基于希尔伯特—黄变换的高阶听觉诱发电位提取研究
发布时间：2019-04-16 12:51
【摘要】：听觉诱发电位(Auditory evoked potentials, AEP)是听觉系统受到特定的声音刺激后中枢神经系统产生的与外界刺激相关的生物电变化。目前AEP已广泛的应用于评价婴幼儿听力、鉴别诊断听神经病变等方面。临床上的AEP记录主要采用刺激间间隔(Stimulus onset asynchrony, SOA)相等的低刺激率方案并通过总体平均方法提高信号的信噪比。当刺激率过高,刺激间间隔小于诱发反应的潜伏期时,相邻刺激所诱发的AEP波形就会出现首尾重叠的现象,这种AEP称为高刺激率AEP (High stimulus rate AEP, HSR-AEP),其所包含的暂态AEP成分称为高阶AEP (High-order AEP, HO-AEP)。高刺激率条件下由于听神经负荷加重从而有助于听觉系统适应性的评估以及一些听觉系统疾病的机理研究和临床诊断,并且HO-AEP有利于麻醉深度监测和睡眠状态评估。考虑到在给予相同刺激个数的情况下,高刺激率记录比常规记录的时间要短许多,人们也期望高刺激率记录可以减少记录时间。因此HO-AEP的研究具有广阔的临床应用前景。 然而,HSR-AEP的瞬时分量产生重叠的问题是无法通过传统的平均方法解决的,其重叠过程在工程学上可视为HO-AEP与刺激序列的循环卷积效应所导致。基于这一模型,人们对刺激序列中的各个SOA采用抖动(Jitter)技术,即应用不规则的SOA代替相等的SOA,以便对HSR-AEP去卷积还原HO-AEP。一种抖动较小的刺激序歹(?)——CLAD (Continuous loop averaging deconvolution)去卷积方案,被广泛的应用于重建HO-AEP。但是该方案去卷积过程是在频域中进行的,会对某些频带的噪声进行放大,从而影响高刺激率记录的效率。如何在少次平均的情况下获得较高质量的去卷积前的信号是目前研究的热点,因此本文的主要研究工作包括： 1、运用希尔伯特-黄变换(Hilbert-Huang transform, HHT)与非线性阈值滤波相结合方法提高信号的信噪比,减少平均次数。HHT包括经验模态分解(Empirical mode decomposition, EMD)和Hilbert变换两部分,其中EMD能够根据脑电信号的局部特征尺度将其分解为一组频率由高到低的固有模态函数(Intrinsic mode functions, IMFs),通过对每个IMF进行Hilbert变换就能确定其频率范围。EMD相当于一个带通滤波过程,能够将不同频率的信号和噪声相对分离,但是它本身并不能区分混叠在同一频率段的有用信号和噪声,因此本文提出了对IMFs进行进一步的阈值滤波处理方法。因为本文所感兴趣的AEP的频率范围是20-100Hz,所以将包含主要AEP频率成分的IMFs分为有用信号层,高频的IMFs归为噪声层,低频的包含较少信息的IMFs归为趋势层。由于相应频带的IMFs的幅值分布近似于均值为零的高斯分布,因此本文根据其标准差和分类来决定相应层的阈值。对噪声层IMFs选取较小滤波阈值,尽可能的去除较多的噪声,加强高频信号的平稳性。信号层包含了主要的AEP成分,但是仍然包含了大量的瞬态干扰,阈值滤波主要是去除对平均处理影响较大的大幅值的干扰。对于信号层本文提出了两种滤波方法：整体滤波和区间滤波。整体滤波是当某个IMF中出现大于所设定阈值的波幅时将整个IMF全部去除。区间滤波是将IMF中出现的大于所设定阈值的波峰去除,保留IMF的其余部分。 基于上述思想该部分设计了三种滤波方案：(1)直接提取信号层的IMFs重建信号；(2)对噪声层采用类似软阈值的处理方法,对信号层分别采取整体滤波和区间滤波的方法处理。最后将重建后的EEG信号分别做总体平均处理,得到估计的HSR-AEP。通过三个受试者的临床数据对以上的滤波方法进行验证,并计算信噪比作为评价标准,结果表明上述的三种滤波方法都能够有效的提高信号质量,减少平均次数。 2、运用HHT与总体相关技术(Ensemble correlation, EC)相结合的方法提取暂态AEP。在EMD分解时,先将连续的EEG数据按一个周期刺激序列对应时间分段,形成一组等长的EEG数据段(称为EEG扫程)。由于每个EEG扫程是由相同刺激序列所诱发,故在背景噪声的掩盖下含有相同的AEP成分。经过EMD分解后,各层IMF的信噪比不同。信噪比较高的IMF之间存在一定的相关性。因此不同EEG扫程经EMD分解后IMF之间的EC函数可以作为一维滤波函数,对IMF进行加权滤波。为了强化这种相关性,我们在EMD分解前先对EEG扫程进行适当的分组平均,以提高EEG扫程的信噪比。与阈值法相比,采用EC滤波法处理的IMF信号,无需人为对IMF进行分类和确定阈值,减少了主观因素的影响。通过对相同的数据集进行检验,结果表明该方法能有效的抑制噪声提高诱发信号的信噪比,并且不需要信号的先验知识和人为干预,具有广泛的应用范围。
[Abstract]:Auditory evoked potentials (AEP) are the bioelectrical changes in the central nervous system that are associated with external stimuli after the hearing system is stimulated by a particular sound. At present, AEP has been widely used in the evaluation of the hearing of infants and the differential diagnosis of auditory neuropathy. The clinical AEP records mainly employ a low stimulation rate scheme equal to the inter-stimulation interval (SOA) and improve the signal-to-noise ratio of the signal by the overall averaging method. When the stimulation rate is too high and the inter-stimulation interval is less than the latent period of the evoked response, the AEP waveform induced by the adjacent stimulation will have an end-to-end overlap, which is known as the high stimulation rate AEP (HSR-AEP), and the transient AEP component contained therein is referred to as a high-order AEP (High-order AEP, HO-AEP). It is helpful to evaluate the adaptability of auditory system and the mechanism and clinical diagnosis of some auditory system diseases under the condition of high stimulation rate, and HO-AEP is beneficial to the monitoring of anesthesia depth and the evaluation of sleep state. In view of the fact that the high stimulation rate recording is much shorter than the conventional recording time in the case of giving the same number of stimuli, it is also expected that the high stimulation rate recording can reduce the recording time. Therefore, the research of HO-AEP has a broad prospect of clinical application. However, the problem of the superposition of the instantaneous components of the HSR-AEP is not solved by the conventional averaging method, and the overlapping process can be considered in engineering as the cyclic convolution effect of the HO-AEP and the stimulation sequence. To this model, a Jitter technique is applied