基于TWQT和熵特征的癫痫脑电快速检测

Rapid detection of epileptic EEG based on TWQT and entropy

该文提出一种基于可调Q因子小波变换的熵特征提取方法用于癫痫脑电的快速检测。首先,利用TWQT算法对脑电信号进行分解,计算分解后得到的小波子带的中心频率;随后,选择对应癫痫发作时异常波段频率范围的子带进行小波重构,并对重构信号提取排序熵和规则性指数等熵特征;最后,选择包括支持向量机、随机森林、极端梯度提升在内的多种分类器进行分类实验。实验中采用德国波恩大学癫痫研究中心的公开数据集和UPenn and Mayo Clinic′s Seizure Detection Challenge数据集进行验证,所提出的方法针对小样本数据集可以达到99.3%的准确率,100%的灵敏度以及98.6%的特异度,体现了该算法的有效性。

In this paper,an entropy feature extraction method based on tunable Q factor wavelet transform is proposed.First,using TWQT algorithm to decompose EEG signals,calculate the center frequency of each wavelet subband,then choose corresponding sub-bands for reconstructing and extract reconstructed signals′permutation entropy and order index as the characteris-tic.In the end,choose a variety of classifieiers including support vector machine(SVM),random forest and XGBoost to classify.The proposed classification method is applied to the detection of epileptic EEG signals,and verified by the open dataset of the Epilepsy Research Center of Bonn University in Germany and UPenn and Mayo Clinic′s Seizure Detection Challenge dataset.The accuracy is 99.3%,sensitivity is 100%and specificity is 98.6%,which shows the effectiveness of the algorithm.