Two-stage wavelet shrinkage and EEG-EOG signal contamination model to realize quantitative validations for the artifact removal from multiresource biosignals. (January 2019)

• Record Type: Journal Article
• Authors: Singh, Balbir; Wagatsuma, Hiroaki
• Is Part Of: Biomedical signal processing and control. Volume 47(2019)
• Page Start: 96
• Page End: 114
• Publication Date: 2019-01
• Subjects: Undecimated wavelet decomposition -- Electroencephalograph (EEG) -- Electro-oculography (EOG) -- Wavelet shrinkage -- EEG-EOG contamination model; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.08.014
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel designed EEG-EOG signal contamination model to quantitatively validate artifact removal from EEGs. A two-stage wavelet shrinkage method proposed for the quality of the reconstructed EEG signal and evaluated in the frequency spectrum, which represents how much the original specific brain-state characteristic can be reconstructed. We checked the threshold multiplier at both stage-I, stage-II and L (decomposition level). Westated the importance of the threshold multiplier and L in the proposed scheme. We used the real EOG and EEG signal to demonstrate the effectiveness of proposed scheme. Abstract: Electroencephalogram (EEG) data inevitably contain large amounts of noise, particularly from ocular potentials in tasks with eye movements, known as electro-oculography (EOG) artifact, which has been a crucial issue in brain–computer-interface studies. This time-frequency characteristic has been substantially dealt with in previous proposed denoising algorithms that relied on a consistent assumption based on the single-noise component model. However, the traditional model is not applicable for biomedical signals that consist of multiple signal components, such as weak EEG signals that are easily recognized as noise because of the signal amplitude with respect to the EOG signal. In consideration of the realistic signal contamination, we designed a novel EEG-EOG signal contamination model to quantitatively validate artifact removal from EEGs. We then proposed the two-stage wavelet shrinkage method with the decomposition of the undecimated wavelet transform (UDWT), which is suitable for signal structure. Open-source clinical intracranial EEGs with the hundred dataset in each behavioral condition were introduced to the validation as "true EEG" before the contamination of artificial EOGs. The quality of the reconstructed EEG signal has evaluated in the frequency spectrum, which represents how much the original specific brain-state characteristic can be reconstructed. Numerical analyses demonstrated that the first stage pursued abrupt changes with high amplitudes provided by assumed EOGs, and the second stage provided the EEG frequency spectrum as observed in the original signal. What the performance exceeded the conventional shrinkage suggests that threshold values is required to be set properly depending on individual amplitudes of contaminated bio-signal sources as our proposed method demonstrated. This result focused on the actual amplitude-frequency structure in the polygenetic signal. It not only provided the decomposition performance, but also revealed how bio-signals are mixed together by using a new standard model for robust validation in the EEG-EOG signal contamination.