• Current MI-BCI systems have limited control due to decoding only basic motor actions.
• Decoding movement speed from unilateral MI tasks using electroencephalogram (EEG) is challenging due to low spatial resolution.

Purpose of the Study:

• To introduce EEGScaler, an end-to-end deep learning framework for decoding speed (slow vs. fast) from unilateral MI tasks.
• To enhance MI-BCI systems by increasing their degrees of freedom through speed decoding.
• To improve the precision and adaptability of BCI-driven neurorehabilitation.

Main Methods:

• EEGScaler adaptively scales EEG samples and electrodes using a Multi-Layer Perceptron (MLP) network.
• Spatiotemporal features are extracted using temporal and depth-wise convolution filters.
• The model utilizes subject-independent pre-training and subject-specific fine-tuning.

Main Results:

• EEGScaler achieved an average cross-validated accuracy of 65.98% in decoding fast vs. slow MI speed tasks.
• The proposed model outperformed existing methods by approximately 6%.
• Subject-specific scaling of EEG data for speed decoding from unilateral MI tasks is demonstrated as novel.

Conclusions:

• EEGScaler effectively decodes movement speed from unilateral MI tasks, enhancing MI-BCI control.
• This advancement offers a more natural and intuitive interface for neurorehabilitation applications.
• The findings support the potential for more precise and adaptive BCI-driven therapies tailored to individual patient needs.