Robust Common Spatial Patterns for EEG signal preprocessing.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Common Spatial Patterns (CSP) is crucial for discriminating electroencephalogram (EEG) signal classes, particularly for motor activities.
• Standard CSP algorithms are sensitive to outliers due to covariance matrix estimation, impacting performance.
• Multivariate outliers can significantly degrade the accuracy of CSP-based EEG analysis.

Purpose of the Study:

• To investigate the impact of multivariate outliers on CSP algorithm performance.
• To develop a robust CSP algorithm less susceptible to outliers.
• To improve the reliability of EEG signal classification in the presence of noisy data.

Main Methods:

• Implemented Minimum Covariance Determinant (MCD) estimator for robust covariance estimation.
• Utilized Median Absolute Deviation (MAD) for robust variance estimation of projected EEG signals.
• Modified the CSP algorithm by replacing classical covariance estimates with robust estimates.

Main Results:

• The proposed robust CSP algorithm demonstrated a significant reduction in accuracy drop compared to the standard CSP.
• With 2.5% outliers, standard CSP accuracy dropped by 9.21%, while the proposed algorithm only dropped by 0.72%.
• The modified algorithm effectively mitigates the influence of outliers on CSP performance.

Conclusions:

• The robust CSP algorithm using MCD and MAD estimators offers improved performance and reliability for EEG signal analysis.
• This modification makes CSP a more dependable tool for classifying motor imagery and other EEG-based applications.
• The proposed method provides a practical solution for handling outlier-corrupted EEG data.