基于单导脑电经验模式分解的癫痫发作预测

Predicting epileptic seizure by empirical mode decomposition and complexity analysis of single-channel scalp electroencephalogram

目的寻找一种通过单通道脑电分析实现癫痫发作预报的新方法。方法从西京医院癫痫中心的临床病例中选择7名受试者记录癫痫发作前后8个通道的脑电,对发作前的各导脑电信号进行经验模式分解,提取分解后各分量的复杂性测度,将其作为一个4层（7-6-2-1）神经网络的输入进行非线性分类,神经网络的训练采用去一循环法（leave one out）。结果研究表明,所提出的方法在预报癫痫发作时的表现为：根据所用脑电导联不同,准确度为71.7%-78.3%,特异性为71.4%-88.1%,敏感性为50%-77.8%范围。另外,系统运算速度足够快,适合临床实时检测需要。结论所提出的方法在预测癫痫发作时有一定优势,但进一步的研究仍然是必要的。

Aim To find a new approach to recognize and predict successful epileptic seizures by using single-channel electroencephalogram（EEG） analysis.Methods Eight channels of EEG from each patient of the seven consenting patients with generalized epilepsy were collected in Epilepsy Center of Xijing Hospital.The raw EEGs were decomposed by the algorithm of empirical mode decomposition（EMD）,the complexity measures were extracted from the seven components of EMD,and then a four layer（7-6-2-1） artificial neural network（ANN） was employed for prediction,using the extracted measures as inputs to the ANN.Training and testing the ANN used the ′leave one out′strategy.Results The performance obtained for the proposed scheme in predicting seizures is： sensitivity 50%～77.8%,specificity 71.4%～88.1% and accuracy 71.7%～78.3%,depending on the different EEG leads.This method is also computationally fast and can be used to monitor epilepsy for real-time clinical application.Conclusion The results show the proposed system has a certain advantage in predicting seizure.Additional studies need to be carried out on a wider patient population to further evaluate and improve the design.