Comprehensive Study On Removal Of Artifacts From EEG Data

Electroencephalogram (EEG) is an important tool in various scientific research fields, such as clinical medicine, research on thought, brain-computer interface .But, the noises in feeble recording EEG, such as power line interference and ocular artifacts, pose a major embarrassment for EEG interpretation and disposal. So, it is important to eliminate artifacts for extracting pure EEG signal.Based on the existing achievements, a series of study on removal of artifacts from EEG data are made in this thesis. The main research work and achievements of this thesis include:(1) Traditional notch filters weakens parts of EEG signals since its spectrum is close to power line interference. In this thesis, three robust algorithms, which are notch filter based on pole-zero placement algorithm, adaptive notch filter and independent component analysis (ICA), are investigated and employed to filter the power line interference .Comprehensive comparisons between these algorithms are made and the experimental results demonstrate that the proposed algorithms can eliminate 50Hz power line interference successfully and do little harm to useful EEG signals. Remarkably, ICA algorithm is the best choice for extracting EEG signals and eliminating artifacts due to its perfect performance shown in simulation experiments.(2) A robust algorithm based on the combination of principal component analysis (PCA) and joint approximative diagonalization of eigen matrix (JADE ) is presented. Besides, the influence of PCA on the performance is discussed. Simulation results demonstrate that the proposed method is efficient for noise removal in EEG signals.(3) A novel method for automatic removal of electroocular (EOG) artifacts from EEG data, which using nonlinear parameters, is presented. Renyi's entropy and sample entropy are adapted to identify ocular artifact automatically in this paper, and the results are compared with a robust method based on fractal dimension. Blind source separation (BSS) algorithm is used to separate real EEG signal firstly, then a universal approach together with three kinds of parameters is employed to identify the artifact components, then the artifacts is eliminated through reconstruction of EEG data . The experimental results show that, for real EEG data of different lengths, and in the condition that ocular artifact is separated into different components, Renyi's entropy and sample entropy can always identify ocular artifact accurately, but fractal dimension may underestimate or overestimate ocular artifact with EEG date of short length or when there are several artifact components .Taking into account computation time, Renyi's entropy is of optimal performance. Meanwhile, compared with existing algorithms, the proposed robust and efficient method does not need any EOG reference channel, and is fully automated without human intervention, which is suitable for real-time applications.