Research On Artifact Removal Methods Of EEG Signal In EEG-fMRI Hybrid Brain-computer Interface

With the development of brain science research,it is increasingly difficult for single-mode EEG signal observation to meet the current research needs.The combination of neuroimaging and electrophysiological recording complements each other and provides a multi-dimentional observation approach,of which the combination of EEG and functional magnetic resonance imaging(fMRI)has attracted extensive attention and research,and played an irreplaceable role in many areas.However,in the process of EEG-fMRI simultaneously acquisition,because of the changing nuclear magnetic environment,the EEG signal has been contaminated with a lot of noise.How to effectively remove the noise to get EEG signal with high quality is the problem of all synchronous reaserch.In this paper,we have analyzed the artifact components in the EEG signal at the EEGfMRI simultaneously acquisition and introduced the most popular method for nuclear magnetic artifacts removal,average artifact subtraction(AAS)and optimal basis sets(OBS)based PCA.The AAS is simple to calculate and it can removal most of the nuclear magnetic noise effectly,but still some part of the artifacts remains.The optimal basis sets method can extract the gradient artifact residuals using temporal PCA after AAS,but because the extraction results are unstable and it cannot adaptively select the artifact components so that it can usually lead to loss of information.This paper focuses on the study of ICA-baesd denoising method.ICA assumes that the source signals are independent of each other and it's independence constraints is much stronger than the irrelevant constraint of PCA.Therefore,the residual artifact components could be extracted better by performing ICA to separate the EEG signals from the noise signals.In this paper,we have also proposed an automatic method to select artifact components after ICA by analyzing the correlation between the frequency distribution of the nuclear magnetic noise and the scanning peorid of MRI.This method can select the gradient artifact components simply and reliably by caculating the ratio of the magnetic artifact energy and the low frequency EEG energy in each independent component,It overcomes the disadvantages of conventional ICA de-nosing method of choosing artifact components manully and improve the denosing performance.But there is still a problem remain after ICA-based denosing,conventionally setting the artifact components to zero could cause the loss of real EEG signal.So this paper has proposed a denoising method combining wavelet and ICA.Using wavelet multiresolution analysis to decompose the artifact components and reconstruct the EEG signal in it with low frequency approximate coefficient could effevtively remove the nuclear magnetic noise and avoid such information loss at the meantime.In the end of this paper,the superiority of the ICA combined with wavelet denoising method is verified by comparing the effect of two kinds of denoising methods by performing off-line experiments and cross-validation.And the results have been analyzed and disscuesd in the end.