基于离散S变换和排列熵的癫痫脑电识别

发布时间：2019-05-06 23:37

【摘要】：脑电图(EEG)分析对癫痫疾病的诊断具有重要的参考价值,对癫痫脑电信号的自动分类可以及时对患者的情况作出判断,在临床上有很重要的意义。为解决脑电信号采用单一特征识别率不高的问题,同时也为避免小波基函数的选取对分类结果的影响,本文提出了一种基于S变换和排列熵(PE)的癫痫脑电信号自动判别方法,首先将原始脑电信号进行离散S变换,再对变换后脑电信号各节律的系数分别求其波动指数,并与脑电信号的排列熵值共同组成特征向量送入Real Ada Boost分类器进行癫痫各时期的判别。本研究采用德国波恩大学癫痫研究中心数据库,对正常人清醒睁眼,癫痫患者发病间歇期致痫灶内及发作期3组脑电信号数据进行方法有效性检验。研究结果表明,各节律的波动指数可有效表征正常、癫痫发作间期和癫痫发作期脑电信号,且多种特征的识别率明显优于单一特征,平均识别率可达到98.13%,相比于仅提取时频特征或非线性特征,识别率分别提高了1.2%和8.1%以上,优于文献中报道的多种方法。因此,本方法在癫痫疾病的诊断方面有较好的应用前景。
[Abstract]:Electroencephalogram (EEG) (EEG) analysis has important reference value in the diagnosis of epileptic diseases. Automatic classification of EEG signals can be used to judge the patient's condition in time, and it is of great significance in clinical practice. In order to solve the problem of low recognition rate of EEG signals by single feature, and to avoid the influence of the selection of wavelet basis function on the classification results. In this paper, an automatic discrimination method for epileptic EEG signals based on S transform and permutation entropy (PE) is proposed. Firstly, the original EEG signals are transformed by discrete S transform, and then the coefficients of each rhythm of the transformed EEG signals are calculated. Together with the permutation entropy of EEG signals, the eigenvector is sent into the Real Ada Boost classifier to distinguish the epileptic phases. Using the database of epilepsy research center of Bonn University in Germany, the validity of three groups of EEG data in the epileptic foci and seizure phases of normal persons and epileptic patients during the intermittent period of onset of epilepsy was tested. The results show that the fluctuation index of each rhythm can effectively represent normal EEG signals in interictal and epileptic periods, and the recognition rate of multiple features is significantly better than that of a single feature, with an average recognition rate of 98.13%. Compared with extracting only time-frequency features or nonlinear features, the recognition rate is increased by 1.2% and 8.1% respectively, which is better than the methods reported in the literature. Therefore, this method has a good prospect in the diagnosis of epilepsy.
【作者单位】：山东大学信息科学与工程学院;山东大学第二医院;
【基金】：国家自然科学基金资助项目(61571274) 山东省自然科学杰出青年科学基金资助项目(201614) 山东大学青年学者未来计划资助项目(2015WLJH39)
【分类号】：R742.1;TN911.6

【相似文献】