言语诱发反应中频率跟随反应部分的瞬时能量谱分析

发布时间：2018-09-04 13:18

【摘要】：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称为听觉诱发电位(Auditory Evoked Potentials, AEP),目前最为人熟知的是听性脑干反应(Auditory Brainstem Response, ABR),其中最常用的是短声(click)刺激引出的ABR(click Evoked Brainstem Response, click-ABR),健康人的click-ABR一般包含7个特征波：Ⅰ～Ⅶ波,各波的潜伏期不同,因为它们都有各自的起源。简单声(比如短声、纯音)诱发的ABR广泛应用于临床听阈估计、检查听觉通路中的病变、估计听觉通路的完整性等。尽管如此,但由于这些简单声与我们日常生活所接触到得复杂声(如言语、环境噪声等)差别甚大,它们引出的ABR能用来表征听得见,但却不能用来表述是否能听懂。人类环境中的复杂声具有丰富的谐波、频率信息变化迅速等特点。这些特性的在脑干中的编码主要分为精确的时域编码和频谱编码,引出的ABR包含瞬态反应和持续反应两个部分。这些都是与刺激的声学特性密切相关的,瞬态反应与非周期特征有关,而持续反应则与刺激的周期性特征有关。尽管短声和纯音在各自引出的ABR中较好的表征了瞬态反应和持续反应模式,用它们却无法估计出同时包含瞬态和持续特征的复杂声引出的ABR。因此,相关学者开始研究复杂声刺激引出的ABR。其中目前较为成熟的是合成言语(speech)单音节/da/诱发的ABR相关研究。时间长度为40ms的合成言语/da/诱发出的speech-ABR(da-ABR)研究已较为成熟。/da/包含瞬态和持续两个部分,前者为辅音/d/,持续时间为10ms,后者为元音/a/,持续时间为30ms。元音/a/包含3个准周期波：d、e、f波,周期约为10ms,这部分含有基频(Fo)以及五个共振峰(F1-F5)的信息,其频率分别：F0=103～121Hz,F1=220-720Hz,F2=1700～1240Hz,F3=2580-2500Hz,F4.5=3600-4500Hz。 /da/引出的言语诱发脑干反应(Speech Evoked ABR, speech-ABR)可用来研究刺激的声学特征是如何在听觉系统中被编码的,并且其结构与/da/的声学结构非常相似,也可分为瞬态和持续两个部分。前者包括起始反应(onset reponse,OR),后者包括频率跟随反应(frequency following response, FFR)。其中OR(V、A波)为辅音/d/引出的反应,FFR部分(D、E、F波)被认为是刺激中的周期部分元音/a/引出的反应。FFR部分的D、E、F波为准周期波,分别为/a/中对应准周期波d、e、f引出,继承了/a/的周期特性,周期为10ms左右,与/a/的周期基本相同。有研究表明FFR部分中的三个准周期波具有显著相关性,提示可将FFR部分视为一个整体性的生理指标并可能源自相同的神经处理机制。FFR部分时域波形的周期性表征了脑干是如何编码元音/a/的基频信息,而基频信息是辨识言语的最重要信息,因此,可根据在speech-ABR的时域波形D、E、F波的完整性来判断脑干对言语/a/编码的程度。而在实际应用中,虽然所有受试者的总体平均波形FFR都非常清晰,三个准周期波容易分辨,但对单个受试者的speech-ABR进行分析时,FFR部分一般都很模糊,难以直观分辨出三个准周期波,这就说明需要采用信号处理技术在时域中突出FFR部分的准周期波,有利于临床应用FFR部分来判断脑干是否对言语/a/进行了编码或者编码完整。由于听觉系统的非线性特点,speech-ABR是非线性、非平稳信号,而希尔伯特黄变换(Hilbert-Huang Transform, HHT)是现在发展较成熟的处理非线性、分平稳信号的方法。HHT分为两个步骤：经验模态分解(Empirical Mode Decomposition, EMD)和希尔伯特(Hilbert)变换。EMD是将非平稳数据分解为有限个本征模态函数(ntrinsic Mode Function, IMF),每一层IMF基本都是平稳信号,并且IMF在任意时间点的振动模式唯一,即瞬时频率唯一。EMD是按照信号本身时间局部特征分解的,是自适应的。speech-ABR原始时域波形中的任意时间点可能包含多个瞬时频率,因为FFR部分的三个准周期波表征了元音/a/的基频信息,则它们的振动模式应该相似,经EMD后,它们应该会集中分布在同一层或相同几层IMF上。本文提出一种基于希尔伯特黄变换的瞬时能量谱方法来分析speech-ABR中的频率跟随反应部分,因为speech-ABR忠实的模拟了刺激信号的波形结构,再加上基于希尔伯特包络的瞬时能量能够突出极值,所以瞬时能量谱可能会比直接在原始时域波形中根据幅值绝对值最大来检测FFR部分跟有效。本文用实验证明此方法可以将FFR部分从speech-ABR中突显出来,有利于进一步的分析和临床应用。该方法的具体步骤如下： (1)将个体speech-ABR进行经验模态(empirical mode decomposition, EMD)分解,得到有限个本征模态函数(intrinsic mode function, IMF); (2)计算每一层IMF的瞬时能量谱； (3)观察个体speech-ABR分解后IMFs的瞬时能量谱,记录准周期波D、E、F出现的层数； (4)利用时窗为10ms,对该层瞬时能量谱17-47ms范围内寻找D、E、F波,记录下潜伏期和极性。实验记录了29例成人的speech-ABR。实验过程包括四个步骤。首先,临床专家根据V波是否明显找出有效数据,总共有25例(86%)。第二,临床专家在有效数据中根据FFR部分是否明显,将其分为两组：典型组和非典型组。其中典型组有7例(28%),非典型组有18例(72%),非典型组的个体speech-ABR中FFR部分不明显。第三,对着25例有效数据采用基于HHT的瞬时能量谱方法进行处理。最后,临床专家在这25例有效数据的瞬时能量谱中观察FFR部分,发现有23例(92%)FFR明显,有2(8%)例无法判断。结果表明,本文提出的方法突出表征了个体speech-ABR中的FFR部分,对检测FFR部分确实有效。该方法与常规方法相比有如下两个突出优点：(1)个体speech-ABR中的FFR部分在瞬时能量谱中明显比在speech-ABR时域波形中清晰可辨；(2)个体speech-ABR中FFR部分的准周期波极性不尽相同,可为正波可为负波,但在瞬时能量谱中,均为正波。可见,此方法比常规方法更有效更准确,可用于临床上直接或辅助检测D、E、F波。