to each SOA in the stimulation sequence, i.e., an irregular SOA is applied in place of the same SOA, so that the HSR-AEP is deconvolved to restore the HO-AE P. A less jittery stimulus. (?) _ CLAD (Continuous loop) deconvolution scheme is widely used in the reconstruction of HO-AE P. However, the deconvolution process of the scheme is carried out in the frequency domain, and the noise of certain frequency bands can be amplified, thereby affecting the effect of high stimulation rate recording. The main research work package of this paper is how to obtain high-quality deconvolution before deconvolution is the hot spot of current research. The method of combining Hilbert-Huang transform (HHT) and non-linear threshold filtering to improve the signal signal-to-noise ratio The HHT consists of the empirical mode decomposition (EMD) and the Hilbert transform, in which the EMD can be decomposed into a set of intrinsic mode functions (IM) from high to low according to the local characteristic scale of the brain electrical signal. Fs), which can be determined by Hilbert transform for each IMF The EMD is equivalent to a band-pass filtering process, which can separate the signal and noise of different frequencies, but it does not distinguish the useful signal and noise of the aliasing in the same frequency band, so this paper puts forward a further threshold filter for IMFs. Since the frequency range of the AEP of interest is 20-100 Hz, the IMFs containing the main AEP frequency components are divided into a useful signal layer, the high-frequency IMFs are classified as a noise layer, and the low-frequency IMFs containing less information are classified into a useful signal layer. The trend layer. Since the amplitude distribution of the IMFs of the corresponding frequency band is similar to the Gaussian distribution with the mean value of zero, the corresponding layer is determined according to the standard deviation and the classification. And the lower filtering threshold is selected for the noise layer IMFs, more noise is removed as much as possible, and the high-frequency signal is enhanced. stationarity. The signal layer contains the primary AEP component, but still contains a large amount of transient interference, the threshold filtering is mainly to remove a significant value that has a greater impact on the average processing In this paper, two kinds of filtering methods are proposed in this paper: the whole filtering and the area Inter-filtering. The overall filtering is the whole of the IMF when the amplitude of a certain IMF is greater than the set threshold the interval filtering is to remove the peak of the IMF that is greater than the set threshold, and reserve the IMF the method comprises the following steps of: (1) directly extracting an IMFs reconstruction signal of a signal layer; (2) adopting a processing method similar to a soft threshold to the noise layer, and finally, the reconstructed EEG signals are respectively subjected to overall average processing to obtain an estimated HS, R-AEP. The above filtering method is validated by the clinical data of three subjects, and the signal-to-noise ratio is calculated as the evaluation standard. The results show that the above three filtering methods can effectively improve the signal quality and reduce the signal-to-noise ratio. the method of combining the HHT with the general correlation technique (EC) The transient AEP is extracted. At the time of EMD decomposition, the continuous EEG data is first segmented according to a periodic stimulation sequence to form a set of equal-length EEG data segments (said for EEG scanning). Since each EEG sweep is induced by the same stimulation sequence, the phase is contained under the masking of the background noise The same AEP composition. After the decomposition of EMD, each layer of IM The signal-to-noise ratio of F is different. Therefore, the EC function between the IMF and the IMF can be used as a one-dimensional filtering