基于支持向量机的癫痫脑电信号模式识别研究

The Recognition Methodology Study of Epileptic EEGs Based on Support Vector Machine

癫痫患者脑电(EEG)信号包含了癫痫发作过程中丰富的生理病理信息,EEG活动的动态变化为癫痫的自动检测系统的研发提供了依据和可能。本文从检测癫痫EEG信号的非线性动力学特征入手,提取EEG信号和小波分解后的各脑电特征波的非线性动力学特征值,作为输入向量构建支持向量机(SVM)分类器。结果表明,基于非线性动力学指标的SVM分类器对癫痫发作间期EEG和发作期EEG的分类准确率可达90%以上,SVM在癫痫EEG信号检测中作为非线性分类器具有较好的泛化能力。

EEG recordings contain valuable physiological and pathological information in the process of seizure. The dynamic changes of brain electrical activity provide foundation and possibility for research and development of auto- matic detection system about epilepsy. In this paper, a nonlinear dynamic method is presented for analysis of the nonlinear dynamic characteristics of EEGs and delta, theta, alpha, and beta sub-bands of EEGs based on wavelet transform. The extracted feature is used as the input vector of a support vector machine （SVM） to construct classifi- ers. The results showed that the classification accuracy of SVM classifier based on nonlinear dynamic characteristics to classify the EEG into interietal EEGs and ietal EEGs reached 90% or higher. The support vector machine has good generalization in detecting the epilepsy EEG signals as a nonlinear classifier.