单导癫痫脑电模糊特征提取的支持向量机发作预测

Epileptic seizure prediction from single-channel scalp EEG using support vector machine based on fuzzy feature extracted with empirical mode decomposition

为了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,提出了一种基于经验模式分解模糊特征提取的支持向量机二分类预测模型。对7名受试者发作前的各导脑电信号进行经验模式分解,提取各本征模式函数分量的模糊特征向量,将其作为支持向量机的输入进行分类.研究表明,系统预测准确度与导联选择有关,但与支持向量机所用的核函数关系不大。最佳导联选择条件下,所提出的方法在预测癫痫发作时的表现为:准确度为82.8～87.1%,特异性为86.7～91.9%,敏感性为74.6～81.3%范围。

To find a new approach to recognize and predict succedent epileptic seizures from single-channel electroencephalogram （EEG） analysis, a binary classification model of support vector machine （SVM） based on fuzzy feature extracted with empirical mode decomposition （EMD） is presented. Eight channels of EEG from each patient of seven consenting patients with epilepsy were monitored in Epilepsy Center of Xijing Hospital. The raw EEGs are decomposed using EMD algorithm, the intrinsic mode functions （IMFs） are extracted and converted into fuzzy feature vectors, which are taken as the input vectors of a trained SVM for classification, the output of the SVM will be the prediction results. Testing results show that the prediction accuracy observably depends on the selection of EEG leads, but is almost not influenced by the change of SVM kernel functions. The performance of the proposed method in predicting seizures is： sensitivity 74.6%-81.3%, specificity 86.7%-91.9% and accuracy 82.8-87.1%, if a most appropriate EEG lead is employed.