Epileptic seizure detection using EEG signals and extreme gradient boosting

The problem of automated seizure detection is treated using clinical electroencephalograms(EEG) and machine learning algorithms on the Temple University Hospital EEG Seizure Corpus(TUSZ).Performances on this complex data set are still not encountering expectations.The purpose of this work is to determine to what extent the use of larger amount of data can help to improve the performances.Two methods are explored:a standard partitioning on a recent and larger version of the TUSZ,and a leave-one-out approach used to increase the amount of data for the training set.XGBoost,a fast implementation of the gradient boosting classifier,is the ideal algorithm for these tasks.The performances obtained are in the range of what is reported until now in the literature with deep learning models.We give interpretation to our results by identifying the most relevant features and analyzing performances by seizure types.We show that generalized seizures tend to be far better predicted than focal ones.We also notice that some EEG channels and features are more important than others to distinguish seizure from background.

Editorial commentary on special issue of Advances in EEG Signal Processing and Machine Learning for Epileptic Seizure Detection and Prediction

This special issue of The Journal of Biomedical Research features novel studies on epileptic seizure detection and prediction based on advanced EEG signal processing and machine learning algorithms.The articles selected present important findings including new experimental results and theoretical studies.

Correlation between level of serum prolactin and epileptic discharges of electroencephalogram from 24 to 36 hours after epileptic onset

BACKGROUND： Researchers discovered that serum prolactin could rise following an epileptic seizure. The prolactin level might reach three times more than basic level within 30 minutes and decrease to the normal value 2 hours after the seizure occurred. The mechanism might result in an increase of serum prolactin concentrations with the activation of the hypothalamic-pituitary axis. OBJECTIVE：To probe into the correlation between changes of serum prolactin and incidence of epileptic discharges of electroencephalogram （EEG） at 24-36 hours after epileptic onset of patients with secondary epilepsy. DESIGN ： Clinical observational study SEI-FING： Department of Neurology, First Hospital affiliated to Soochow University PARTICIPANTS： A total of 21 patients with secondary epilepsy were selected from the Department of Neurological Emergency or Hospital Room of the First Hospital affiliated to Soochow University from November 2005 to April 2006. There were 14 males and 7 females aged from 25 to 72 years. All patients met International League Anti-epileptic （ILAE） criteria in 1981 for secondary generalized tonic clonic seizure through CT or MRI and previous EEG. All patients were consent. Primary diseases included cerebral trauma （3 cases）, tumor （2 cases）, stroke （7 cases） and intracranial infeion （9 cases）. METHODS ： Venous blood of all patients was collected at 24-36 hours after epileptic onset. Serum prolactin kit （Beckman Coulter, Inc in USA） was used to measure value of serum prolactin according to kit instruction. Then, value of serum prolactin was compared with the normal value （male： 2.64-13.13 mg/L; female： 3.34- 26.72 mg/L）; meanwhile, EEG equipment （American Nicolet Incorporation） was used in this study. MAIN OUTCOME MEASURES ： ① Abnormal rate of serum prolactin of patients with secondary epilepsy; ②Comparison between normal and abnormal level of serum prolactin and incidence of EEG epileptic discharge of patients with secondary epilepsy. RESULTS：All 21 patients with secondary epilepsy were involved in the final analysis. ① Results of serum prolactin level： Among 21 patients with of secondary epilepsy, 10 of them had normal serum prolactin and 11 had abnormal one, and the abnormal rate was 52% （11/21）. ② Detecting results of EEG： EEG results showed that 6 cases were normal and 15 were abnormal, and the abnormal rate was 71% （15/21）. The symptoms were sharp wave, spike wave or sharp slow wave, spike slow wave of epileptic discharges in 8 cases, which was accounted for 38%. ③ Correlation between abnormality of serum prolactin and EEG epileptic wave： Eleven cases had abnormal serum prolactin, and the incidence was 64% （7/11）, which was higher of epileptic wave than that of non-epileptic wave [36% （4/11）, P 〈 0.05]; however, 10 cases had normal serum prolactin, and the incidence was 10% （1/10）. Epileptic wave was lower than non-epileptic wave [90% （9/10）, P 〈 0.01]. CONCLUSION ： The level of serum prolactin of patients with secondary epilepsy is abnormally increased at 24- 36 hours after epileptic onset; in addition, incidence of epileptic discharge is also increased remarkably.

