Variance thresholded EMD-CCA technique for fast eye blink artifacts removal in EEG

Abstract

Eye blink (EB) artifacts generated during eye blinks often contaminate electroencephalogram (EEG) signal. Previously Empirical Mode Decomposition (EMD) and Canonical Correlation Analysis (CCA), hybrid EMD-CCA were developed for EB artifact removal in EEG. However, EMD restricts the hybrid algorithm for real time implementation due to its slow processing nature, hence the algorithm has to be enhanced so that it can be a viable solution for real-time EB artifact removal. In this research work, to avoid applying EMD repetitively as and when EB artifacts occur, a method to use EMD minimally is approached. A suitable EB artifact region is detected through a variance threshold algorithm. This region is then subjected to EMD, where an EB artifact template is extracted out. This template is used by CCA to remove all EB artifacts that are present in that particular EEG signal, avoiding the need to apply EMD repetitively. The proposed method, (varEMD-CCA) is analyzed in terms of Signal-to-Noise Ratio (SNR) and the time consumed in removing EB artifacts from the entire length of the frontal channel, Fp1 of the EEG signal. Analysis shows that the proposed method (varEMD-CCA) is at least 40 times faster than the previous EMD-CCA method. The SNR of both methods are also comparable, which means the proposed method could comparably removes EB artifacts as the previous method does.