function, which can be used as a one-dimensional filtering function. F performs weighted filtering. In order to enhance this correlation, we have an appropriate packet averaging of the EEG sweep before the EMD is resolved to improve the EE The signal-to-noise ratio of the G sweep is compared with the threshold method. The IMF signal processed by the EC filtering method is not required to classify and determine the threshold value for the IMF, so that the signal-to-noise ratio is reduced. The results show that the method can effectively suppress the noise, improve the signal-to-noise ratio of the induced signal, and does not need the prior knowledge of the signal and the human intervention, and the method has the advantages that the method can effectively suppress the noise and improve the signal-to-noise ratio of the induced signal,
【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：R764
本文编号：2458791
[Abstract]:Auditory evoked potentials (AEP) are the bioelectrical changes in the central nervous system that are associated with external stimuli after the hearing system is stimulated by a particular sound. At present, AEP has been widely used in the evaluation of the hearing of infants and the differential diagnosis of auditory neuropathy. The clinical AEP records mainly employ a low stimulation rate scheme equal to the inter-stimulation interval (SOA) and improve the signal-to-noise ratio of the signal by the overall averaging method. When the stimulation rate is too high and the inter-stimulation interval is less than the latent period of the evoked response, the AEP waveform induced by the adjacent stimulation will have an end-to-end overlap, which is known as the high stimulation rate AEP (HSR-AEP), and the transient AEP component contained therein is referred to as a high-order AEP (High-order AEP, HO-AEP). It is helpful to evaluate the adaptability of auditory system and the mechanism and clinical diagnosis of some auditory system diseases under the condition of high stimulation rate, and HO-AEP is beneficial to the monitoring of anesthesia depth and the evaluation of sleep state. In view of the fact that the high stimulation rate recording is much shorter than the conventional recording time in the case of giving the same number of stimuli, it is also expected that the high stimulation rate recording can reduce the recording time. Therefore, the research of HO-AEP has a broad prospect of clinical application. However, the problem of the superposition of the instantaneous components of the HSR-AEP is not solved by the conventional averaging method, and the overlapping process can be considered in engineering as the cyclic convolution effect of the HO-AEP and the stimulation sequence. To this model, a Jitter technique is applied to each SOA in the stimulation sequence, i.e., an irregular SOA is applied in place of the same SOA, so that the HSR-AEP is deconvolved to restore the HO-AE P. A less jittery stimulus. (?) _ CLAD (Continuous loop) deconvolution scheme is widely used in the reconstruction of HO-AE P. However, the deconvolution process of the scheme is carried out in the frequency domain, and the noise of certain frequency bands can be amplified, thereby affecting the effect of high stimulation rate recording. The main research work package of this paper is how to obtain high-quality deconvolution before deconvolution is the hot spot of current research. The method of combining Hilbert-Huang transform (HHT) and non-linear threshold filtering to improve the signal signal-to-noise ratio The HHT consists of the empirical mode decomposition (EMD) and the Hilbert transform, in which the EMD can be decomposed into a set of intrinsic mode functions (IM) from high to low according to the local characteristic scale of the brain electrical signal. Fs), which can be determined by Hilbert transform for each IMF The EMD is equivalent to a band-pass filtering process, which can separate the signal and noise of different