Evidence of Variabilities in EEG Dynamics During Motor Imagery-Based Multiclass Brain-Computer Interface

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Inter-subject and inter-session variabilities significantly challenge electroencephalogram (EEG)-based brain-computer interface (BCI) system performance.
• High-dimensional EEG data increases computational burden due to the large number of channels.

Purpose of the Study:

• Investigate inter-session and inter-subject variabilities in EEG dynamics during motor imagery (MI) tasks.
• Explore the impact of day-to-day EEG variability on BCI performance across different cognitive states.
• Examine the feasibility of inter-subject BCIs and the effect of channel reduction on spatial brain dynamics.

Main Methods:

• Analyzed EEG data from four MI tasks (left/right hand, feet, tongue).
• Applied three different spatial filtering techniques for preprocessing.
• Evaluated BCI performance under inter-subject and inter-session conditions with varying channel counts.

Main Results:

• Achieved a maximum classification accuracy of approximately 58% for inter-subject BCI scenarios.
• Observed a 31% deviation in classification accuracy across sessions for inter-session BCI analysis.
• Demonstrated the influence of channel reduction on spatial brain dynamics.

Conclusions:

• BCIs can be made more generic and user-friendly ('plug-and-play') by minimizing subject- and session-specific calibration.
• Employing fewer channels is crucial for reducing computational load and enhancing efficiency.
• The study highlights a path towards more accessible and robust BCI systems.