【作者】 杨文静；

【导师】 田心；

【作者基本信息】 天津医科大学 ， 生物医学工程， 2010， 硕士

【摘要】 研究目的：大脑的工作记忆是由不同模态的神经信号,以及不同脑区同一模态的神经信号相互协同来编码的。本论文基于大鼠工作记忆过程中在体前额叶皮层上获得的两类不同模态的多通道神经信号：多通道局部场电位(local field potentials, LFPs)和离散点电位形式的多通道峰电位(Spikes)为实验数据,发展和应用LFPs-Spikes的时变频谱相干(time-varying spectrum coherence)编码方法,及其对工作记忆任务的协同编码模式,为研究工作记忆的神经编码机制提供神经计算支持。研究方法：1.实验数据：来自本实验室,应用清醒动物在体多通道采集技术,在大鼠工作记忆过程中记录的前额叶皮层16通道神经信号。2.多通道LFPs和Spikes的时变谱相干方法：(1)16通道LFPs的时变频谱相干分布对16通道原始记录数据进行低通滤波(0-300Hz),获得16通道LFPs。利用加权最小二乘的局部线性回归与拟合方法消除16通道LFPs信号中夹杂的基线漂移与工频干扰,获取零均值16通道LFPs。分别计算每个通道LFPs的功率谱密度(Power Spectral Density, PSD),选取PSD分布集中的θ频段作为LFPs的特征频段。选取功率最大、变化最明显的通道作为计算相干频谱分布的参考通道；选取计算窗口为50ms,移动步长为12.5ms,从初始点开始逐个计算每个窗口中每个通道LFP以及LFP的θ分量对参考通道的频谱相干分析,分别绘制多通道LFPs及其θ分量的时变频谱相干动态分布图。(2)25个神经元动作电位时空序列的获取及其时变频谱相干编码对16通道原始记录数据进行高通滤波(≥300Hz),获得16通道高通信号。选取超出阈值-65μV且信噪比大于3.0认定为每个通道的峰电位(Spikes),或称为multi-unit。本论文中可以有效获得Spikes序列的通道是14个。应用美国Plexon公司提供的Offline Sorter神经信号分类软件对14个通道的Spikes逐一进行单个神经元放电的分类(sorting),获得25个神经元发放的动作电位时空序列。分别计算每个神经元窗口放电率{r(r)}i的PSD分布,选取PSD分布集中的θ频段为神经元群体放电时空序列的特征频段。从25个神经元中选取PSD最大、变化最明显的神经元作为参考神经元。选取计算窗口为50ms,移动步长为12.5ms,从初始点开始逐个计算每个窗口中每个神经元的平均放电频率,及其对参考神经元的频谱相干分析,获取神经元群体的时变频谱相干动态分布。(3)14通道LFPs-Spikes之间的时变频谱相干分布将14通道LFPs和14通道Spikes进行配对,得到14通道LFPs-Spikes数据对。选取计算窗口为50ms,移动步长为12.5ms,从初始点开始逐个计算每个窗口中每个通道LFPs-Spikes的频谱相干值,获取14通道LFPs-Spikes的时变频谱相干动态分布。计算每个通道LFP的整体PSD分布,同时利用点电位平均发放率的方法计算Spikes的PSD分布,选取特征频段(θ频段),如上计算θ频段LFPs-Spikes的时变相干动态分布。结果：1.16通道LFPs时变频谱相干模式(1)PSD峰值频率为(8.45±2.46)Hz,特征θ频段(4-13 Hz)占整体频段(0-120 Hz)的(48.89±3.05)%；(2)37次重复实验中,大鼠工作记忆事件前后1 s的时变频谱相干平均值分别为：①LFPs的θ分量：0.2404±0.0102与0.7825±0.0104,经t检验,事件点后1s的时变频域相干值比事件点前1 s显著增加(P<0.05)；②LFPs的全频段：0.3913±0.0189与0.4729±0.0178,经t检验,事件点前后相干值无显著增加(P>0.05)。③LFPs的其它分量,如δ分量：0.2113±0.0140与0.2621±0.0121,经t检验,事件点前后相干值无显著增加(P>0.05)。2.25个神经元动作电位时空序列时变频谱相干编码模式(1)PSD峰值频率为(8.25±3.12)Hz,特征θ频段(4-13 Hz)占整体频段(0-120 Hz)的(60.03±6.98)%；(2)37次重复实验中,大鼠工作记忆事件前后1 s的时变频谱相干平均值分别为：①时空序列的θ分量：0.1952±0.0064与0.7357±0.0083,经t检验,事件点后1 s的时变频域相干值比事件点前1 s显著增加(P<0.05)；②平均放电频率：0.2711±0.0046与0.3265±0.0038,经t检验,事件前后相干值无显著改变(P>0.05)。3.14通道LFPs-Spikes时变频谱相干模式(1) LFPs的PSD峰值频率为(7.86±2.74) Hz, Spikes的PSD峰值频率为(8.26±1.64)Hz,特征θ频段(4-13 Hz)占整体频段(0-120 Hz)的(46.21±4.07)%；(2)37次重复实验中,大鼠工作记忆事件前后1s的时变频谱相干平均值分别为：①LFPs-Spikes的θ分量：0.2222±0.0108与0.7786±0.0129,经t检验,事件点后1s的时变频域相干值比事件点前1s显著增加(P<0.05)；②LFPs-Spikes的全频段：0.2987±0.0077与0.3332±0.0088,经t检验,事件点前后相干值无显著改变(P>0.05)。结论：本论文研究16通道LFPs,25个神经元动作电位以及14通道LFPs-Spikes的时变频谱相干编码,结论如下：1.16通道LFPs及θ分量的时变相干谱分布模式(1) LFPs的θ频段是大鼠工作记忆编码的特征频段,16通道θ分量的时变频谱相干有效地编码了工作记忆事件：事件后1s的平均频谱相干值显著高于事件前(P<0.05)；(2)16通道LFPs编码大鼠工作记忆事件效果在P>0.05水平上不显著；其它分量(如δ分量)编码低于LFPs整体编码效果。2.25个神经元动作电位时空序列及其θ分量时变频谱相干编码模式(1)点电位表示的动作电位序列放电频率的θ分量有效地编码了工作记忆事件：事件后1s的平均频谱相干值显著高于事件前(P<0.05)；(2)平均放电频率编码大鼠工作记忆事件效果在P>0.05水平上不显著。3.14通道LFPs-Spikes时变相干谱协同模式(1) LFPs-Spikes的θ分量时变频谱相干有效地编码了工作记忆事件：事件后1s的平均频谱相干值显著高于事件前(P<0.05)；(2) LFPs-Spikes全频段编码大鼠工作记忆事件效果在P>0.05水平上不显著。更多还原

