1.College of Electronic and Optical Engineering & College of Microelectronics, Nanjing University of Posts and Telecommunications, Nanjing 210023, China 2.National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology, Nanjing University of Posts and Telecommunications, Nanjing 210023, China 3.Post Big Data Technology and Application Engineering Research Center of Jiangsu Province, Nanjing University of Posts and Telecommunications, Nanjing 210003, China 4.Post Industry Technology Research and Development Center of the State Posts Bureau (Internet of Things Technology), Nanjing University of Posts and Telecommunications, Nanjing 210003, China
Fund Project:Project supported by the National Natural Science Foundation of China (Grant Nos. 61972208, 61672299)
Received Date:14 June 2020
Accepted Date:26 August 2020
Available Online:12 December 2020
Published Online:05 January 2021
Abstract:Epilepsy is an extensive nervous system disease nowadays. Electroencephalogram (EEG) can capture the abnormal discharge of nerves in the brain duration of seizure and provide a non-invasive way to identify epileptogenic sites in the brain. In order to distinguish between focal epilepsy EEG signal and non-focal epilepsy EEG signal, in this paper we propose an automated epileptic EEG detection method based on the elastic variational mode decomposition (EVMD). The proposed EVMD algorithm is a method of analyzing the signals and also a processing method in time-frequency domain, in which the elastic net regression is used to reconstruct a constrained variational model in variational mode decomposition (VMD). Used in the VMD algorithm is the Tikhonov regularization that is also statistically called ridge regression as a solution of recovering the unknown signal and assessing the bandwidth of a mode, namely the variational equation constructed by VMD only has L2 norm. However, the ridge regression cannot select variables when the equation has multiple variables. Another regression method, called lasso regression, only has L1 norm and can select a more accurate model from multiple variables, but it has worse performance when variables have group effect or co-linearity. The elastic net regression has advantages of ridge regression and lasso regression, in other word, the variational equation constructed by EVMD has both L1 regularization item and L2 regularization item, so in this paper we propose the EVMD by elastic net regression. In addition, in this paper the EVMD is used to distinguish between focal epilepsy EEG signal and non-focal epilepsy EEG signal. Firstly, the original EEG signals are divided into several sub-signals where the test signals are divided into sub-signals with shorter durations by time series and a reasonable time overlap is kept between successive sub-signals. After that each sub-signal is decomposed into intrinsic mode functions by using the EVMD. Furthermore, the refined composite multiscale dispersion entropy (RCMDE) as feature is extracted from each intrinsic mode function where a Student’s t-test is used to assess the statistical differences between RCMDEs extracted from focal and non-focal EEG signals respectively. Finally, the support vector machine (SVM) is used to classify their features. For an epilepsy EEG signalspublic data set, the final experimental results show that the performance indices of accuracy, sensitivity, and specificity can reach 92.54%, 93.22% and 91.86% respectively. Keywords:elastic variational mode decomposition/ refined composite multiscale dispersion entropy/ epileptic electroencephalogram
表1从各VMF中提取的RCMDE特征p值 Table1.The p values of RCMDE computed from VMF.
指标
准确度/%
灵敏度/%
特异度/%
EMVD
92.54
93.22
91.86
VMD
89.49
87.56
88.12
表2EVMD与VMD实验结果对比 Table2.Comparison of experimental result between EVMD and VMD.
23.3.分类结果 -->
3.3.分类结果
选用数据集中全部3750对病灶性癫痫脑电数据段和3750对非病灶性癫痫脑电数据段, 按上述方法进行处理, 并将得到的每段脑电数据的RCMDE特征送入SVM进行特征分类, SVM选用线性核函数, 通过网格搜索法将惩罚参数设定为0.52, 并重复进行10次5折交叉验证实验, 采用准确度、灵敏度和特异度这三个指标对最终分类结果进行度量, 结果如图4所示. 由图4可知10次5折交叉验证实验的平均准确度、灵敏度和特异度分别可达92.54%, 93.22%, 91.86%. 图 4 10次5折交叉验证实验结果折线图 Figure4. The line chart of the results by 5-fold cross validation for 10 times.