Grouped Automatic Relevance Determination and Its Application in Channel Selection for P300 BCIs

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) benefit from utilizing multiple electrode signals.
• Practical BCI applications favor a reduced channel subset for machine learning feasibility and usability.

Purpose of the Study:

• To propose an embedded channel selection approach for BCIs using grouped automatic relevance determination.
• To enable simultaneous channel selection and feature classification within a Bayesian linear model.

Main Methods:

• A Gaussian conjugate group-sparse prior was developed for embedded channel selection.
• Bayesian linear model with marginal likelihood maximization for hyper-parameter estimation.
• Applied to P300 speller BCIs using public and in-house datasets.

Main Results:

• The proposed method achieves competitive classification performance compared to state-of-the-art techniques.
• Selected channels demonstrate biological relevance to P300 signals.
• Simultaneous channel selection and classification were successfully achieved.

Conclusions:

• The embedded channel selection approach offers an efficient method for optimizing BCI signal processing.
• This technique enhances both the performance and practicality of brain-computer interfaces.
• The method provides biologically relevant channel subsets for P300-based BCIs.