注意力残差网络结合LSTM的EEG情绪识别研究

基于脑电信号的情感识别已成为情感计算和人机交互领域的一个重要挑战。由于脑电信号中具有时间、空间、频率维度信息,采用结合注意力残差网络与长短时记忆网络的混合网络模型(ECA-ResNet-LSTM)对脑电信号进行特征提取与识别。首先,提取时域分段后脑电信号不同频带微分熵特征,将从不同通道中提取出的微分熵特征转化为四维特征矩阵;然后通过注意力残差网络(ECA-ResNet)提取脑电信号中空间与频率信息,并引入注意力机制重新分配更相关频带信息的权重,长短时记忆网络(LSTM)从ECA-ResNet的输出中提取时间相关信息。实验结果表明:在DEAP数据集唤醒维和效价维二分类准确率分别达到了97.15%和96.13%,唤醒-效价维四分类准确率达到了95.96%,SEED数据集积极-中性-消极三分类准确率达到96.64%,相比现有主流情感识别模型取得了显著提升。

Detection of EEG signals in normal and epileptic seizures with multiscale multifractal analysis approach via weighted horizontal visibility graph

Electroencephalogram(EEG) signals contain important information about the regulation of brain system. Thus, automatic detection of epilepsy by analyzing the characteristics obtained from EEG signals has important research implications in the field of clinical medicine. In this paper, the horizontal visibility graph(HVG) algorithm is used to map multifractal EEG signals into complex networks. Then, we study the structure of the networks and explore the nonlinear dynamics properties of the EEG signals inherited from these networks. In order to better describe complex brain behaviors, we use the angle between two connected nodes as the edge weight of the network and construct the weighted horizontal visibility graph(WHVG). In our studies, fractality and multifractality of WHVG are innovatively used to analyze the structure of related networks. However, these methods only analyze the reconstructed dynamical system in general characterizations,they are not sufficient to describe the complex behavior and cannot provide a comprehensive picture of the system. To this effect, we propose an improved multiscale multifractal analysis(MMA) for network, which extends the description of the network dynamics features by focusing on the relationship between the multifractality and the measured scale-free intervals.Furthermore, neural networks are applied to train the above-mentioned parameters for the classification and identification of three kinds of EEG signals, i.e., health, interictal phase, and ictal phase. By evaluating our experimental results, the classification accuracy is 99.0%, reflecting the effectiveness of the WHVG algorithm in extracting the potential dynamic characteristics of EEG signals.

基于有效注意力和GAN结合的脑卒中EEG增强算法

在基于脑电的卒中分类诊断任务中,以卷积神经网络为基础的深度模型得到广泛应用,但由于卒中类别病患样本数量少,导致数据集类别不平衡,降低了分类精度。现有的少数类数据增强方法大多采用生成对抗网络(GAN),生成效果一般,虽然可通过引入缩放点乘注意力改善样本生成质量,但存储及运算代价往往较大。针对此问题,构建一种基于线性有效注意力的渐进式数据增强算法LESA-CGAN。首先,算法采用双层自编码条件生成对抗网络架构,分别进行脑电标签特征提取及脑电样本生成,并使生成过程逐层精细化;其次,通过在编码部分引入线性有效自注意力(LESA)模块,加强脑电的标签隐层特征提取,并降低网络整体的运算复杂度。消融与对比实验结果表明,在合理的编码层数与生成数据比例下,LESA-CGAN与其他基准方法相比计算资源占用较少,且在样本生成质量指标上实现了10%的性能提升,各频段生成的脑电特征样本均更加自然,同时将病患分类的准确率和敏感度提高到了98.85%和98.79%。

Feature Selection with Deep Belief Network for Epileptic Seizure Detection on EEG Signals

The term Epilepsy refers to a most commonly occurring brain disorder after a migraine.Early identification of incoming seizures significantly impacts the lives of people with Epilepsy.Automated detection of epileptic seizures(ES)has dramatically improved the life quality of the patients.Recent Electroencephalogram(EEG)related seizure detection mechanisms encountered several difficulties in real-time.The EEGs are the non-stationary signal,and seizure patternswould changewith patients and recording sessions.Further,EEG data were disposed to wide noise varieties that adversely moved the recognition accuracy of ESs.Artificial intelligence(AI)methods in the domain of ES analysis use traditional deep learning(DL),and machine learning(ML)approaches.This article introduces an Oppositional Aquila Optimizer-based Feature Selection with Deep Belief Network for Epileptic Seizure Detection(OAOFS-DBNECD)technique using EEG signals.The primary aim of the presented OAOFS-DBNECD system is to categorize and classify the presence of ESs.The suggested OAOFS-DBNECD technique transforms the EEG signals into.csv format at the initial stage.Next,the OAOFS technique selects an optimal subset of features using the preprocessed data.For seizure classification,the presented OAOFS-DBNECD technique applies Artificial Ecosystem Optimizer(AEO)with a deep belief network(DBN)model.An extensive range of simulations was performed on the benchmark dataset to ensure the enhanced performance of the presented OAOFS-DBNECD algorithm.The comparison study shows the significant outcomes of the OAOFS-DBNECD approach over other methodologies.In addition,the result of the suggested approach has been evaluated using the CHB-MIT database,and the findings demonstrate accuracy of 97.81%.These findings confirmed the best seizure categorization accuracy on the EEG data considered.

