An Electroencephalogram Artifacts Removal Algorithm for Electroencephalogram Signals Based on Sample Entropy-Complete Ensemble Empirical Mode Decomposition with Adaptive Noise

An improved algorithm for removing electroencephalogram (EEG) artifacts based on sample entropy is proposed to solve the problem that EOG signals are easily polluted by electroencephalography (EOG) artifact, and conventional artifact removal algorithms lead to a large amount loss of useful EEG information. At first, EEG is decomposed into independent components by an independent component analysis (ICA) algorithm, then, sample entropy analysis for each independent component is employed. Then a threshold value is introduced to automatically identify the artifact component. The identified artifact component is decomposed by the complete ensemble empirical mode decomposition algorithm with adaptive noise (CEEMDAN), and then is denoised by the wavelet transform. Finally, both the inverse CEEMDAN and the inverse ICA algorithms are used to reconstruct the signal to achieve the purpose of artifact removal. Experiments are performed using 60 sets of data in the public BCI2000 motion imagination data set. Results show that the accuracy of automatic identification of EOG artifacts is 80%, and is about 26.7% higher than kurtosis-based algorithms. Experiment results performed on 15 sets of data in the public Klados EEG data set show that the correlation coefficient between the reconstructed EEG signal and the pure EEG signal is 0.841, and the root mean square error is reduced by about 56.82% compared with contaminated EEG signals. It is proved that the proposed algorithm achieves high performance not only in artifact removal but also in useful EEG information retainment. © 2020, Editorial Office of Journal of Xi'an Jiaotong University. All right reserved.