EEG DENOISING USING WAVELET ENHANCED ICA

Abstract

In this work we have presented a new approach towards wavelet enhanced ICA, we have used mMSE and kurtosis to detect the artifactual components automatically, Mahajan et al [28] displayed their performance in terms of sensitivity (90%) and specificity (98%), nMSE is good at recognizing EEG patterns because of its randomness and kurtosis is good at recognizing peaked signal because they have high kurtosis values. We compared our result with ICA based method zeroing ICA in terms of correlation, mutual information and coherence. Our result is far superior to it in all three terms, in correlation measure our method not only gives better results for unaffected recording channel but it improves the result from 0.44 to 0.58 for most affected recording channel, which means our method only suppresses the noise without introducing additional noise. When we compare the results in terms of mutual information it improves from 0.30 to 0.42 for most affected recording channel. When we study the coherence graph we notice that the wICA method is affecting those frequencies too which are not present in ocular artifacts frequency range but our method has only affected the frequency range 0-16 Hz which is ocular artifact’s frequency band. We can extend this work by using different mother wavelets to best approximate eye blink and other ocular artifacts.