基于时延嵌入式隐马尔科夫模型的癫痫脑电分类算法

EEG classification algorithm for epilepsy basedon time-delayed embedded hidden Markov model

癫痫脑电的分类识别能够为癫痫的预警和病程的发展监测提供强有力的技术支持。传统的癫痫脑电分类识别方法需要从较长的时间序列中提取特征,难以刻画大脑的瞬态变化,检测低效且耗时,降低了癫痫预警的有效性。针对上述问题,提出一种基于隐马尔科夫模型的癫痫脑电分类算法。该方法通过时延嵌入式隐马尔科夫模型(time-delay embedded hidden Markov model,TDE-HMM)对脑电进行状态估计,并提取状态序列中的状态切换特征,通过多层感知机(multiple layer perceptron,MLP)实现对不同癫痫发作阶段脑电的有效辨识。实验结果表明,相较于小波变换、微分熵等传统特征,所提方法准确率高,能够有效刻画癫痫不同阶段的大脑状态变化,为癫痫脑电的分类识别和状态分析提供了新的备选方案。

Electroencephalogram(EEG)classification for epilepsy can offer powerful technical assistance for both its early warning and progression monitoring.However,traditional recognition methods for epilepsy EEG need to extract features from long-term time series,which cannot characterize the transient changes of brain and result in lower efficiency for epilepsy recognition and higher time consumption.These shortages further restrict the effectiveness of early warning for epilepsy.To address these problems,we proposed a novel epilepsy classification method based on hidden Markov model(HMM),which adopted the time-delay embedded HMM(TDE-HMM)to extract features of state transformation from estimated state series and utilized multiple layer perceptron(MLP)to further identify different seizure stages.The experimental results proved that compared with discrete wavelet transformation,power spectral density and differential entropy,our proposed method holds higher classification and capability of characterizing the state transformations of different seizure stages,which offers a novel alternative for the epilepsy classification and state analysis.