A novel channel selection method for optimal classification in different motor imagery BCI paradigms

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Classifying motor imagery (MI) is key for sensorimotor rhythm-based brain-computer interfaces (BCI).
• Determining the optimal number of scalp electrodes (channels) for MI classification is critical.
• Previous research focused on non-feedback MI tasks; this study examines optimal channel selection in real-time feedback paradigms (two-class and four-class control).

Purpose of the Study:

• To investigate optimal channel selection for motor imagery (MI) classification in brain-computer interface (BCI) systems with real-time feedback.
• To propose and evaluate a novel channel selection method, IterRelCen, for MI tasks.
• To compare the number of optimal channels required across different MI paradigms (MI task, two-class control, four-class control).

Main Methods:

• Analyzed three datasets from MI task, two-class control, and four-class control experiments.
• Extracted frequency-spatial synthesized features and applied the enhanced IterRelCen method for channel selection.
• Utilized a multiclass support vector machine classifier and One-way ANOVA for significance testing.

Main Results:

• The IterRelCen method achieved high classification accuracies: 85.2% (MI task), 94.1% (two-class control), and 83.2% (four-class control).
• The proposed method outperformed other channel selection techniques.
• Significantly different average numbers of optimal channels were found across the three MI paradigms.

Conclusions:

• IterRelCen demonstrates strong feature selection capabilities for BCI applications.
• The number of optimal channels required for classification accuracy increases with the complexity of the MI paradigm (MI task < two-class < four-class control).
• Findings offer insights for optimizing electroencephalography (EEG)-based BCI systems and enhancing noninvasive BCI performance.