基于皮层脑电的癫痫脑网络研究及癫痫预测探索

发布时间：2018-06-29 03:48

本文选题：癫痫 + 皮层脑电信号(ECoG)　；参考：《电子科技大学》2017年硕士论文

【摘要】：癫痫是一种神经系统疾病,癫痫病人发作时会有一系列临床表现,有些行为会对癫痫病人有一定的损害,如烧伤、淹死、骨折、事故、死亡等等。随着现代医学的发展和进步,许多癫痫学家开始研究癫痫,治疗癫痫的方法和手段也随之增多,许多癫痫患者因此而得到治疗恢复了正常的生活。虽然现代医学的发展并没有解决癫痫病人的所有问题,但是,人们仍然对全面控制癫痫充满了希望。近十年来,通过对临床脑电信号的分析,人们对大脑活动的理解从大脑活动的不同点的鉴别到包含信息处理和任务机制的大脑网络的鉴别上。大量研究证明,脑网络方法能够很好的定位癫痫,因此本文研究了癫痫脑网络的癫痫灶定位方法,同样验证了脑网络方法对定位的有效性,并通过癫痫脑网络机制分析探索癫痫的发作机制。另外对于一些难治性癫痫病人来说,研究出一种精确有效的癫痫预测方法是很重要的,现有的癫痫预测方法由于数据库的限制和算法本身的缺陷至今没有一种能有效应用于临床的算法,因此本文提出了一种基于癫痫脑网络的癫痫预测方法,此方法能够准确预测癫痫发作。本文围绕癫痫发作区域定位和癫痫预测等两个方面,深入地开展了如下三项内容的研究:首先,本文对癫痫脑网络理论进行分析研究,为癫痫发作区域(Seizure Onset Zone,SOZ)定位和癫痫预测提供基础。然后,本文研究了基于癫痫脑网络的癫痫发作区域定位方法。利用有向传递函数方法(Directed Transfer Function,DTF)分析皮层脑电信号(Electrocorticogram,ECoG)之间的连接性,并利用图论方法(度中心性、中介中心性和接近中心性)计算网络属性,根据癫痫发作前后不同电极位置网络属性的变化定位癫痫发作区域。利用该方法对三个临床癫痫病人的7个发作数据进行了实验验证,结果表明了此方法在定位上的有效性。最后,根据发作前后的癫痫脑网络特征,提出了一种基于癫痫脑网络的癫痫预测方法,此方法利用支持向量机(Support Vector Machine,SVM)对三个临床病人7个癫痫发作数据的三个时期进行分类处理,三个病人的平均敏感性、特异性和准确性分别为94.82%、91.55%和93.14%,预测时间分别为35秒,60秒和85秒。进一步把此方法应用在头皮脑电数据库中,计算了数据库中的四个癫痫病人的数据,敏感性、特异性和准确性的平均值分别为73.70%、88.32%和80.54%。研究结果表明基于癫痫网络的方法在皮层脑电的定位和预测方面特异性较强,但是对头皮脑电的预测效果较差。本文通过对基于癫痫脑网络的癫痫定位和预测研究,得到的成果为今后癫痫机制分析和预测准确性的研究提供了基础和研究的方向。同时,本文提出的癫痫预测算法有一定的工程应用前景。
[Abstract]:Epilepsy is a nervous system disease, epileptic patients will have a series of clinical manifestations, some behavior will have certain damage to epileptic patients, such as burns, drowning, fractures, accidents, deaths and so on. With the development and progress of modern medicine, many epileptists have begun to study epilepsy, and the methods and means of treating epilepsy have increased, and many epileptic patients have been treated and returned to normal life. Although the development of modern medicine does not solve all the problems of epileptic patients, there is still hope for full control of epilepsy. In the past decade, through the analysis of clinical EEG, people's understanding of brain activity is from the different points of brain activity to the identification of brain network which includes information processing and task mechanism. A large number of studies have proved that the brain network method can locate epilepsy very well, so this paper studies the epileptic focus localization method of epileptic brain network, and also verifies the effectiveness of brain network method in localization. The seizure mechanism of epilepsy was explored by analyzing the mechanism of epileptic brain network. In addition, for some patients with refractory epilepsy, it is important to develop an accurate and effective method for predicting epilepsy. Due to the limitations of the database and the shortcomings of the algorithm, the existing epileptic prediction methods have not been effectively applied to clinical applications. Therefore, a epileptic prediction method based on epileptic brain network is proposed in this paper. This method can accurately predict epileptic seizures. This paper focuses on two aspects of epileptic seizure location and epileptic prediction, in-depth research on the following three aspects: first, this paper analyzes the theory of epileptic brain network. To provide the basis for the location and prediction of epileptic seizure area (Seizure Onset Zonesoz). Then, this paper studies the epileptic seizure region location method based on epileptic brain network. Using directed transfer function (DTF) to analyze the connectivity between electrocorticogram (ECoG), and using graph theory (degree centrality, intermediary centrality and proximity centrality) to calculate the network properties. According to the changes of the network properties of different electrode positions before and after epileptic seizure, the epileptic seizure region was located. Seven seizure data of three clinical epileptic patients were tested by this method, and the results showed that the method was effective in localization. Finally, according to the characteristics of epileptic brain network before and after seizure, a epileptic prediction method based on epileptic brain network is proposed. In this method, support Vector Machine (SVM) was used to classify the three periods of 7 epileptic seizures in three clinical patients, and the average sensitivity of the three patients was obtained. The specificity and accuracy were 91.55% and 93.14%, respectively, and the predicted time was 35 seconds, 60 seconds and 85 seconds, respectively. Furthermore, the method was applied to the scalp EEG database. The data of four epileptic patients in the database were calculated. The average values of sensitivity, specificity and accuracy were 73.7088.32% and 80.54%, respectively. The results show that the method based on epileptic network is more specific in the localization and prediction of cortical EEG, but it has a poor prediction effect on scalp EEG. In this paper, the localization and prediction of epilepsy based on epileptic brain network are studied. The results provide the basis and research direction for the analysis and prediction accuracy of epilepsy mechanism in the future. At the same time, the epileptic prediction algorithm proposed in this paper has a certain engineering application prospect.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R742.1;O157.5

【相似文献】