基于多重分形去趋势波动分析的脑电信号特征提取及分类方法  

Feature extraction and classification of electroencephalogram signal based on multifractal detrended fluctuation analysis

作　　者：陈敬凯 孟雪[1] 王常青 钟亚鼎[2] CHEN Jingkai;MENG Xue;WANG Changqing;ZHONG Yading(School of Biomedical Engineering,Anhui Medical University,Hefei 230032,China;Department of Radiology,the First Affiliated Hospital of Anhui Medical University,Hefei 230032,China)

机构地区：[1]安徽医科大学生物医学工程学院,安徽合肥230032 [2]安徽医科大学第一附属医院放射科,安徽合肥230032

出　　处：《中国医学物理学杂志》2021年第11期1387-1391,共5页Chinese Journal of Medical Physics

基　　金：国家自然科学基金(62001005);安徽省高校自然科学研究项目(KJ2017A209);安徽省自然科学基金(2008085QH425);安徽医科大学科研基金(XJ201811)。

摘　　要：目的:针对脑电信号普遍存在的数据维度高、难以预测的问题,提出一种多重分形去趋势波动分析特征提取方法与长短时记忆网络(LSTM)相结合的脑电信号分类方法。方法:首先对信号样本进行多重分形去趋势波动分析计算得到脑电信号样本的多重分形谱,计算广义Hurst指数hq和广义维数Dq之间的函数关系;然后对多重分形谱进行分析,找出最具代表性的坐标值作为信号的特征向量;最后将其用于LSTM进行训练和分类测试。实验采用波恩大学采集的经过处理的癫痫脑电数据集。结果:当训练样本占总体样本比例超过10%之后,LSTM分类器的测试准确率均稳定在98%以上;当占比超过80%时LSTM分类器的测试准确率达到了100%;即使训练样本较少时也有95%之上的准确率。结论:该算法有良好的准确率和稳定性。Objective To propose a electroencephalogram(EEG)signal classification method based on the combination of feature extraction by multifractal detrended fluctuation analysis(MF-DFA)and long short-term memory network(LSTM)for solving the problems existing in EEG signal such as high data dimensionality and difficulty in prediction.Methods The multifractal spectrum of the EEG signal samples was firstly obtained by MF-DFA,and the functional relationship between the generalized Hurst exponent hq and the generalized dimensionality Dq was calculated.Then the multifractal spectrum was analyzed to find the most representative coordinate value as the signal eigenvector.Finally,the obtained signal eigenvector was used for LSTM training and classification test.The experiment was carried out on a processed epileptic EEG data set collected by University of Bonn.Results When the training samples accounted for more than 10%of the total samples,the test accuracy of LSTM classifiers stabilized at 98%and above;and when the proportion was more than 80%,the test accuracy of LSTM classifier reached 100%.Even with a small number of training samples,the accuracy was higher than 95%.Conclusion The proposed algorithm has good accuracy and stability.

关 键 词：脑电信号 多重分形去趋势波动 长短时记忆网络 特征提取 信号分类 

分 类 号：R318[医药卫生—生物医学工程] TP301.6[医药卫生—基础医学]