【Abstract】 ObjectiveWorking memory in brain is coded by different modal neural signals and same modal neural signals in different brain areas collaboratively. The study developed and applied time-varying spectrum coherence of LFPs-Spikes based on two types of multi-channel neural singals:local field potentials and discrete spikes, which acquired on prefrontal cortex of rats during working memory task in vivo. The collaborative coding pattern provides neural computation support for analyzing the neural coding mechanisms of working memory.Methods1. Experimental data, which come from our laboratory, were 16-channel neural signals on prefrontal cortex of rats during working memory task based on multi channel recording technology on awake animals in vivo.2. Methods on multi-channel time-varying spectrum coherence:(1) Time-varying spectrum coherence distribution of 16-channel LFPs16-channel LFPs were acquired through lowpass filtering the original data (0-300Hz). Local linear regression based on the weighted least square method was used to remove the baseline drift and power-line interference on LFPs.Power spectral density (PSD) on each channel LFP was performed respectively for characteristic theta band of LFPs. After choosing the reference channel LFP which had the largest PSD value, time-varying spectrum coherence dynamic distribution was available between each channel LFP and reference channel LFP with 50-ms multi-taper window sliding and 25% overlap, and the theta band of LFPs as well.(2) Time varying spectrum coherence of 25-neuron action potential time-space series.16-channel highpass signals were acquired through highpass filtering the original data (≥300Hz). Spikes, or multi-unit, were recognized above -65μV thresholds and 3.0 SNR.25-neuron action potential time-space series were got through sorting the 14-channel valid spikes by Offline Sorter software (Plexon, USA).PSD on each neuron firing rate was performed for characteristic theta band respectively. After choosing the reference neuron which had the largest PSD, time-varying spectrum coherence dynamic distribution was available with 50 ms multi-taper window sliding and 25% overlap.