去卷积序列的性质及其对听觉诱发反应的影响

发布时间：2018-05-23 15:31

本文选题：听觉诱发电位 + 线性卷积模型　；参考：《南方医科大学》2017年博士论文

【摘要】：听觉系统在短促刺激声作用下,经头皮电极记录可得到有限持续时间的电信号,即为一种暂态型的听觉诱发电位(Auditory evoked potential,AEP)。AEP十分微弱需要多次重复刺激,平均去噪后方能识别。但如果刺激重复率过快,AEP未及平息又被激发则导致AEP成分的相互重叠失去暂态性。工程上将这种重叠反应描述为暂态反应和刺激序列的线性卷积模型,如果考虑到相邻刺激的相互作用的影响,还需要采用诱发反应的非线性成分及非线性卷积模型。AEP按潜伏期可分为听觉脑干反应(Auditory brainstem-response,ABR)和中潜伏期反应(Middle-latency response,MLR)等。目前已提出若干去卷积算法恢复暂态的AEP成分,这些方法各具特色也各有不足。本文系统介绍了本领域的线性/非线性去卷积技术和实验范式,对其中部分关键问题提出解决办法,具体工作包括以下三个方面:1.真实噪声条件下CLAD刺激序列的评估和优化CLAD(Continuous loop averaging deconvolution)是近期关注较多的线性去卷积方法,采用刺激序列随机抖动方式保证解的存在。然而,该法应用时存在两个关键问题:刺激序列的噪声抑制性能评估和刺激序列优化产生。本项研究指出其序列的噪声抑制能力需要结合噪声本身的统计特性。并特别证实在ABR和MLR的记录条件下,脑电噪声符合1/f模型。为此,我们提出一个新的噪声增益因子度量序列性能。通过充分的实测数据检验,证实新的度量和实际CLAD刺激序列的噪声抑制能力高度相关。以此为目标函数,我们以经典的遗传算法为例,用于给定约束条件下序列的优化生成。本项研究较好地解决了上述两个关键问题,有利于促进CLAD方法推广应用。2.40Hz中潜伏期反应的特性对比研究表明刺激率对中潜伏期反应存在显著影响,刺激率为40 Hz时明显强化的稳态反应被认为和MLR的同相增强有关。但是也有确凿的实验说明40 Hz诱发反应的产生机制另有解释。现有线性去卷积技术可恢复40 Hz的暂态AEP成分。因此,本项工作试图回答两个基本问题:与常规5 Hz记录条件下的MLR相比,40 Hz暂态反应的基本特性如何;不同抖动方式的40 Hz刺激序列,对求解的暂态AEP的影响是否可鉴。我们选择比较CLAD范例和另一种不同抖动方式的 MSAD(multi-rate steady-state averaging deconvolution)范例下去卷积恢复的40 Hz暂态AEP,并与常规5 Hz记录条件下的AEP进行对比实验。结果表明MLR在快速刺激下形态更清晰稳定,并且刺激序列的不同抖动方式对MLR的一些成分有显著影响,说明MLR受刺激率显著影响,也受刺激序列设计的影响。3.非线性AEP重建中最大长序列的噪声衰减分析最大长序列(Maximum length sequence,MLS)是一种经典的去卷积刺激序列,可用互相关技术恢复线性和非线性AEP成分。MLS是基于不同阶数本原多项式产生的伪随机序列,不同序列也会导致计算时噪声的改变。本项研究旨在评估MLS的噪声衰减特性与阶数的关系。我们给出了叠加平均过程和互相关过程噪声衰减系数的理论公式,用仿真和真实的ABR非线性成分提取实验加以验证。理论推导和实验结果都表明噪声衰减与阶数无关,而是由有效的EEG数据总长度和MLS的刺激率决定的。本研究为MLS的选择提供了依据,并且可以用来估计MLS实验所需要的时间和信噪比。
[Abstract]:Under the action of short stimulus, the auditory system can obtain a limited duration of electrical signal through the recording of the scalp electrode, that is, a transient auditory evoked potential (Auditory evoked potential, AEP).AEP is very weak in need of repeated stimulation and can be identified after the average denoising. But if the stimulation repetition rate is too fast, AEP is not appease and stimulated. The generation leads to the loss of transient characteristics of the overlapping of AEP components. This overlap reaction is described as a linear convolution model of the transient response and the stimulus sequence. If the interaction of adjacent stimuli is considered, the nonlinear component and the nonlinear convolution model,.AEP, can be divided into auditory brainstem according to the latent period. The reaction (Auditory brainstem-response, ABR) and the medium latency reaction (Middle-latency response, MLR) etc. have been proposed for the restoration of transient AEP components. These methods have various characteristics and shortcomings. This paper systematically introduces the linear / non linear deconvolution and experimental paradigm in this field, and some of them are key questions. The solutions include the following three aspects: 1. the evaluation and optimization of the CLAD stimulus sequence under the real noise condition and the optimization CLAD (Continuous loop averaging deconvolution) are the most recent linear deconvolution methods, using the random jitter of the stimulus sequence to ensure the existence of the solution. However, there are two key points in the application of the method. Key problems: noise suppression performance evaluation and stimulus sequence optimization