基于EEG和DE-CNN-GRU的情绪识别

近年,情绪识别研究已经不再局限于面部和语音识别,基于脑电等生理信号的情绪识别日趋火热.但由于特征信息提取不完整或者分类模型不适应等问题,使得情绪识别分类效果不佳.基于此,本文提出一种微分熵(DE)、卷积神经网络(CNN)和门控循环单元(GRU)结合的混合模型(DE-CNN-GRU)进行基于脑电的情绪识别研究.将预处理后的脑电信号分成5个频带,分别提取它们的DE特征作为初步特征,输入到CNN-GRU模型中进行深度特征提取,并结合Softmax进行分类.在SEED数据集上进行验证,该混合模型得到的平均准确率比单独使用CNN或GRU算法的平均准确率分别高出5.57%与13.82%.

Research on time-frequency cross mutual of motor imagination data based on multichannel EEG signal

At present,multi-channel electroencephalogram(EEG)signal acquisition equipment is used to collect motor imagery EEG data,and there is a problem with selecting multiple acquisition channels.Choosing too many channels will result in a large amount of calculation.Components irrelevant to the task will interfere with the required features,which is not conducive to the real-time processing of EEG data.Using too few channels will result in the loss of useful information and low robustness.A method of selecting data channels for motion imagination is proposed based on the time-frequency cross mutual information(TFCMI).This method determines the required data channels in a targeted manner,uses the common spatial pattern mode for feature extraction,and uses support vector ma-chine(SVM)for feature classification.An experiment is designed to collect motor imagery EEG da-ta with four experimenters and adds brain-computer interface(BCI)Competition IV public motor imagery experimental data to verify the method.The data demonstrates that compared with the meth-od of selecting too many or too few data channels,the time-frequency cross mutual information meth-od using motor imagery can improve the recognition accuracy and reduce the amount of calculation.

基于拓扑数据分析的驾驶疲劳EEG数据处理与优化分析研究

为提高驾驶疲劳脑电(EEG)数据处理与分析的准确性和鲁棒性,提出一种基于拓扑数据分析(TDA)的驾驶人疲劳脑电分析方法。首先利用汽车性能虚拟仿真平台开展驾驶实验,通过驾驶人状态反馈和面部特征视频,标记脑电数据,形成清醒和疲劳二分数据集。之后利用EEGLAB预处理数据,剔除噪声并保留0.3~30 Hz频带,直接从时域EEG数据中提取拓扑特征。此外还提取了经典频域特征α波能量和α/β用于对比分析。最后使用支持向量机进行分类。结果表明,基于持久同源(PH)的拓扑特征取得了高达88.7%的准确率和91.4%的召回率,与经典频域特征性能相当,且对脑电伪影的鲁棒性明显更好,在未剔除EEG伪影的情况下仍取得了87.4%的准确率和89.7%的召回率。综上所述,本文提出的用于驾驶疲劳脑电信号处理与分析的TDA方法抗干扰特性好、处理成本低、经济性高,有助于稳定、高效地处理驾驶人脑电数据并检测驾驶疲劳状态,具有较大的科学实际应用价值。

基于多频带路径签名特征的癫痫脑电图信号分类方法

基于脑电图(EEG)信号的癫痫自动检测对癫痫的临床诊断和治疗有很大的帮助。由于大部分癫痫识别算法忽略了EEG信号的时序关系,为此,文中提出了一种基于多频带路径签名特征的癫痫EEG信号分类方法。此方法首先将EEG信号分解成5个不同频段的频带信号,再通过路径签名算法进行特征提取,然后采用局部主成分分析去除特征相关性并进行特征融合,最后将融合特征送入集成分类器中进行预测分类。由于路径签名可以更深入地挖掘EEG信号的相关关系,结合局部主成分分析后,文中方法可以获取更有鉴别性的癫痫分类特征。分别在时长超过2 000 s癫痫发作片段的本地医院私有数据集和开源的CHB-MIT癫痫数据集上,选用10折交叉进行实验验证,结果表明:在私有数据集上,文中方法的平均分类准确率达到97.25%,比经典的基于经验模态分解(EMD)的方法提高了3.44个百分点,比最新的基于长短期记忆网络(LSTM)+卷积神经网络(CNN)的方法提高了1.35个百分点;在CHB-MIT数据集上,文中方法的平均分类准确率达到98.11%,比经典的基于EMD的方法提高了5.20个百分点,比最新的基于LSTM+CNN的方法提高了2.64个百分点;在两个数据集上文中方法的分类准确率均优于其他对比方法。

