Research On Key Algorithms Of Removing Ballistocardiogram Artifact From EEG

Electroencephalogram(EEG)and functional magnetic resonance imaging(f MRI)are the two major neuroimaging techniques.Simultaneous EEG-f MRI has been widely applied in clinical diagnosis with integrating the advantages of EEG and f MRI in spatiotemporal resolution.With high temporal and spatial resolution simultaneously,simultaneous EEG-f MRI still faces some challenges.The acquired EEG signals in simultaneous EEG-f MRI contain two main aitifacts: gradient artifact(GA)and ballistocardiogram(BCG)artifact.The gradient artifact(GA)is time-invariant,and can be corrected by various algorithms.BCG artifact removal is the important technology of EEG-f MRI fusion,which directly determines the quality of processed signals.It is difficult to effectively remove BCG artifact utilizing common filtering techniques due to its characteristics of low signal-to-noise ratio(SNR)and complexity of mixing signals.Therefore,BCG artifact removal is the greatest challenge in EEG artifact suppression method for simultaneous EEG-f MRI.To effectively remove BCG artifacts in acquired EEG signals by simultaneous EEGf MRI,the BCG artifact removal method utilizing adaptive comb filter(ACF)combined with the improved J-peak detection algorithm is proposed.Firstly,the reference BCG signal is obtained using principal component analysis(PCA).Then the adaptive threshold detection method is applied to detecte the R peaks of ECG signal.And the average delay between ECG signal and BCG artifacts of each channel is deter-mined by the correlation analysis on ECG signal and reference BCG signal.According to the R peak positions of ECG signal and the average delay,the ECG signal and reference BCG signal are segmented.The J peak positions of the BCG artifact are preliminary estimated by segmented ECG signal and reference BCG signal.And the accurate J peak positions are estimated by correcting preliminary J peak positions in accordance with the comparative analysis between RR interval and JJ interval.The parameters of ACF are adaptively adjusted according to the accurate J peak positions,and then the BCG artifact is estimated by that the raw EEG signal is filtered adaptively with the filter.The clear EEG is finally obtained by suppressing the estimated BCG artifact.Experiment on the BCG artifact suppression of clinic EEG signal is conducted.The visual effects,normalized power spectrum ratio(INPS),peak-to-peak value and combined ratio are selected as the criterial to evaluate the performance of BCG artifact removal methods.The experimental results show that the proposed BCG artifact removal method utilizing adaptive comb filter(ACF)combined with the improved J-peak detection algorithm is effective and better than other conventional BCG artifact removal method.In this thesis,the BCG artifact removal methods utilizing nonlinear estimation are studied.With the deep learning is introducted,the BCG artifacts removal methods utilizing deep feedforward neural network(DFN)and one-dimensional convolutional neural network(1D-CNN)respectively are proposed.The nonlinear transformation between ECG signal and BCG artifact is estimated using deep neural network,and then BCG artifact is obtained with the estimated nonlinear transformation.The clear EEG is finally extracted by removing the estimated BCG artifact from raw EEG.The experimental results show that the proposed BCG artifact removal methods utilizing DFN and 1-D CNN respectively are effective and better than other conventional BCG artifact removal method on both subjective and objective evaluation.