of stimulus sequences. This study indicates that the noise suppression ability of the sequence needs to be combined with the statistical characteristics of the noise itself. It is particularly confirmed that the EEG noise conforms to the 1/f model under the recording conditions of ABR and MLR. For this reason, we propose a new noise gain factor measure. The performance. Through the full test data test, it is proved that the new measurement and the noise suppression ability of the actual CLAD stimulus sequence are highly correlated. As the objective function, we use the classical genetic algorithm as an example to optimize the generation of the sequence under a given constraint condition. This study has solved the above two key problems well and is beneficial to promote the promotion. The comparison of the characteristics of the latent period reaction in the application of the CLAD method in.2.40Hz shows that the stimulation rate has a significant effect on the medium latency reaction. When the stimulation rate is 40 Hz, the obviously enhanced steady state reaction is considered to be related to the phase enhancement of MLR. But there are also definite experiments showing that the mechanism of the 40 Hz induced reaction is otherwise explained. Deconvolution technology can restore the transient AEP component of 40 Hz. Therefore, this work tries to answer two basic questions: what is the basic characteristic of the 40 Hz transient response compared to the conventional 5 Hz recording conditions; the 40 Hz stimulus sequence of different jitter modes is a reference to the response of the transient AEP. We choose to compare the CLAD paradigm and the other one. The MSAD (multi-rate steady-state averaging deconvolution) example of different jitter is carried out to convolution the 40 Hz transient AEP and compare with the conventional 5 Hz recording AEP. The results show that the MLR is more clear and stable under the rapid stimulus, and the different jitter modes of the sequence are displayed to some components of MLR. It shows that MLR is influenced significantly by the stimulus rate, and the maximum long sequence of noise attenuation analysis (Maximum length sequence, MLS), which affects the maximum long sequence of the.3. nonlinear AEP reconstruction, is influenced by the stimulus sequence design (MLS) is a classical deconvolution sequence, and the regression of linear and nonlinear AEP components.MLS is based on different orders by intercorrelation technology. This study aims to evaluate the relationship between the noise attenuation characteristics of MLS and the relation between the noise attenuation characteristics and the order. We give the theoretical formula of the noise attenuation coefficient of the superposition mean process and the intercorrelation process, and extract the experiment with the real and real ABR nonlinear components. The theoretical deduction and experimental results show that the noise attenuation is independent of the order, but is determined by the total length of the EEG data and the stimulus rate of MLS. This study provides the basis for the selection of MLS and can be used to estimate the time and signal to noise ratio required by the MLS experiment.
【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R318.0

【相似文献】