基于MPE-ANN-SVM的癫痫脑电检测分类研究  

Study on EEG classification of epilepsy based on MPE-ANN-SVM

作　　者：石雨菲 王瑶 胡珊[1] 田翔华[3] 陈子怡[4] 周毅[1] SHI Yufei;WANG Yao;HU Shan;TIAN Xianghua;CHEN Ziyi;ZHOU Yi(Zhongshan Medical College,Sun Yat-sen University,Guangzhou 510080,China;School of Biomedical Engineering,Sun Yat-sen University,Guangzhou 510006;College of Medical Engineering and Technology,Xinjiang Medical University,Urumqi 830011,China;First Affiliated Hospital of Sun Yat-sen University,Guangzhou 510080)

机构地区：[1]中山大学中山医学院,广州510080 [2]中山大学生物医学工程学院,广州510006 [3]新疆医科大学医学工程技术学院,乌鲁木齐830011 [4]中山大学附属第一医院,广州510080

出　　处：《生物医学工程研究》2022年第4期353-358,共6页Journal Of Biomedical Engineering Research

基　　金：国家重点研发计划项目(2021YFC2009402,2022YFC3601600);国家自然科学基金资助项目(61876194);广东省自然科学基金资助项目(2021A1515011897);广州市重点研发计划(20220610028)。

摘　　要：针对以往检测癫痫的脑电自动分类算法存在泛化性和鲁棒性不足的问题,本研究综合了非线性动力学与机器学习的优点,使用多尺度排列熵(MPE)与人工神经网络提取脑电特征、支持向量机进行分类决策,提出了基于MPE-ANN-SVM的癫痫脑电分类模型。该模型在13例临床数据上多次测试,平均识别率、敏感性和特异性分别达到94.01%、94.51%和93.52%。该模型分类指标均优于同样本下的人工神经网络模型分类结果与支持向量机基线模型,能够为癫痫脑电检测算法研究提供支持。In view of the problem of insufficient generalization and robustness of the previous EEG automatic classification model used for seizure diagnosis,we synthesized the advantages of nonlinear dynamics and machine learning.The multi-scale permutation entropy(MPE)and artificial neural network were used to extract EEG features,the support vector machine was used for classification decision-making,finally an epileptic EEG classification model was proposed based on MPE-ANN-SVM.The model was tested multiple times on 13 clinical datasets,obtained the average recognition rate,sensitivity and specificity were 94.01%,94.51%and 93.52%respectively.The classification indexes of this model are better than the classification results of artificial neural network models and the baseline model of support vector machine under the same basis,which can provide support for the research of epilepsy EEG detection algorithm.

关 键 词：癫痫 脑电信号 脑电检测 非线性动力学 机器学习 多尺度排列熵 人工神经网络 支持向量机 

分 类 号：R318[医药卫生—生物医学工程] TP391[医药卫生—基础医学]