融合互信息和支持向量机的癫痫自动检测算法  

An Automatic Detection Algorithm for Epil TIFy EEG Based on MI and SVM

作　　者：沈洋洋[1] 黄丽亚[1] 郭迪[1] 笪铖璐 陈志阳[1] 戴加飞[2] 

机构地区：[1]南京邮电大学电子科学与工程学院,江苏南京210003 [2]南京军区南京总医院神经内科,江苏南京210002

出　　处：《计算机技术与发展》2016年第6期133-137,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61003237);江苏省高校自然科学研究项目(10KJB510018)

摘　　要：脑电图(Electroencephalogram,EEG)是通过电极记录下来的脑神经细胞群的自发性、节律性电活动,是癫痫诊断中最重要的一项检查工具。文中提出了一种新的基于互信息(Mutual Information,MI)和支持向量机(Support Vector Machine,SVM)的特征提取和分类的方法,可以高效地区分正常脑电信号和癫痫脑电信号,并分别对比了相同维度下不同特征向量组合以及不同维度的特征向量组合的分类效果。除此之外,还对比了文中算法与其他常用算法的分类效果和算法效率。实验结果表明,由两类脑电信号的互信息序列提取的以均值、方差组成的二维特征向量,具有运算简单、分类准确率高的优点,同时文中算法比其他常用算法具有更快的运算速度,这对于临床实时监控癫痫是否发作具有积极的指导意义。Electroencephalogram （EEG） is the most important tools for seizure detection by recording spontaneous and rhythmic electrical activity of brain cells through electrodes. A new method for feature extraction and classification is proposed based upon Mutual Informa- tion （MI） and Support Vector Machine （SVM）, which can distinguish epilepsy EEG from normal EEG quickly and efficiently. Then the comparison on the classification results is made using various combinations of feature vector in the same dimension and in the different di- mension. In addition, the classification results and efficiency are compared between proposed algorithm and other common algorithm. The experiment shows that the two-dimensional feature vectors combining mean and variance extracted from MI sequence of epilepsy EEG and normal EEG, has advantages of simple operation and high classification result, and this algorithrn is also faster than others, which is useful for clinical seizure detection in real time.

关 键 词：互信息 支持向量机 脑电信号 特征提取 癫痫自动检测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]