Multi-View & Transfer Learning for Epilepsy Recognition Based on EEG Signals

Epilepsy is a central nervous system disorder in which brain activity becomes abnormal.Electroencephalogram(EEG)signals,as recordings of brain activity,have been widely used for epilepsy recognition.To study epilep-tic EEG signals and develop artificial intelligence(AI)-assist recognition,a multi-view transfer learning(MVTL-LSR)algorithm based on least squares regression is proposed in this study.Compared with most existing multi-view transfer learning algorithms,MVTL-LSR has two merits:(1)Since traditional transfer learning algorithms leverage knowledge from different sources,which poses a significant risk to data privacy.Therefore,we develop a knowledge transfer mechanism that can protect the security of source domain data while guaranteeing performance.(2)When utilizing multi-view data,we embed view weighting and manifold regularization into the transfer framework to measure the views’strengths and weaknesses and improve generalization ability.In the experimental studies,12 different simulated multi-view&transfer scenarios are constructed from epileptic EEG signals licensed and provided by the Uni-versity of Bonn,Germany.Extensive experimental results show that MVTL-LSR outperforms baselines.The source code will be available on https://github.com/didid5/MVTL-LSR.

抑郁症EEG诊断的类脑学习模型

抑郁症是一种全球性精神疾病,传统诊断方法主要依靠量表与医生的主观评估,无法有效识别症状,甚至存在误诊的风险。基于生理信号的深度学习辅助诊断有望改善传统缺乏生理学依据的方法。然而,传统深度学习方法依赖巨大算力,且大多是端到端的网络学习。这些学习方法也缺乏生理可解释性,限制了辅助诊断临床应用。提出一种用于抑郁症脑电图(electroencephalogram,EEG)诊断的类脑学习模型,在功能层面,构建脉冲神经网络对抑郁症与健康个体进行分类,精度超过97.5%,相比深度卷积方法,脉冲方法降低了能耗;在结构层面,利用复杂网络建立脑连接的空间拓扑并分析其图特征,找出了抑郁症个体潜在的脑功能连接异常机制。

Emotion Measurement Using Biometric Signal

In recent years, research on the estimation of human emotions has been active, and its application is expected in various fields. Biological reactions, such as electroencephalography (EEG) and root mean square successive difference (RMSSD), are indicators that are less influenced by individual arbitrariness. The present study used EEG and RMSSD signals to assess the emotions aroused by emotion-stimulating images in order to investigate whether various emotions are associated with characteristic biometric signal fluctuations. The participants underwent EEG and RMSSD while viewing emotionally stimulating images and answering the questionnaires. The emotions aroused by emotionally stimulating images were assessed by measuring the EEG signals and RMSSD values to determine whether different emotions are associated with characteristic biometric signal variations. Real-time emotion analysis software was used to identify the evoked emotions by describing them in the Circumplex Model of Affect based on the EEG signals and RMSSD values. Emotions other than happiness did not follow the Circumplex Model of Affect in this study. However, ventral attentional activity may have increased the RMSSD value for disgust as the β/θ value increased in right-sided brain waves. Therefore, the right-sided brain wave results are necessary when measuring disgust. Happiness can be assessed easily using the Circumplex Model of Affect for positive scene analysis. Improving the current analysis methods may facilitate the investigation of face-to-face communication in the future using biometric signals.

基于3D特征融合与轻量化CNN的情绪EEG识别

情绪变化可引起头皮脑电信号的改变,基于脑电信号的情绪识别是近年来情绪研究的一个重要方向。为此,提出一种基于3D特征融合与轻量化卷积神经网络的情绪EEG识别方法,使用2 s窗口的3D特征图作为输入,并根据效价和唤醒提供情绪状态作为输出。在DEAP公开数据集上对所提方法进行受试者依赖实验,结果表明情绪识别性能评估效价和唤醒识别准确率分别为(97.08±0.32)%和(96.78±0.34)%。所提方法具有较高的情绪识别准确度和较低的计算复杂度,适用于实际场景中的情绪识别。

