自回归模型和隐马尔可夫模型在癫痫脑电识别中的应用

摘要

目的 研究自回归(autoregressive model,AR)模型和隐马尔可夫模型(hidden Markov model,HMM)在癫痫脑电(electroencephalogram,EEG)识别中的应用,以期减轻医生工作量,减少人工识别主观因素的影响.方法 使用基于联合信息准则(combined information criterion,CIC)的最佳阶数AR模型对脑电信号进行特征提取,连续密度隐马尔可夫模型(continuous density hidden Markov model,CD-HMM)作为正常脑电和癫痫脑电的分类工具,对南京军区总医院的临床脑电数据(8组采样频率为512 Hz的16导正常、癫痫脑电信号)进行分析和识别.实验时对每一例样本选取T3、T4、FP1、FP2、C3、C4六个导联的数据.使用训练集中的15段样本进行HMM建模,剩下35段用作测试.结果 癫痫脑电的识别率可达90%.结论 AR模型结合HMM建模的方法对正常脑电信号和癫痫脑电的识别率较高,在脑-机接口设备的开发中有一定的应用前景.%Objective To study the application of hidden Markov model(HMM) and autoregressive(AR) model in detection of epileptic electroencephalogram( EEG). The automated epileptic EEG recognition method can reduce the workload of doctors and reduce the influence of subjective factors of artificial identification. Methods This paper used the optimal order AR model which is based on combined information criterion ( CIC ) criterion to do the EEG feature extraction,and used CD-HMM as the classification tool. The clinical EEG data of Nanjing General Hospital of Nanjing Military Command ( 8 groups of 16 leads normal and epileptic EEG signals with a sampling frequency of 512 Hz ) were analyzed and identified. In the experiment,the six lead ( T3,T4,FP1,FP2,C3,C4) data of each sample were selected. HMM modeling was performed using 15 samples of training concentration,and the remaining 35 sections were used for testing. Results The recognition rate of epilepsy EEG reached 92.8%. Conclusions The recognition rate is higher under the optimal order AR model and CD-HMM, which have certain application prospect in the development of brain-machine interface equipment.