基于脑电信号特征的高铁调度员疲劳状态识别

为增强铁路行车的稳定性与安全性,有效识别调度员的疲劳状态对行车组织的影响,基于脑电(EEG)信号特征,提出一种调度员疲劳状态识别方法,根据作业时间段划分调度员的疲劳状态,设计高铁调度模拟试验获取脑电信号数据,通过小波级数展开和傅里叶变换提取高铁调度被试的3种脑电波频域幅值作为特征值,结合调度员作业特征和脑电信号特征,验证疲劳状态的划分结果,通过Python语言环境搭建ResNet18+SoftMax和MobileNet V2+SoftMax这2种模型,基于深度学习方法,将输入特征转换为三维立体矩形模型,并优化调整权重,获得最优模型,从而判断高铁调度员的疲劳状态。研究结果表明:ResNet18+SoftMax和MobileNet V2+SoftMax神经网络模型对高铁调度试验参与人员的疲劳状态识别准确率分别为92.78%和99.17%;相较于支持向量机(SVM)模型,这2种模型可提升清醒状态和疲劳状态的识别精度,并降低运算时间,其中,MobileNet V2+SoftMax模型的识别准确率和运行速度最优。以MobileNet V2+SoftMax模型原理为内核,可以更快速准确地识别高铁调度员在长时间作业条件下的潜在疲劳风险。

基于脑电节律能量与模糊熵的VR诱发晕动症水平检测研究

晕动症一直是影响虚拟现实用户体验及限制虚拟现实行业发展的一个关键因素。为解决这一问题,本文研究了虚拟现实晕动症对大脑神经活动的影响,并利用脑电特征对晕动症水平进行检测。为得到可度量眩晕水平的特征,记录受试者在体验眩晕测试场景前及过程中的脑电信号,计算节律能量和模糊熵,并利用统计分析进行特征选择,最后分类验证该特征的有效性。结果表明,受试者产生晕动症时,CP4和Oz的θ、α频段能量及C4的β、γ频段能量显著降低(p<0.01);在模糊熵方面,δ频段有FC4、Cz模糊熵值显著升高(p<0.0001),β频段有O1模糊熵值显著降低(p<0.0001)。对比线性判别分析(Linear discriminant analysis,LDA)、逻辑回归(Logistic regression,LR)和支持向量机(Support vector machine,SVM),K最近邻(K-nearest neighbor,KNN)算法的分类效果较好,它在节律能量和模糊熵上的分类准确率分别为89%和91%。本研究表明脑电节律能量及模糊熵有望成为晕动症水平检测的有效指标,为研究虚拟现实晕动症成因及缓解方案提供客观依据。

基于EEG模糊相似性的癫痫发作预测

本研究提出基于EEG序列模糊相似性指数方法预测癫痫发作。首先,结合复自相关法和Cao法对EEG序列进行了相空间重构;然后,计算相关积分时用Gaussian函数代替Heavyside函数,克服了Heavyside函数的刚性边界问题,使得计算相似性指数更加准确和可靠;最后,分析大鼠癫痫EEG信号,检测癫痫发作前期状态。分析结果表明模糊相似性指数方法能够比动态相似性指数方法获得更长的预测时间和更低的错误预测率。

