摘要
癫痫是一种常见的大脑神经紊乱疾病,癫痫性发作主要由大脑中反常的神经元的超同步放电引起。为了更好地完成癫痫性发作的自动检测,文中提出了一种新的癫痫脑电融合特征提取方法。一方面,在基于Hjorth参数的振幅移动性与振幅复杂度的基础上,结合Hilbert变化提出了一种新的频率移动性与频率复杂度,然后将他们合成定义为改进的Hjorth参数特征;另一方面,结合二阶差分提出了一种改进的二阶差分样本熵。最后将改进的Hjorth参数特征与二阶差分样本熵一起作为融合特征放入超限学习机(ELM)中进行分类。数值实验结果表明,文中所提出的融合特征与ELM结合的癫痫性发作的自动检测方法与已有方法相比,检测性能有了很大提高,准确率可达到97.42%。
Epilepsy is one of the most common neurological disorders. It is characterized by recurrent epileptic seizures, which are caused by hypersynchronous discharges of an excessive group of cells in the brain. In this paper, a novel fusion feature extraction method is proposed for realizing the automatic seizure detection using epileptic EEGs successfully. On one hand, the frequency mobility and frequency complexity are defined. And further, combining with the Hjorth parameters, an improved Hjorth-parameters-based feature is designed. On the other hand, the second-order differential sample entropy is presented, which is based on the sample entro- py and second-order differential method. Then the fusion feature, which combing the improved Hjorth-parame- ters-based feature and the second-order differential sample entropy, are fed into extreme learning machine （ELM） to complete the epileptic seizure detection. Experimental results show that the proposed seizure detec- tion method achieves the detection accuracy at 97.42%, whose performance is greatly improved compared with some other existing methods.
出处
《西北大学学报（自然科学版）》
CAS
CSCD
北大核心
2016年第6期801-808,共8页
Journal of Northwest University（Natural Science Edition）
基金
国家自然科学基金资助项目(61473223)
陕西省自然科学基础研究计划基金资助项目(2014JM1016)
关键词
癫痫脑电
Hjorth参数
差分
样本熵
超限学习机(ELM)
epileptic electroencephalogram（EEG）
Hjorth parameters
differential
sample entropy
extreme learning machine （ELM）
