基于k最近邻法的癫痫脑电信号研究  

Study on epileptic EEG signals based on k-nearest neighbor algorithm

作　　者：卢灿爱 姚文坡 乙万义[2] 白登选 王琼 戴加飞 王俊[2] LU Canai;YAO Wenpo;YI Wanyi;BAI Dengxuan;WANG Qiong;DAI Jiafei;WANG Jun(School of Communications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003;School of Geographic and Biologic Information,Nanjing University of Posts and Telecommunications,Nanjing 210023;Department of Neurology,Jinling Hospital,Medical School of Nanjing University,Nanjing 210008)

机构地区：[1]南京邮电大学通信与信息工程学院,南京210003 [2]南京邮电大学地理与生物信息学院,南京210023 [3]南京大学医学院金陵医院神经内科,南京210008

出　　处：《北京生物医学工程》2025年第1期55-60,67,共7页Beijing Biomedical Engineering

摘　　要：目的利用k最近邻法(k-nearest neighbor,KNN)计算符号序列部分互信息对癫痫脑电信号间的耦合关系进行分析,以期探究癫痫脑电信号与健康人的脑电信号耦合程度的差异,为癫痫脑电信号的研究提供借鉴方法。方法传统方法是通过计算变量间的概率分布密度求得部分互信息。本文是利用k最近邻法计算部分互信息,该算法对数据的需求量要求不高,并且算法的精度和效率比较高。首先将原始脑电信号序列符号化,符号化的目的就是把序列转变成符号序列,这可以有效降低噪声的影响;然后对符号序列进行编码处理;最后利用k最近邻算法计算其部分互信息来获得脑电信号的耦合关系。结果对于传统方法求部分互信息,在数据长度大于4000时,在枕区O1、O2求出的P值满足小于0.05。对于k最近邻方法求部分互信息,在数据长度大于2000时,在枕区O1、O2求出的P值满足小于0.05。相对于传统方法,k最近邻法可以利用较短数据长度区分出实验数据中的癫痫脑电信号和健康人的脑电信号。同时发现健康人的脑电信号耦合程度显著高于癫痫患者。结论k最近邻法求解符号化部分互信息可以有效分析癫痫脑电信号,并且算法的精度和效率比较高。Objective The k-nearest neighbor algorithm is used to calculate the partial mutual information of symbol sequence to analyze the coupling relationship between epileptic EEG signals,to explore the difference in the coupling degree between epileptic EEG signals and healthy people’s EEG signals,and to provide a reference method for the study of epileptic EEG signals.Methods The traditional method is to obtain partial mutual information by calculating the probability distribution density between variables.In this paper,the k-nearest neighbor method is used to calculate part of mutual information.The algorithm does not require much data,and the accuracy and efficiency of the algorithm are relatively high.First,the original epileptic EEG signal sequence is symbolized.The purpose of symbolization is to transform the sequence into a symbol sequence,which can effectively reduce the impact of noise.Then,the symbol sequence is coded.Finally,the k-nearest neighbor algorithm is used to calculate partial mutual information to obtain the coupling relationship of EEG signals.Results For traditional methods of obtaining partial mutual information,when the data length is greater than 4000,the P-values obtained in the pillow regions O1 and O2 are less than 0.05.For the k-nearest neighbor method to obtain partial mutual information,when the data length is greater than 2000,the P-values obtained in the pillow regions O1 and O2 are less than 0.05.Compared to traditional methods,the k-nearest neighbor method can distinguish between epileptic EEG signals and healthy EEG signals in experimental data using a shorter data length.At the same time,it is found that the coupling degree of EEG signals in healthy individuals is significantly higher than that in epilepsy patients.Conclusions The k-nearest neighbor method can effectively analyze epileptic EEG signals.The accuracy and efficiency of the algorithm are relatively high.

关 键 词：癫痫脑电信号 序列符号化 互信息 K最近邻算法 耦合关系 

分 类 号：R318.04[医药卫生—生物医学工程]