基于脑电的局灶性癫痫灶点定位研究

发布时间：2018-02-28 16:29

本文关键词： 灶点定位动态源成像功能连接黎曼流形癫痫网络动态　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：癫痫是中国当前第二大神经科疾病,由神经元放电异常引发,发作时患者会出现痉挛、意识丧失等症状,对患者的生理和心理都会造成严重的影响。目前常见的治疗手段为药物治疗、手术治疗、电刺激等,约70%的患者能够通过药物治疗得到有效控制,剩下的药物难治性癫痫患者需要手术切除治疗或者电刺激治疗。在手术切除治疗中,癫痫灶点定位是临床手术前极其重要的诊断步骤。由于癫痫发病原因和发作机制十分复杂,诊断和治疗非常困难。随着电子工艺、及计算机信息处理技术的发展,利用信号处理手段进行癫痫灶点定位的研究正在逐步兴起。头皮脑电(sEEG)空间覆盖率高,采集方便无植入式创伤,但是空间分辨率较低,适合用于粗略定位癫痫灶点区域;皮层脑电(iEEG)的时空分辨率高,但是空间覆盖率低且需要开颅手术,适合用于在初步定位的区域植入电极后精确定位癫痫灶点。本文主要围绕基于脑电信号(EEG)的局灶性癫痫灶点定位问题,着重探索基于头皮脑电(sEEG)动态源成像的癫痫灶点粗定位和基于皮层脑电(GEEG)黎曼流形的癫痫灶点精定位两个子问题。(1)sEEG信号采集点都是标准的电极位置,首先使用独立成分分析将sEEG信号分离成头皮表面的相互独立的信号成分,然后利用癫痫信号成分与非癫痫信号成分在时频特征上的差异挑选出癫痫信号成分,最后根据用边界元法构造的三层三维电子大脑模型将癫痫信号成分逆向映射回皮层的源信号,从而完成癫痫灶点定位。在源映射的过程中,我们根据大脑结构和生理机制的特点添加了平滑约束。(2)在通过sEEG初步定位的癫痫灶点区域,埋入植入式电极采集皮层脑电(iEEG)。首先计算表征通道之间相互关系的功能连接矩阵,然后根据功能连接矩阵分析癫痫网络与灶点有关的空间上的特征和时间上的特征。在分析癫痫网络动态时通过引入自定义度量-对数密度比ρ1发现癫痫发作期后半段与发作前期和发作期前半段有显著的差异,为我们将发作期聚类成两个状态,并取后一个状态进行灶点定位提供依据。我们提出了基于黎曼距离的k-means聚类算法,利用整个功能连接矩阵而不是有损压缩的特征向量中心性等进行状态聚类。最后我们基于聚类得到的后一个状态进行灶点定位,取得了比较理想的效果。
[Abstract]:Epilepsy is currently the second largest neurologic disease in China. It is caused by abnormal firing of neurons, and when the seizure occurs, the patient will have symptoms such as spasm, loss of consciousness, etc. At present, the common treatment methods are drug therapy, surgical treatment, electrical stimulation, etc. About 70% of the patients can be effectively controlled through drug therapy. The remaining patients with drug-refractory epilepsy need surgical resection or electrical stimulation. In surgical resection, the location of epileptic foci is an extremely important diagnostic step before clinical operation. Because of the complexity of the cause and mechanism of epilepsy, Diagnosis and treatment are very difficult. With the development of electronic technology and computer information processing technology, the study of epileptic foci location using signal processing means is gradually rising. The acquisition is convenient without implantation trauma, but the spatial resolution is low, which is suitable for rough location of epileptic foci. The cortical EEG has high spatial and temporal resolution, but the space coverage is low and the craniotomy is required. This paper focuses on the localization of focal epileptic foci based on EEG-based EEG. The two sub-problems of epileptic focal point coarse location based on scalp EEG dynamic source imaging and epileptic focal fine location based on EMG Riemann manifold are all standard electrode positions. Firstly, the sEEG signal was separated into independent signal components on the scalp surface by independent component analysis (ICA), and then the epileptic signal components were selected by using the difference between the epileptic signal components and the non-epileptic signal components in time-frequency characteristics. Finally, according to the three-layer three-dimensional electronic brain model constructed by the boundary element method, the epileptic signal components are reversely mapped back to the source signal of the cortex to locate the epileptic foci. According to the characteristics of brain structure and physiological mechanism, we added smoothing constraint. 2) in the region of epileptic foci preliminarily located by sEEG, we embedded an implanted electrode to collect EEGG. First, we calculated the functional connection matrix, which was used to characterize the relationship between channels. Then, the spatial and temporal characteristics of epileptic network related to focal points are analyzed according to the functional connection matrix. In the analysis of epileptic network dynamics, the seizure period is found by introducing the custom metric logarithmic density ratio 蟻 1. There were significant differences between the half segment and the preictal and the first half of the attack. It provides the basis for us to cluster the seizure period into two states, and then take the latter state to locate the focal point. We propose a k-means clustering algorithm based on Riemann distance. The whole functional connection matrix rather than the feature vector centrality of lossy compression is used for state clustering. Finally, we use the latter state to locate the focal point, and obtain a better result.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R742.1;TN911.7

【相似文献】