[Abstract]:Auditory Evoked Potentials (AEP) are the bioelectrical changes in the central nervous system associated with external stimuli produced by specific sound stimulation of the auditory system. At present, the most well-known is Auditory Brainstem Response (ABR), the most common of which is the click-induced ABR (ABR). Click Evoked Brainstem Response (click-ABR), click-ABR in healthy people generally contains seven characteristic waves: _~_waves, the latency of each wave is different, because they have their own origins. Simple sound (such as short sound, pure tone) induced ABR is widely used in clinical auditory threshold estimation, examining the pathological changes in auditory pathways, estimating the integrity of auditory pathways. Nevertheless, because these simple sounds are so different from the complex sounds we are exposed to in our daily life (such as speech, ambient noise, etc.), their ABR can be used to indicate hearing, but it can not be used to express whether we can understand.
Complex sound in human environment has the characteristics of abundant harmonics and rapid change of frequency information.The encoding of these characteristics in the brain stem is mainly divided into precise time domain encoding and spectrum encoding.The ABR derived from ABR consists of transient response and continuous response.These are closely related to the acoustic characteristics of stimuli.Transient response and non-cyclic response are closely related. Although the short and pure tones represent the transient and persistent response patterns better in their respective ABRs, it is impossible to estimate the ABR induced by complex sounds with both transient and persistent characteristics. The most mature ABR. is the ABR related study of synthetic speech (speech) monosyllabic /da/.
The study of speech-ABR (d a-ABR) induced by 40 ms synthetic speech/d a/contains two parts: transient and persistent, the former being consonant/d/, lasting 10 ms, the latter being vowel/a/, lasting 30 ms. The frequencies of the five formants (F1-F5) are: F0 = 103-121Hz, F1 = 220-720Hz, F2 = 1700-1240Hz, F3 = 2580-2500Hz, F4.5 = 3600-4500Hz.
