Information Theoretic Approaches for Motor-Imagery BCI Systems: Review and Experimental Comparison

Area of Science:

• Neuroscience
• Signal Processing
• Information Theory

Background:

• Brain-Computer Interfaces (BCIs) are gaining significant attention for human-computer interaction.
• Common Spatial Patterns (CSP) is a standard technique for spatial filtering electroencephalogram (EEG) data in BCI applications.
• CSP's heuristic origins can be strengthened by information-theoretic foundations.

Purpose of the Study:

• To review the theoretical connections between CSP and information-theoretic measures.
• To explore alternative feature selection methods that maximize class label information.
• To experimentally compare the performance of various CSP and information-theoretic approaches.

Main Methods:

• Review of information-theoretic divergences: Kullback-Leibler, Beta, and Alpha-Beta log-det (AB-LD).
• Investigation of feature selection techniques prioritizing informativeness about class labels.
• Experimental validation and performance comparison of the discussed methods.

Main Results:

• Demonstration of strong theoretical links between CSP and information-theoretic concepts.
• Identification of divergences and feature selection strategies that enhance BCI performance.
• Empirical evidence supporting the efficacy of information-theoretic CSP variants.

Conclusions:

• Information theory provides a robust framework for understanding and improving CSP in BCI.
• Alternative divergence measures and feature selection methods offer promising avenues for enhanced EEG analysis.
• Further research into information-theoretic CSP is warranted for advancing BCI technology.