Improved Adaptive Filtering based Artifact Removal from EEG Signals

Removal of the artifacts caused by eye movement is the necessary step in EEG (electroencephalogram) preprocessing. In this paper, we study the adaptive filtering algorithm for artifacts removal of eye movement and present a new LMS (Least Mean Square) based algorithm, by using eye movement artifact signal as a reference signal and take the error signal of the LMS system as the estimated EEG signal to achieve a significantly higher signal to noise ratio (SNR). In the experiments with real EEG data, we measure the mutual information (MI) and coherence (COH) that show the output of the new algorithm have better consistency with the original EEG signal. We also calculate the approximate entropy that indicates the output of the new algorithm better maintains nonlinear characteristics of the EEG signal.