Formulation of Common Spatial Patterns for Multi-Task Hyperscanning BCI

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Common Spatial Patterns (CSP) is a key feature extraction technique in brain-computer interfacing (BCI).
• Existing CSP methods face challenges in multi-subject scenarios with complex motor tasks and interference.
• Neurological studies often require robust methods for analyzing simultaneous brain activity from multiple individuals (hyperscanning).

Purpose of the Study:

• To develop an advanced CSP formulation, termed hyperCSP, for enhanced feature extraction in multi-subject BCI.
• To effectively isolate common motor tasks between simultaneously recorded subjects.
• To mitigate the impact of spurious or undesired tasks on BCI performance.

Main Methods:

• A novel hyperCSP formulation is proposed, integrating individual covariance and mutual correlation matrices from multi-subject electroencephalograms (EEG).
• The hyperCSP method was applied to analyze motor-related hyperscanning data.
• Classification was performed using hyperCSP features combined with a support vector machine (SVM) classifier.

Main Results:

• The hyperCSP method demonstrated effective isolation of common motor tasks among multiple subjects.
• Achieved a classification accuracy of 81.82% over 8 trials, even with significant undesired task interference.
• The technique offers satisfactory classification performance with reduced data size and computational complexity.

Conclusions:

• HyperCSP presents a promising advancement for feature extraction in multi-task BCI applications.
• This method has the potential to significantly reduce training errors in complex BCI scenarios.
• The publicly available motor-related hyperscanning dataset will facilitate further research in the field.