Artifact Removal of Nonlinear Transcranial Electrical Stimulations Using Adaptive Variational Mode Decomposition
Chen Ni,Fan Zeping,Cao Xinran,Qin Yurong
DOI: https://doi.org/10.2139/ssrn.3986960
2021-01-01
SSRN Electronic Journal
Abstract:Background: Transcranial alternating current stimulation is a widely used noninvasive brain stimulation method, and its stimulation effect is usually evaluated by electroencephalography (EEG). However, due to massive electrical stimulation artifacts overlapping the original EEG, direct insights into the effects during stimulation from EEG is methodologically challenging. The existing artifact removal methods focus on removing the first order stimulation artifact, and they are difficult to remove non-linear artifacts effectively.New method: According to the spectral characteristics of nonlinear artifact, an adaptive variational mode decomposition (AVMD) method is proposed to remove the nonlinear artifacts of transcranial electrical stimulation. In this method, the envelope of artifacts is extracted by Hilbert transform (HT), and then, the envelope signal is processed by window Fourier transform (WFT). The number of dominant oscillation frequencies is obtained by WFT spectrum analysis, and then, the K value that represents the VMD modes is determined. VMD is used to decompose the recorded data to obtain multiple intrinsic mode signals. According to the amplitude characteristics, the artifact components are selected, and the effective EEG components are recovered.Result: AVMD algorithm were tested on the synthetic and experimental data. The correlation coefficient between recovered EEG and real EEG was used to measure the artifact removal effect for the synthetic data. The mean absolute error (MAE) of the statistical characteristics between recovered EEG and sham EEG was used to evaluate the artifact removal effect for the experimental data. Under the conditions of amplitude modulation depth ma ∈ [0.001, 0.01], phase modulation depth mp ∈ [0.001, 0.01] and stimulation frequency farti ∈ [10, 100], the average correlation coefficients for harmonic-modulated artifact are 0.9885, 0.8935, 0.9484, respectively. The average correlation coefficients for noise-modulated artifact are 0.7984, 0.2872 and 0.9561, respectively. The MAE of the statistical characteristics between recovered EEG and sham EEG are 0.9896(kurtosis), 2.9918(root mean square amplitude), 0.1751(sample entropy) for the experimental data with the stimulation frequency 11Hz, and are 0.9407 (kurtosis), 2.4731 (root mean square amplitude) and 0.0841 (sample entropy) for the experimental data with the stimulation frequency 62Hz.Comparison with existing methods: When the ma, mp and SNR values of the original recorded data are small, AVMD algorithm has better artifact removal performance than SMA, AF and EMD methods.Conclusions: The performance of artifact removal algorithm is related to artifact characteristics (ma, mp, farti, SNR). AVMD algorithm has the best removal effect for harmonic-modulated artifact.