基于频域Granger因果分析研究颞叶癫痫多通道脑电的功能连接和因果网络特性

发布时间：2018-03-13 08:28

本文选题：颞叶癫痫　切入点：多通道脑电　出处：《天津医科大学》2014年硕士论文　论文类型：学位论文

【摘要】：研究目的：癫痫是常见脑部疾病,对其发病机制的研究具有重要的临床价值。本论文从多通道脑电的功能性连接和因果网络的角度,以颞叶癫痫间歇期、发作前期、发作期和正常对照组的多通道脑电为研究对象,研究癫痫发作期过度放电的脑网络机制,为癫痫机制研究提供思路和技术上的支持。研究方法： 1实验数据：实验数据来自天津医科大学总医院神经外科功能室,临床确诊的颞叶癫痫组(年龄在19岁至47岁),正常对照组(年龄在24岁至30岁)各8例。癫痫致痫灶均位于前颞区(头皮电极F7、F8)。分别采集颞叶癫痫组间歇期、发作前期、发作期和正常对照组共4个组各10段,每段10秒的16通道脑电(EEGs)数据。 2数据预处理：对EEGs进行去基线漂移和去工频干扰,并应用Matlab FastICA算法去除明显伪迹。 3时频分析：应用短时傅里叶变换对EEGs逐一通道进行时频分析,研究能量最大通道的颞叶癫痫组间歇期、发作前期、发作期和正常对照组EEGs的时频分析,从中获取癫痫发作期的主要频率范围：6和θ。 4应用带通滤波获取EEGs的δ和0频段分量。 5分别对4个组每1段EEGs的6和0频段分量进行Granger频域因果关系分析： (1)对16通道EEGs时间序列进行频域因果分析,计算两个节点之间因果值γij,平均后得到定向传递函数(Directed Transfer Function, DTF)值。 (2)应用γij构建因果网络,计算网络的BC度量值(Betweenness Centrality,BC)。 (3)BC值的K均值聚类分析：对颞叶癫痫组间歇期、发作前期、发作期整个过程的BC度量值进行K均值聚类分析,将网络各个节点分成两类：活跃节点和不活跃节点,并且分别分析BC度量值的变化趋势；以及对发作期的BC度量值进行K均值聚类分析。 (4)计算发作期网络的因果密度(Causal Density, CD): 计算颞叶癫痫组发作期各个节点的因果密度值,并对因果密度值进行K均值聚类分析,和发作期BC度量值的活跃节点比较。 (5)计算网络的因果流(Causal Flow, CF): 计算发作期活跃节点和非活跃节点的因果流值,以及间歇期、发作前期与活跃节点相对应节点的因果流值并比较分析。研究结果： 1时频分布：为颞叶癫痫发作期能量集中在低频段：δ和0频段。 24个组多通道脑电的频域功能性连接 (1)6频段：颞叶癫痫组：间歇期DTF均值(7.3404±1.9629)；发作前期DTF均值(4.8755±1.0541)；发作期DTF均值(8.1770±1.6978)。正常对照组DTF均值(2.1591±0.5561)。 (2)θ频段：颞叶癫痫组：间歇期DTF值(8.1263±0.4356)；发作前期DTF值(3.5955±0.8756)；发作期DTF值(8.2826±0.4899)。正常对照组DTF值(2.7373±0.6542)。 3颞叶癫痫组间歇期、发作前期和发作期BC度量值K均值聚类分析 (1)6频段活跃节点,间歇期BC度量值(0.0499±0.0149)；发作前期BC度量值(0.0469±0.0095)；发作期BC度量值(0.0800±0.0200)。 (2)0频段活跃节点,间歇期BC度量值(0.0467±0.0108)；发作前期BC度量值(0.0437±0.0076)；发作期BC度量值(0.0651±0.0126)。 4颞叶癫痫组发作期BC度量值与CD值的K均值聚类分析比较在δ和0各个频段,BC度量值的活跃节点和CD值的活跃节点重合率很高,两者相同,或者前者包括了后者,或者后者包括了前者；两者的位置与致痫灶位置并不相符。 5颞叶癫痫组发作期活跃节点的因果流分析 (1)δ频段发作期活跃节点CF值(0.6864±0.3037),间歇期、发作前期与活跃节点相对应的区域CF值(0.1495±0.1358)、(0.1174±0.0648)经过验证皆符合正态分布,发作期活跃节点CF值分别与间歇期、发作前期相对应的区域CF值进行t检验,p0.05,均有显著性差异。 (2)0频段发作期活跃节点CF值(0.6661±0.1645),间歇期、发作前期与活跃节点相对应的区域CF值(0.0774±0.1973)、CF值(0.1489±0.0570),经过验证皆符合正态分布,发作期活跃节点CF值分别与间歇期、发作前期相对应的区域CF值进行t检验,p0.05,均有显著性差异。结论： 1颞叶癫痫发作期能量集中频带为δ和0频段。 2颞叶癫痫组发作期比间歇期、发作前期功能连接特性增强；发作期比正常对照组功能连接特性增强；间歇期和发作前期功能连接性无明显差异；间歇期、发作间期比正常对照组功能连接特性增强。 3颞叶癫痫发作期活跃节点因果网络的BC度量值比间歇期、发作前期增大,间歇期和发作前期BC度量值无明显差异。 4颞叶癫痫发作期根据因果密度和BC度量值分别K均值聚类分析得到的活跃节点位置相近。 5颞叶癫痫发作期活跃节点因果流值较大属于因果源；非活跃节点因果流值较小属于因果汇。
[Abstract]:The purpose of the study is:
Epilepsy is a common disease of the brain, the research on the pathogenesis of this disease has important clinical value. This paper from the function of multi-channel connection and causal network perspective, with intermittent temporal lobe epilepsy, early attack, attack of multi-channel EEG and normal control group as the research object, the mechanism of brain network excessive discharge of epileptic seizures, provide ideas and technical support for the mechanism of epilepsy.
Research methods:
1 experimental data: the experimental data from the Department of Neurosurgery of General Hospital Affiliated to Tianjin Medical University function room, clinical diagnosis of temporal lobe epilepsy group (aged 19 to 47 years old), normal control group (aged 24 to 30 years). All 8 cases of epileptogenic focus were located in the anterior temporal region (scalp electrodes F7, F8) were collected. Period of intermittent temporal lobe epilepsy pre ictal and normal control group 4 group 10, 16 channel EEG every 10 seconds (EEGs) data.
2 data preprocessing: baseline drift and power frequency interference to EEGs, and Matlab FastICA algorithm to remove obvious artifacts.
3 time frequency analysis of EEGs: one by one channel based on short time Fourier transform time-frequency analysis, temporal lobe epilepsy group of intermittent period, maximum energy channel before the attack, attack analysis and control group when the frequency of EEGs, get the main frequency range of seizure period from 6 and 0.
4 the band pass filter is used to obtain the Delta and 0 band components of EEGs.
5 the Granger frequency domain causality analysis of the 6 and 0 band components of each 1 segments of EEGs in 4 groups was analyzed, respectively.
(1) frequency domain causality analysis for 16 channel EEGs time series. We calculate the causal value ij between two nodes, and get the Directed Transfer Function (DTF) value after averaging.
(2) the causality network is constructed with gamma ij, and the BC measurement value (Betweenness Centrality, BC) of the network is calculated.
(3) K mean clustering analysis of BC value:
The temporal lobe epilepsy group intermittent period, early attack, attack the whole process of the BC measure of K clustering, every node in the network will be divided into two categories: active and inactive node node, and the variation trend of the BC value respectively; and to attack the BC measure of K clustering analysis.
(4) calculate the causality density (Causal Density, CD) of the attack network:
The causal density values of each node in the temporal lobe epilepsy group were calculated, and the K density clustering analysis of the causal density values was compared with the active nodes in the BC period.
(5) computing network causality flow (Causal Flow, CF):
The causal flow values of active nodes and inactive nodes in the attack phase are calculated, and the causal flow values corresponding to the nodes in the early stage and the active nodes are compared and analyzed.
The results of the study:
1 time frequency distribution: the energy of the temporal lobe epilepsy is concentrated at low frequency segments: Delta and 0 frequency bands.
Frequency domain functional connection of 24 groups of multichannel EEG
(1) 6 frequency bands:
Temporal lobe epilepsy group: DTF mean (7.3404 + 1.9629) in intermission, DTF mean (4.8755 + 1.0541) in pre attack, and DTF mean (8.1770 + 1.6978) at attack stage.
The mean value of DTF in the normal control group was (2.1591 + 0.5561).
(2) frequency band:
Temporal lobe epilepsy group: DTF value in intermission (8.1263 + 0.4356), DTF value (3.5955 + 0.8756) in pre attack and DTF (8.2826 + 0.4899) at attack stage.
The value of DTF in the normal control group was (2.7373 + 0.6542).
3 temporal lobe epilepsy group intermission, K mean cluster analysis of BC measurement at prophase and episodes
(1) active node in 6 frequency band, BC measure (0.0499 + 0.0149) in intermittent period, BC measure (0.0469 + 0.0095) in pre attack, and BC measure (0.0800 + 0.0200) at attack stage.
(2) active node in 0 frequency band, BC measure (0.0467 + 0.0108) in intermittent period, BC measure (0.0437 + 0.0076) in pre attack, and BC measure (0.0651 + 0.0126) at attack stage.
Comparison of BC and CD values of temporal lobe epilepsy group in 4 temporal lobe epilepsy group by K mean cluster analysis
In every band of delta and 0, the coincidence rate of active nodes and active nodes of BC value is very high. The coincidence rate between them is the same, or the former includes the latter, or the latter includes the former; the location of the CD is not consistent with the location of epileptogenic foci.
Causality flow analysis of active nodes in the 5 temporal lobe epilepsy group
(1) delta band ictal active node CF values (0.6864 + 0.3037), intermittent period, regional CF early attack and active node value corresponding to (0.1495 + 0.1358), (0.1174 + 0.0648) were tested with normal distribution, exacerbation of active node CF respectively and intermittent period, hair area the CF value corresponding to the t test, P0.05, there were significant differences.
(2) 0 band ictal active node CF values (0.6661 + 0.1645), intermittent period, regional CF early attack and active node value corresponding to (0.0774 + 0.1973), CF (0.1489 + 0.0570), after verification with normal distribution, exacerbation of active node CF respectively and intermittent period, attack region pre CF value corresponding to the t test, P0.05, there were significant differences.
Conclusion:
1 the energy concentration band of temporal lobe epilepsy is delta and 0 frequency bands.
2 temporal lobe epilepsy onset period than the intermittent period, connecting the characteristics to enhance the early attack; attack than the normal control group functional connectivity enhancements; intermittent period and early attack functional connectivity had no significant difference; intermittent period, interictal than the normal control group the function of connection characteristics increase.
3 the BC measurement values of active node causality network in temporal lobe epilepsy were more than that in the intermission period, the pre attack increased, and there was no significant difference in the BC measurement between the intermittent period and the pre attack stage.
4 the active nodes of temporal lobe epilepsy are similar to the active nodes according to the K mean clustering analysis of the causal density and the BC measure respectively.
5 the causality flow value of active node in temporal lobe epilepsy is larger than that of causality, and the value of causality flow of non active node is smaller than that of causality.

【学位授予单位】：天津医科大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R742.1

【共引文献】