frequencies, but it does not distinguish the useful signal and noise of the aliasing in the same frequency band, so this paper puts forward a further threshold filter for IMFs. Since the frequency range of the AEP of interest is 20-100 Hz, the IMFs containing the main AEP frequency components are divided into a useful signal layer, the high-frequency IMFs are classified as a noise layer, and the low-frequency IMFs containing less information are classified into a useful signal layer. The trend layer. Since the amplitude distribution of the IMFs of the corresponding frequency band is similar to the Gaussian distribution with the mean value of zero, the corresponding layer is determined according to the standard deviation and the classification. And the lower filtering threshold is selected for the noise layer IMFs, more noise is removed as much as possible, and the high-frequency signal is enhanced. stationarity. The signal layer contains the primary AEP component, but still contains a large amount of transient interference, the threshold filtering is mainly to remove a significant value that has a greater impact on the average processing In this paper, two kinds of filtering methods are proposed in this paper: the whole filtering and the area Inter-filtering. The overall filtering is the whole of the IMF when the amplitude of a certain IMF is greater than the set threshold the interval filtering is to remove the peak of the IMF that is greater than the set threshold, and reserve the IMF the method comprises the following steps of: (1) directly extracting an IMFs reconstruction signal of a signal layer; (2) adopting a processing method similar to a soft threshold to the noise layer, and finally, the reconstructed EEG signals are respectively subjected to overall average processing to obtain an estimated HS, R-AEP. The above filtering method is validated by the clinical data of three subjects, and the signal-to-noise ratio is calculated as the evaluation standard. The results show that the above three filtering methods can effectively improve the signal quality and reduce the signal-to-noise ratio. the method of combining the HHT with the general correlation technique (EC) The transient AEP is extracted. At the time of EMD decomposition, the continuous EEG data is first segmented according to a periodic stimulation sequence to form a set of equal-length EEG data segments (said for EEG scanning). Since each EEG sweep is induced by the same stimulation sequence, the phase is contained under the masking of the background noise The same AEP composition. After the decomposition of EMD, each layer of IM The signal-to-noise ratio of F is different. Therefore, the EC function between the IMF and the IMF can be used as a one-dimensional filtering function, which can be used as a one-dimensional filtering function. F performs weighted filtering. In order to enhance this correlation, we have an appropriate packet averaging of the EEG sweep before the EMD is resolved to improve the EE The signal-to-noise ratio of the G sweep is compared with the threshold method. The IMF signal processed by the EC filtering method is not required to classify and determine the threshold value for the IMF, so that the signal-to-noise ratio is reduced. The results show that the method can effectively suppress the noise, improve the signal-to-noise ratio of the induced signal, and does not need the prior knowledge of the signal and the human intervention, and the method has the advantages that the method can effectively suppress the noise and improve the signal-to-noise ratio of the induced signal,
【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：R764
【参考文献】
相关期刊论文 前2条
1 钟佑明,秦树人,汤宝平;希尔伯特黄变换中边际谱的研究[J];系统工程与电子技术;2004年09期
2 赵进平,黄大吉;MIRROR EXTENDING AND CIRCULAR SPLINE FUNCTION FOR EMPIRICAL MODE DECOMPOSITION METHOD[J];Journal of Zhejiang University Science;2001年03期
，本文编号：2458791
本文链接：https://www.wllwen.com/yixuelunwen/yank/2458791.html
最近更新
教材专著
