Extraction of evoked potentials based on subspace method and wavelet denoising **★

Abstract

AIM：In order to reduce the number of trials required for the extraction of the brain Evoked Potentials (EPs) and remove the spontaneous electroencephalogram (EEG) noise, this paper proposed and verified a new approach to extract visual evoked potential.

METHODS： Singular value decomposition (SVD) based on subspace is effective on denoising. Its mechanism explains, SVD on data matrix of noisy signal can obtain signal subspace and noise subspace, and orthogonal projection of noisy signal to signal subspace leads to denoising. But the signal-to-noise ratio (SNR) of EPs records on scalp is usually low, and it is not useful to denoise and extract EPs only by the subspace method. The spontaneous EEG is the main noise in the measurements of EPs. But the Lipschitz exponential for EEG singularity is ambitious, either be positive or be negative. It is difficult to extract EPs components from the measured record only by wavelet denoising. This paper proposed an approach to extract EPs by the combination of subspace method and wavelet transform. First, the raw data are divided into signal subspace and noise subspace by applying SVD. Then the raw data are orthonormally projected onto the estimated signal subspace to obtain an enhanced version of the raw data. Next, we utilize the wavelet method to reduce the noise of the previous denoised version. Finally, we extract EPs efficiently by combining wavelet transform and subspace method.

RESULT: The simulating results shows that the proposed methods can efficiently extract EPs when the SNR is lower than －10 dB by adding spontaneous EEG noise and white noise into the EPs model. Experiments show that the combined method is more effective than solo application, and can decurtate the trial times of extracting EPs within 4 or 5 trials, which is previously 20.

CONCLUSION:The combination of subspace method based on SVD with wavelet denoising provides much better capability on extracting EPs, and reduces the times of extracting.

Xiong XB, Chen YG.Extraction of evoked potentials based on subspace method and wavelet denoising.Zhongguo Zuzhi Gongcheng Yanjiu yu Linchuang Kangfu 2007;11(13):2430-2433(China) [www.zglckf.com/zglckf/ejournal/upfiles/07-13/13k-2430(ps).pdf]

摘要
目的：为减少提取诱发电位所需的试验次数，有效去除自发脑电噪声，提出一种新的视觉诱发电位提取方法并进行验证。
方法：基于奇异值分解的子空间方法可以用于去除信号中的噪声。①其基本原理是，由含噪信号形成的数据矩阵进行奇异值分解可以获得信号子空间和噪声子空间，将含噪信号正交投影到信号子空间中即可得到去除噪声。因为在头皮测量得到的诱发电位记录信号的信噪比很低，所以仅使用基于奇异值分解的子空间方法来去除噪声并不能有效地提取诱发电位。②实验记录中对诱发电位成分影响较大的自发脑电是有色噪声，描述其奇异性的Lipschitz指数具有不确定性，可能为正，也可能为负，因此仅用小波去噪方法提取诱发电位也不能取得理想的结果。③为此，提出了一种基于奇异值分解的子空间正交投影和小波去噪复合方法来提取诱发电位。首先应用基于奇异值分解的子空间方法将包含噪声的记录信号分解为信号子空间和噪声子空间，将含噪信号投影到信号子空间可得到初步去噪的信号，再应用小波变换进一步去除噪声，即可提取诱发电位。
结果：采用自发脑电模型产生有色的自发脑电噪声，与白噪声一起加入仿真的诱发脑电信号中，在低信噪比小于－10 dB的情况下，可有效地提取出诱发脑电信号。仿真和实验结果表明这种复合方法的效果好于单独采用其中的一种方法，能将提取诱发电位的实验次数由20次左右缩短为四五次。
结论：将基于奇异值分解的子空间方法和小波去噪结合起来，能有效提取诱发电位，减少提取诱发电位所需的试验次数。
关键词：视觉诱发电位；奇异值分解；小波去噪；康复工程

熊新兵，陈亚光.诱发电位提取的子空间和小波去噪复合方法[J].中国组织工程研究与临床康复，2007，11(13):2430-2433
[www.zglckf.com/zglckf/ejournal/upfiles/07-13/13k-2430(ps).pdf]