Speech Evoked ABR (speech-ABR) can be used to study how the acoustic characteristics of stimuli are encoded in the auditory system, and its structure is very similar to the acoustic structure of / DA / and can be divided into transient and persistent parts. Following response (FFR). In which OR (V, A) is a consonant / D / induced response, FFR (D, E, F) is considered to be a periodic part of the vowel / A / induced response in stimulation. In FFR, D, E, F waves are quasi-periodic waves, which are corresponding to quasi-periodic waves d, e, f in / A / respectively, inheriting the periodic characteristics of / A / with a period of about 10 ms. The periodicity of FFR partial time domain waveforms indicates how the brainstem encodes the fundamental frequency information of vowel/a/and the fundamental frequency information. Interest is the most important information for speech recognition, so the degree of speech/a/coding in the brain stem can be judged by the integrity of the D, E and F waves in the speech-ABR time domain. It is difficult to distinguish three quasi-periodic waves intuitively because of the ambiguity of the FFR part. This indicates that signal processing technology should be used to highlight the quasi-periodic wave of the FFR part in time domain, which is helpful for clinical application of the FFR part to judge whether the speech/a/is encoded or encoded intact by the brain stem.
Because of the nonlinearity of auditory system, speech-ABR is a non-linear and non-stationary signal. Hilbert-Huang Transform (HHT) is a well-developed method to deal with non-linear and stationary signals. HHT is divided into two steps: Empirical Mode Decomposition (EMD) and Hilbert (Hilbert). EMD decomposes non-stationary data into a finite number of intrinsic mode functions (IMFs). Each IMF layer is basically a stationary signal, and the IMF oscillation mode at any point in time is unique, that is, the instantaneous frequency is unique. EMD is decomposed according to the local time characteristics of the signal itself, and is adaptive. Speech-ABR original time domain wave. Any point of time in the form may contain multiple instantaneous frequencies, because the three quasi-periodic waves in the FFR part represent the fundamental frequency information of vowels/a/, then their vibration modes should be similar. After EMD, they should be concentrated on the same layer or the same IMF.
An instantaneous energy spectrum method based on Hilbert-Huang transform is proposed to analyze the frequency-following response in speech-ABR. Because speech-ABR faithfully simulates the waveform structure of the stimulus signal and the instantaneous energy based on Hilbert envelope can highlight the extreme value, the instantaneous energy spectrum may be more direct than the original time. In this paper, the experimental results show that this method can highlight the FFR part from speech-ABR, which is conducive to further analysis and clinical application.
(1) The individual speech-ABR is decomposed into empirical mode decomposition (EMD) and a finite number of intrinsic mode functions (IMF) are obtained.
(2) calculate the instantaneous energy spectrum of each layer of IMF.
(3) observe the instantaneous energy spectrum of IMFs after speech-ABR decomposition, and record the number of layers of D, E and F.
(4) The latency and polarity of D, E and F waves in the 17-47 MS instantaneous energy spectrum of the layer are recorded by using a time window of 10 ms.
Twenty-nine adults were recorded for speech-ABR. The procedure consisted of four steps. Firstly, clinical experts identified valid data based on whether the V wave was evident in 25 patients (86%). Secondly, according to whether the FFR part was evident in the valid data, clinical experts divided them into two groups: the typical group (7 cases) and the atypical group (28%). Thirdly, the instantaneous energy spectrum based on HHT was used to process 25 valid data. Finally, clinical experts observed the FFR in the instantaneous energy spectrum of 25 valid data, and found that 23 cases (92%) had significant FFR and 2 cases (8%) could not judge.
The results show that the proposed method highlights the FFR part of the individual speech-ABR and is effective for detecting the FFR part. Compared with the conventional method, this method has two outstanding advantages: (1) the FFR part of the individual speech-ABR is clearly distinguished in the instantaneous energy spectrum than in the speech-ABR time domain waveform; (2) the individual speech-ABR part is clearly distinguished in the instantaneous energy spectrum; (2) the individual speech-ABR part is distinguished in the speech-ABR time domain waveform. The polarity of the quasi-periodic wave in the FFR part is different from that in the FFR part, it can be positive wave, but it is positive wave in the instantaneous energy spectrum.
【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R318.0

【相似文献】