(3) Time-varying spectrum coherence distribution of 14-channel LFPs-Spikes14-channel LFPs-Spikes were matched between 14-channel LFPs and 14-channel spikes. Time-varying spectrum coherence dynamic distribution was available with 50 ms multi-taper window sliding and 25% overlap.PSD on each channel LFP and spike can be obtained through calculating firing rate for charactisric theta band, whose time-varying coherence distribution was computed as above methods.Results1.16-channel LFPs time-varying coherence pattern(1) PSD peak frequency:(8.45±2.46) Hz, occupied for (48.89±3.05) percents among all frequency band (0-120Hz);(2) Average coherence at the onset of working memory event±1s in repeated tests:①Theta band of LFPs:0.2404±0.0102 and 0.7825±0.0104, P<0.05;②All frequency band of LFPs:0.3913±0.0189 and 0.4729±0.0178, P>0.05;③Other band such as delta of LFPs:0.2113±0.0140 and 0.2621±0.0121,P>0.05.2.25-neuron action potential time-space series time-varying coherence pattern(1) PSD peak frequency:(8.25±3.12) Hz, occupied for (60.03±6.98) percents among all frequency band (0-120Hz);(2) Average coherence at the onset of working memory event±1s in repeated tests:①Theta band of series:0.1952±0.0064 and 0.7357±0.0083, P<0.05;②Average firing rate:0.2711±0.0046 and0.3265±0.0038,P>0.05.3.14-channel LFPs-Spikes time-varying coherence pattern(1) PSD peak frequency:(7.86±2.74) Hz for LFPs, (8.26±1.64) Hz for spikes, occupied for (46.21±4.07) percents among all frequency band (0-120Hz);(2) Average coherence at the onset of working memory event±1s in repeated tests:①Theta band of series:0.2222±0.0108 and 0.7786±0.0129, P<0.05;②All frequency band of LFPs-Spikes:0.2987±0.0077 and 0.3332±0.0088,P>0.05. ConclusionsThis paper aimed at time-varying spectrum coherence encoding among 16-channel LFPs,25-neuron action potentials time-space series and 14-channel LFPs-Spikes, and the conclusions as follows:1.16-channel LFPs and its theta band component(1) Theta band of LFPs is the characteristic band during working memory of rats, whose time-varying spectrum coherence encodes the working memory event effectively:the coherence value after event is better to the one before event significantly.(2) All frequency band of LFPs coherence doesn’t encode the working memory event, and other band such as delta of LFPs is lower than the former.2.25-neuron action potential time-space series and its theta band component(1) Time-varying spectrum coherence on theta band of action potential time-space series firing rate encodes the working memory event effectively:the coherence value after event is better to the one before event significantly.(2) Average firing rate of series coherence doesn’t encode the working memory event.3.14-channel LFPs-Spikes and its theta band component(1) Time-varying spectrum coherence on theta band of LFPs-Spikes encodes the working memory event effectively:the coherence value after event is better to the one before event significantly.(2) All frequency band of LFPs-Spikes coherence doesn’t encode the working memory event.更多还原