引入迁移学习的癫痫EEG信号自适应识别

在脑电图(EEG)信号识别中,EEG信号的采样环境、病人状态的多样性导致分类器训练所用的源域与分类器测试所用的目标域不匹配,分类器在目标域上表现不佳。为此,引入邻域适应策略,提出一种基于子空间相似度的改进主成分分析特征提取方法(SSM-PCA),在选择主成分时,考虑源域和目标域数据的几何和统计特性,并结合迁移学习分类器大间隔投射迁移支持向量机(LMPROJ),给出以SSM-PCA为基础的LMPROJ分类识别方法。实验结果表明,与结合PCA特征抽取技术和K近邻分类器实现的识别方法相比,该方法在识别正确率方面得到较大提升。

基于多模式EEG的脑-机接口虚拟键鼠系统设计

现有的脑-机接口系统大都只基于单模式的脑电特征,系统能实现的功能非常有限,从而制约了脑-机接口系统的应用。采用基于多种模式脑电信号(electroencephalogram,EEG)的脑-机接口技术来实现虚拟键鼠系统,使得被试可以利用自身的脑电信号控制鼠标和键盘的操作。研究了脑-机接口中常用的3种脑电信号,分别是P300波、alpha波以及稳态视觉诱发电位(steady state visual evoked potential,SSVEP),通过设计实验成功的诱发出了被试相应的特征脑电信号。利用SSVEP的脑电特征设计6频率LED闪烁刺激的虚拟鼠标系统,实现控制鼠标光标移动、单击左键和单击右键的任务;利用P300波的脑电特征设计6×6的字符矩阵虚拟键盘系统,实现字符输入的任务;利用被试自主闭眼增强alpha波的脑电特征,实现鼠标和键盘应用切换的任务。研究了适宜这3种脑电特征的最佳测量电极组合及模式识别算法,使得对3种脑电信号的识别正确率均达到了85%以上。测试结果显示,文中设计的基于多模式EEG的脑-机接口虚拟键鼠系统能有效地实现鼠标控制以及键盘输入的任务。

基于微状态的抑郁症静息态脑电信号分析

抑郁症(MDD)患者存在认知功能障碍,但其瞬时神经异常活动尚未研究清楚,对此采用脑电(EEG)微状态方法对抑郁症患者的脑电数据进行研究。比较22名抑郁症患者和25名正常人的128导闭眼脑电数据微状态特征,进行差异性分析并探索与量表得分之间的相关性。结果发现,相对于健康对照组,抑郁症患者微状态C的出现次数和涵盖比更高,且与其他微状态之间的转换概率较高,而其微状态D的平均持续时间较低,且与微状态B之间的转换次数减少。此外,微状态C和微状态D与抑郁量表和焦虑量表均呈显著相关性,表明基于脑电微状态方法可以捕捉到抑郁症患者异常大脑动态特性,为抑郁症临床早期诊治提供客观参考。

基于SDAE与RELM的EEG情感识别方法

针对情感识别中堆叠式自动编码器存在反向传播方法收敛速度慢和容易陷入局部最优的问题,提出一种基于堆叠式降噪自动编码器(SDAE)和正则化极限学习机(RELM)的情感状态识别方法。从脑电信号的时域、频域和时频域中提取表征情感状态的初始特征,使用SDAE进行无监督特征学习,提取初始特征的高层抽象表示。在网络的回归层,使用RELM进行情感分类。在DEAP数据集上的实验结果表明,与SDAE以及DT、KNN等传统基于机器学习的方法相比,该方法在实时性、准确性和泛化性能等方面均有明显提升。

A new approach for epileptic seizure detection: sample entropy based feature extraction and extreme learning machine

The electroencephalogram (EEG) signal plays a key role in the diagnosis of epilepsy. Substantial data is generated by the EEG recordings of ambulatory recording systems, and detection of epileptic activity requires a time-consuming analysis of the complete length of the EEG time series data by a neurology expert. A variety of automatic epilepsy detection systems have been developed during the last ten years. In this paper, we investigate the potential of a recently-proposed statistical measure parameter regarded as Sample Entropy (SampEn), as a method of feature extraction to the task of classifying three different kinds of EEG signals (normal, interictal and ictal) and detecting epileptic seizures. It is known that the value of the SampEn falls suddenly during an epileptic seizure and this fact is utilized in the proposed diagnosis system. Two different kinds of classification models, back-propagation neural network (BPNN) and the recently-developed extreme learning machine (ELM) are tested in this study. Results show that the proposed automatic epilepsy detection system which uses sample entropy (SampEn) as the only input feature, together with extreme learning machine (ELM) classification model, not only achieves high classification accuracy (95.67%) but also very fast speed.

