Single-Trial Classification of Multi-User P300-Based Brain-Computer Interface Using Riemannian Geometry.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Simultaneous electroencephalographic (EEG) data classification from multiple users presents a significant challenge in Brain-Computer Interface (BCI) research.
• Event-Related Potential (ERP) classification from single trials is crucial for real-time BCI applications.

Purpose of the Study:

• To compare different classification approaches for single-trial ERPs in a collaborative BCI game setting with two users.
• To introduce and evaluate novel Minimum Distance to Mean (MDM) classifier extensions (MDM-hyper and MDM-multi) within a Riemannian framework.

Main Methods:

• The study employed a Riemannian framework for EEG data analysis.
• Two novel MDM classifier variants were developed: MDM-hyper, incorporating inter-subject spatio-temporal statistics, and MDM-multi, merging multiple classifiers.
• Classification performance was evaluated on single-trial ERPs from two subjects engaged in a collaborative BCI game.

Main Results:

• Both MDM-hyper and MDM-multi classifiers demonstrated significant performance improvements over the average performance of the two users.
• These novel classifiers outperformed analogous classifiers based on step-wise linear discriminant analysis.
• The MDM-multi classifier achieved superior performance compared to the best individual player within the pair.

Conclusions:

• The developed MDM-multi and MDM-hyper classifiers offer enhanced accuracy for multi-user EEG classification in BCI.
• These advanced Riemannian-based classifiers show promise for improving collaborative BCI game performance and applications.
• The findings suggest that merging classifiers (MDM-multi) is a highly effective strategy for overcoming individual user limitations in BCI.