一种基于改进经验模态分解的癫痫脑电识别新方法

在临床癫痫诊断过程中，为了提高癫痫脑电的识别率，能在癫痫发作前期就预测到癫痫疾病，其特征波的提取至关重要。针对这一问题，提出将平行延拓与镜像延拓相结合来改进EMD算法。首先，使用平行延拓的方法，在原始脑电信号的左、右端点处分别预测出一个极值；然后，使用基于镜像延拓的EMD方法，对信号进行镜像延拓，以避免经验模态分解过程中的端点效应；最后，采用支持向量机进行信号的分类识别。算法验证数据取自德国伯恩大学癫痫研究中心的脑电数据库，其中50例是正常脑电信号、50例是癫痫发作间期的脑电信号。实验研究表明：该方法对总测试脑电信号的识别率达到94%。其中，正常脑电信号和癫痫脑电信号的独立识别率均为94%，比传统EMD算法处理后的脑电识别率提高了5%，可见该方法可以有效地预测癫痫脑电。

A New Method of Epileptic EEG Identification Based on Improved EMD

In order to improve the recognition rate of epileptic EEG and to predict epileptic disease in the early stage of epileptic seizures, the characteristic wave extraction is very important in the process of clinical diagnosis of epilepsy. To solve this problem, a method that combined parallel extension with mirror extension to improve EMD algorithm was proposed. Firstly, extreme values were predicted respectively in the left and right endpoints of the original EEG using the parallel extension method. Then, the EMD method based on the mirror extension was used in order to avoid the end effect in the process of EMD. Finally, the SVM classifier was used for signal classification and recognition. The algorithm validation data were from EEG database of the Epilepsy Research Center, University of Bonn, Germany (50 cases normal EEG signals and 50 cases epileptic EEG signals). The result shows that the recognition rate (the sum of normal EEG and epileptic EEG) of total test EEG signals by this method can reach 94%, and the recognition rates of normal EEG and epileptic EEG are both 94%. This result is 5% higher than the recognition rate of EEG processed by the traditional EMD algorithm. Therefore the method can predict epileptic EEG effectively.