Multilayer network-based channel selection for motor imagery brain-computer interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-based brain-computer interfaces (MI-BCI) rely on electrode channel selection for performance and usability.
• Existing channel selection methods often overlook inter-channel interactions and cross-frequency band information.
• This limitation can lead to suboptimal decoding accuracy in MI-BCI systems.

Purpose of the Study:

• To introduce a novel Multilayer Network-based Channel Selection (MNCS) method for MI-BCI systems.
• To address the limitations of univariate channel selection by incorporating network interactions across frequency bands.
• To enhance both the decoding performance and practical convenience of MI-BCI applications.

Main Methods:

• A multilayer network framework was constructed by integrating brain networks from four frequency bands.
• Graph learning estimated the multilayer network from multi-band filtered electroencephalogram (EEG) data.
• The multilayer participation coefficient identified channels with minimal redundancy; Common Spatial Pattern (CSP) and Support Vector Machine (SVM) were used for feature extraction and classification.

Main Results:

• The MNCS method demonstrated superior performance compared to using all channels across multiple datasets (e.g., 85.8% vs. 93.1%).
• Significantly higher decoding accuracies were achieved by MNCS compared to state-of-the-art methods in MI-BCI systems (p < 0.05).
• Validation was performed on publicly available BCI Competition datasets and a dataset of stroke patients.

Conclusions:

• The proposed MNCS method effectively selects optimal EEG channels for MI-BCI.
• This channel selection strategy significantly improves decoding accuracy and enhances the usability of MI-BCI systems.
• MNCS offers a promising approach for advancing the development of practical brain-computer interfaces.