Combining CEEMDAN with PCA for effective cardiac artifact suppression from single-channel EEG

Abstract

The large signal due to cardiac activity can easily distort the signals originating from the relatively weak electrical activity of the brain, commonly measured as an Electroencephalogram (EEG). The artifact due to cardiac activity in EEG is called cardiac artifact, which contaminates the EEG data and makes interpretation of the EEG difficult for clinicians. Hence it is crucial to remove the cardiac artifact from EEG data. To suppress the cardiac artifact, we propose a novel approach to effectively extract cardiac artifacts from single-channel contaminated EEG data without using reference Electrocardiogram (EKG) data. The proposed methodology uses Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to decompose EEG data contaminated by cardiac activity into the Intrinsic Mode Functions (IMFs). Principal Component Analysis (PCA) is performed on these IMFs to obtain the principal components arranged in the order of decreasing variance. Effective cardiac artifact extraction is achieved by optimizing the signal reconstruction process so that only those principal components that capture the cardiac activity are retained with the constraint that distortion introduced in EEG data should be minimum. The comparison clearly shows that the proposed method outperforms conventionally employed methods like wavelet-based approach.