Local temporal correlation common spatial patterns for single trial EEG classification during motor imagery.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery-based brain-computer interfaces (BCI) rely on feature extraction from electroencephalography (EEG) signals.
• Common Spatial Pattern (CSP) is a widely used method, but its performance degrades due to noise sensitivity in covariance matrix estimation.

Purpose of the Study:

• To introduce a novel feature extraction method, Local Temporal Correlation-based CSP (LTCCSP), to enhance BCI performance.
• To address the noise sensitivity issue in CSP by incorporating local temporal correlation (LTC) for improved covariance matrix estimation.

Main Methods:

• LTCCSP utilizes correlation as a metric for spatial pattern similarity, outperforming Euclidean distance in previous variants like Local Temporal CSP (LTCSP).
• Quantitative comparisons were performed using simulated datasets with added outliers (BCI Competition III Dataset IVa and BCI Competition IV Dataset IIa).
• The methods were also evaluated on an in-house recorded EEG dataset.

Main Results:

• LTCCSP consistently achieved the highest average classification accuracies across varying outlier frequencies in simulated datasets.
• On the in-house EEG dataset, LTCCSP also demonstrated superior performance compared to CSP and LTCSP.
• The results indicate LTCCSP's robustness against noise and outliers.

Conclusions:

• LTCCSP offers a significant improvement over traditional CSP and its variants for motor imagery BCI.
• The incorporation of local temporal correlation provides a more robust estimation of covariance matrices.
• LTCCSP is a promising method for enhancing the reliability and accuracy of practical motor imagery BCI systems.