具有多层次优化能力的EEG生成模型

基于生成对抗网络(generative adversarial networks,GAN)的脑电信号(electroencephalogram,EEG)生成技术存在生成样本特征单一、幅值差异过大以及拟合速度慢等问题,其质量难以满足深度学习模型训练和优化的要求.因此,本文通过对WGAN-GP的优化,使其更适应脑电信号生成,从而解决以上问题.具体而言:(1)在WGAN-GP网络的框架的基础上,通过将长短期记忆网络(long short-term memory,LSTM)代替卷积神经网络(convolutional neural networks,CNN),以保证时间相关特征的完整性,从而解决脑电生成特征单一的问题;(2)将标准化处理后的真实脑电信号输入至判别器,以解决幅值差异过大问题;(3)将脑电噪声部分作为先验知识输入至网络生成器,以提高生成模型的拟合速度.本文分别通过sliced Wasserstein distance(SWD)、mode score(MS)以及EEGNet对生成模型做多层次定量评估.与目前已有生成网络WGAN-GP相比较,基于本模型的生成数据更为接近真实数据.

Study on Singularity of Chaotic Signal Based on Wavelet Transform

Based on the variations of wavelet transform modulus maxima at multi-scales, the singularity of chaotic signals are studied, and the singularity of these signals are measured by the Lipschitz exponent.In the meantime, a nonlinear method is proposed based on the higher order statistics, on the other aspect, which characterizes the higher order singular spectrum (HOSS) of chaotic signals. All computations are done with Lorenz attractor, Rossler attractor and EEG(electroencephalogram) time series and the comparisions among these results are made. The experimental results show that the Lipschitz exponents and the higher order singular spectra of these signals are significantly different from each other, which indicates these methods are effective for studing the singularity of chaotic signals.

一种针对EEG伪迹的多窗卷积压制算法

脑电信号(EEG)会受到伪迹的严重干扰,可以利用独立成分分析分离出观察信号的源成分,并根据源成分的时、空等特征区分神经信号成分和伪迹成分,随之删除伪迹成分可恢复较干净的观察信号。但在删除整个伪迹成分的同时势必会丢失一些有用的神经信号,这会造成神经信号泄露。为此,提出一种降低伪迹噪声和减少神经信号泄露的均衡方法,在盲源信号分离的基础上,利用局部化的时间特征二分类伪迹成分和神经信号成分,以多种窗函数组合优化并卷积压制局部伪迹成分。通过对观察信号计算信号伪迹比(SAR),8导、32导非校正脑电信号SAR分别为-19.765dB、-19.016dB,校正后8导、32导脑电信号SAR分别提高到2.832dB、2.743dB。结果表明该方法可以消除眼电等伪迹且降低神经信号泄露的影响。

Evolutionary Algorithsm with Machine Learning Based Epileptic Seizure Detection Model

Machine learning (ML) becomes a familiar topic among decisionmakers in several domains, particularly healthcare. Effective design of MLmodels assists to detect and classify the occurrence of diseases using healthcaredata. Besides, the parameter tuning of the ML models is also essentialto accomplish effective classification results. This article develops a novelred colobuses monkey optimization with kernel extreme learning machine(RCMO-KELM) technique for epileptic seizure detection and classification.The proposed RCMO-KELM technique initially extracts the chaotic, time,and frequency domain features in the actual EEG signals. In addition, the minmax normalization approach is employed for the pre-processing of the EEGsignals. Moreover, KELM model is used for the detection and classificationof epileptic seizures utilizing EEG signal. Furthermore, the RCMO techniquewas utilized for the optimal parameter tuning of the KELM technique insuch a way that the overall detection outcomes can be considerably enhanced.The experimental result analysis of the RCMO-KELM technique has beenexamined using benchmark dataset and the results are inspected under severalaspects. The comparative result analysis reported the better outcomes of theRCMO-KELM technique over the recent approaches with the accuy of 0.956.