基于改进型生成式对抗网络的EEG-fNIRS多模态信号数据增广研究

目的基于深度学习的脑电图-功能性近红外光谱技术(electroencephalogram-functional near-infrared spectroscopy,EEG-fNIRS)多模态脑机接口在康复工程中具有广泛的应用前景,但存在数据量不足的问题。为此,本文提出一种基于改进条件生成式对抗网络(conditional generative adversarial network,CGAN)的EEG-fNIRS多模态信号数据增广方法,以解决EEG-fNIRS多模态脑机接口与深度学习结合时面临的数据量匮乏的问题。方法首先,对EEG和fNIRS数据进行滤波、归一化和下采样等预处理。然后,针对EEG的非平稳特点,在CGAN生成器和判别器中增加自注意力机制,获得EEG数据增广模型CGAN_(E),加强捕捉和学习时变关键信息的能力。同时,针对fNIRS采样率低、信息量不充分问题,在CGAN生成器和判别器中增加上采样卷积层,获得fNIRS数据增广模型CGAN_(f),加强模型的信息挖掘能力,并将CGAN_(E)和CGAN_(f)的条件信息设置为类标签;进而,利用CGAN_(E)和CGAN_(f)对每导EEG和fNIRS分别进行增广,并将多导EEG扩增数据和多导fNIRS[包括氧合血红蛋白浓度(oxyhemoglobin concentration,HbO)和脱氧血红蛋白浓度(deoxyhemoglobin concentration,HbR)两种]扩增数据串接融合,获得EEG-fNIRS多模态增广数据。最后,对公开的EEG-fNIRS多模态信号数据集TU-Berlin-A前6名受试者数据进行增广实验,并设计一维卷积神经网络分类器评估增广数据的质量。结果基于EEG-fNIRS多模态信号公开数据集TU-Berlin-A前6名受试者的左右手运动想象数据进行实验研究表明,当数据扩增5倍时,本文方法取得94.81%的平均分类准确率。结论CGAN_(E)和CGAN_(f)能够生成接近真实数据分布的EEG和fNIRS信号,验证了对CGAN改进和本文所提EEG-fNIRS多模态数据增广方法的有效性。

基于时空和频域特征的EEG帕金森疾病识别

脑电图(EEG)中蕴含着有关脑功能的丰富信息,这些信息对不同类型神经系统疾病的检测和诊断非常重要。针对单一特征无法充分表达脑电信号的问题,本文融合了频域特征和时空信息来更好的对信号进行表征,并提出一种基于时空和频域特征的注意力网络(STFACN)用于帕金森疾病(PD)的自动检测。在频域角度,利用快速傅里叶变换法从多通道脑电图中求取Delta、Theta、Alpha频段的平均功率特征。同时构建基于时空特征的紧凑型卷积神经网络,并将通道注意力机制嵌入到网络中,自适应提取表征PD的时空特征。最后将基于频域特征的模型与基于时空特征的紧凑型卷积神经网络模型进行融合,在新墨西哥州大学(UNM)数据集上进行实验,特异性、敏感性、准确率分别达到87.97%、84.39%、86.89%。在爱荷华大学(UI)数据集上进行跨数据集实验,准确率达到77.33%。实验结果表明:与现有的方法相比,本文提出的方法能够从原始脑电图中挖掘出有效特征,在基于EEG的帕金森疾病识别问题上准确率高,泛化能力强。

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.

A Method for Quantifying the Emotional Intensity and Duration of a Startle Reaction with Customized Fractal Dimensions of EEG Signals

The assessment of emotions with fractal dimensions of EEG signals has been attempted before, but the quantification of the intensity and duration of sudden and short emotions remains a challenge. This paper suggests a method for this purpose, by using a new fractal dimension algorithm and by adjusting the amplitude of the EEG signal in order to obtain maximal separation of high and low fractal dimensions. The emotion was induced by embedding a scary image at 20 seconds in landscape videos of 60 seconds length. The new method did not only detect the onset of the emotion correctly, but also revealed its duration and intensity. The intensity is based on the magnitude and impulse of the fractal dimension signal. It is also shown that Higuchi’s method does not always detect emotion spikes correctly;on the contrary, the region of the expected emotional response can be represented by fractal dimensions smaller than the rest of the signal, whereas the new method directly reveals distinct spikes. The duration of these spikes was 10 - 11 seconds. The magnitude of these spikes varied across the EEG channels. The build-up and cool-down of the emotions can occur with steep and flat gradients.

