Whitening Technique Based on Gram-Schmidt Orthogonalization for Motor Imagery Classification of Brain-Computer...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) rely on accurate classification of brain signals, such as motor imagery (MI).
• Accuracy variance among subjects poses a significant challenge for robust BCI performance.
• Existing methods like electroencephalogram eigenface analysis (EFA) can be sensitive to this variability.

Purpose of the Study:

• To introduce a novel whitening technique for motor imagery classification in BCIs.
• To reduce the accuracy variance across different subjects.
• To enhance the performance of electroencephalogram eigenface analysis for MI classification.

Main Methods:

• Proposed a BCI channel whitening (BCICW) scheme utilizing Gram-Schmidt orthogonalization.
• Implemented the BCICW method to minimize inter-subject accuracy variance.
• Validated the technique on BCI competition datasets (IIIa and 4 IIa) using MATLAB simulations.

Main Results:

• The BCICW method demonstrated a substantial reduction in accuracy variance compared to the EFA method.
• For dataset IIIa, variance decreased from 58.33 (EFA) to 11.21 (BCICW).
• For dataset 4 IIa, variance decreased from 17.48 to 9.38 with the BCICW method.

Conclusions:

• The proposed BCICW method effectively reduces accuracy variance in motor imagery classification.
• This technique shows promise for improving the reliability and performance of BCI applications.
• Gram-Schmidt orthogonalization is a valuable tool for enhancing BCI signal processing.