基于平滑非线性能量算子划分的尖峰相关特征癫痫发作自动检测算法

Automatic seizure detection algorithm based on spike-related features of smoothed nonlinear energy operator division

针对癫痫发作自动检测算法多集中于时域、频域等传统特征,无法全面表征癫痫脑电信号的信息等问题,本文结合癫痫脑电图中异常波振幅和频率提高的现象,提出一种基于平滑非线性能量算子划分的尖峰相关特征癫痫发作自动检测算法。该算法使用传统的时域、频域特征,结合尖峰相关性特征对脑电信号进行刻画,使用有监督的机器学习分类器,测试癫痫发作自动检测的有效性和可靠性。本文将提出的方法在开源数据集CHBMIT上进行了评估,获得了96.52%的准确率、95.65%的敏感性和97.09%的特异性。实验结果表明,基于平滑非线性能量算子划分的尖峰相关特征,能够作为癫痫脑电信息的补充,提高癫痫发作检测的性能。

Most current seizure automatic detection algorithms focus on traditional features such as time domain and frequency domain,which cannot fully characterize the information of epileptic EEG signals.This paper proposes an automatic seizure detection algorithm based on spike correlation features divided by a smooth nonlinear energy operator,taking into account the phenomenon that the amplitude and frequency of abnormal waves in epileptic EEG will increase.The algorithm uses traditional time-domain and frequency-domain features,combined with spike correlation features to characterize the EEG signal,and uses supervised machine learning classifiers to test its effectiveness and reliability for automatic seizure detection.The research evaluates the proposed method on the open source dataset CHBMIT and obtains 96.52%on accuracy,95.65%on sensitivity and 97.09%on specificity.The experimental results show that the proposed spike-related features based on the smoothed nonlinear energy operator segmentation can be used as a complement to the epileptic EEG information to improve the performance of seizure detection.