Data Space Adaptation for Multiclass Motor Imagery-based BCI.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Non-stationarity in electroencephalogram (EEG) signals poses a significant challenge for brain-computer interfaces (BCIs).
• Existing adaptation techniques primarily cater to binary-class BCIs, limiting their applicability to more complex, multiclass scenarios.

Purpose of the Study:

• To propose a novel supervised multiclass data space adaptation (MDSA) technique for EEG-based BCIs.
• To enhance the robustness and accuracy of multiclass BCIs by minimizing distribution differences between training and testing data.

Main Methods:

• Developed a linear transformation method to adapt multiclass EEG test data to the training data distribution.
• Evaluated the MDSA technique on the BCI Competition IV dataset 2a.

Main Results:

• The proposed MDSA technique improved classification accuracy by an average of 4.3% using 20 trials per class for adaptation.
• MDSA outperformed the multi pooled mean linear discrimination (MPMLDA) technique, even with as few as 10 trials per class for transformation matrix calculation.

Conclusions:

• The MDSA algorithm effectively addresses the non-stationarity issue in multiclass EEG-based BCIs.
• MDSA offers a promising solution for improving the performance and reliability of advanced brain-computer interfaces.