摘要
自动癫痫检测对癫痫病发作的诊断及减轻医务人员繁杂的工作有着重大的意义。本研究提出一种基于多特征的长程颅内脑电癫痫检测的新算法。该算法首先对颅内脑电信号进行小波分解和半波处理,然后提取脑电信号的微分方差、相对能量和波动指数组成特征向量,利用贝叶斯原理求得待检信号特征向量的后验概率,通过阈值判断达到癫痫检测的目的。利用德国弗莱堡长程脑电数据进行实验,检测灵敏度为94.2%,特异性为95.6%,误检率为每小时1.16次。实验表明,该算法能够有效检测出长程颅内脑电中的癫痫信号,并具有较低的运算复杂度,有利于实时脑电检测。
The automatic seizure detection and classification are significant in both diagnosis of epilepsy and relieving heavy working load of doctors. In this paper we proposed a new seizure detection method based on multi-features of long-term intracranial EEG. After wavelet and half-wave decomposition, differeutial variance, relative energy and relative fluctuation index were used to characterize seizure activity as three features. Then the feature vector was fed to Bayesian formulation which was used as a classifier. A sensitivity of 94.2% , average specificity of 95.6 % and a false detection rate of 1.16 per hour were achieved with long-term intracranial EEG from Freiburg dataset. The experimental results indicated that this method is able to detect epileptic seizures effectively and its low computational complexity made it suitable for real-time seizure detection.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2013年第3期279-283,共5页
Chinese Journal of Biomedical Engineering
基金
国家科技支撑计划项目(2008BAI52B03)
山东省攻关计划项目(2010GSF10243)
山东大学自主创新基金(2012DX008)
关键词
颅内脑电
自动癫痫检测
微分方差
波动指数
intracranial EEG
automatic seizure detection
differeutial variance
fluctuation index