Adaptive Signal Enhancement Unit for EEG Analysis in Remote Patient Care Monitoring Systems

In this paper we propose an efcient process of physiological artifact elimination methodology from brain waves(BW),which are also commonly known as electroencephalogram(EEG)signal.In a clinical environment during the acquisition of BW several artifacts contaminates the actual BW component.This leads to inaccurate and ambiguous diagnosis.As the statistical nature of the EEG signal is more non-stationery,adaptive ltering is the more promising method for the process of artifact elimination.In clinical conditions,the conventional adaptive techniques require many numbers of computational operations and leads to data samples overlapping and instability of the algorithm used.This causes delay in diagnosis and decision making.To overcome this problem in our work we propose to set a threshold value to diminish the problem of round off error.The resultant adaptive algorithm based on this strategy is Non-linear Least mean square(NL2MS)algorithm.Again,to improve this algorithm in terms of ltering capability we perform data normalization,using this algorithm several hybrid versions are developed to improve ltering and reduce computational operations.Using the method,a new signal enhancement unit(SEU)is realized and performance of various hybrid versions of algorithms examined using real EEG signals recorded from the subject.The ability of the proposed schemes is measured in terms of convergence,enhancement and multiplications required.Among various SEUs,the MCN2L 2MS algorithm achieves 14.6734,12.8732,10.9257,15.7790 dB during the artifact removal of RA,EMG,CSA and EBA components with only two multiplications.Hence,this algorithm seems to be better candidate for artifact elimination.

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%.

THE EFFECT OF ACUPUNCTURING ACUPOINTS ON THE CHANGE OF ELECTROENCEPHALOGRAM (EEG) IN ENDOTOXIC SHOCKED RATS

In present work,EEG and BP were used as the indexes to observe the relationbetween the change of EEG and the change of BP in the endotoxic shocked rats。At maintainingshock for 1 hr,dysrhythmia of EEG appeared in 38/46 cases.Simultaneously,there was a markeddrop in Bp,P【0.05.Following the shocked time prolonged,dysrhythmia was getting severe。AfterEA”Rengzhong"(n=14)or“Zusanli”(n=12),BP was significantly increased(P【0.05),anddysrhythmia of EEG showed clear improvement in most of the rats。There was a close relation be-tween the changes of EEG and BP,the change of EEG had a direct bearing on the change of BP.

Classification of Imagined Speech EEG Signals with DWT and SVM

With the development of human-computer interaction technology,brain-computer interface(BCI)has been widely used in medical,entertainment,military,and other fields.Imagined speech is the latest paradigm of BCI and represents the mental process of imagining a word without making a sound or making clear facial movements.Imagined speech allows patients with physical disabilities to communicate with the outside world and use smart devices through imagination.Imagined speech can meet the needs of more complex manipulative tasks considering its more intuitive features.This study proposes a classification method of imagined speech Electroencephalogram(EEG)signals with discrete wavelet transform(DWT)and support vector machine(SVM).An open dataset that consists of 15 subjects imagining speaking six different words,namely,up,down,left,right,backward,and forward,is used.The objective is to improve the classification accuracy of imagined speech BCI system.The features of EEG signals are first extracted by DWT,and the imagined words are clas-sified by SVM with the above features.Experimental results show that the proposed method achieves an average accuracy of 61.69%,which is better than those of existing methods for classifying imagined speech tasks.

基于改进Renyi熵算法的EEG心算任务识别

结构熵是度量网络复杂度的重要手段,为了弥补传统结构熵仅仅关注网络单一特性的问题,提出了一种改进Renyi熵算法来研究心算任务下的EEG脑网络,引入了两个重要网络属性——分形维数和介数中心性来提高网络复杂性的度量能力。之后,基于心算EEG数据计算两两电极间的相位锁定值(PLV),构建了复杂脑网络,并进行复杂度分析。结果表明,在α频段,心算状态下额叶与顶枕叶的脑同步性低于休息状态,心算状态的脑网络复杂性高于休息状态。利用支持向量机(SVM)实现了休息、心算状态的识别,算法识别准确率达到了88.42%。

EEGbands: A Computer Program to Statistically Analyze Parameters of Electroencephalographic Signals

The quantitative analysis of electroencephalographic activity (EEG) is a useful tool for the study of changes in brain electrical activity during cognitive and behavioral functions in several experimental conditions. Their recording and analysis are currently carried out primarily through the use of computer programs. This paper presents a computerized program (EEGbands) created for Windows operating systems using the Delphi language, and designed to analyze EEG signals and facilitate their quantitative exploration. EEGbands applies Rapid Fourier Transformation to the EEG signals of one or more groups of subjects to obtain absolute and relative power spectra. It also calculates both interhemispheric and intrahemispheric correlation and coherence spectra and, finally, applies parametrical statistical analysis to these spectral parameters calculated for wide frequency EEG bands. Unlike other programs, EEGbands is simple and inexpensive, and rapidly and precisely generates results files with the corresponding statistical significances. The efficacy and versatility of EEGbands allow it to be easily adapted to different experimental and clinical needs.

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.

3DMKDR:3D Multiscale Kernels CNN Model for Depression Recognition Based on EEG

Depression has become a major health threat around the world,especially for older people,so the effective detection method for depression is a great public health challenge.Electroencephalogram(EEG)can be used as a biomarker to effectively explore depression recognition.Motivated by the studies that multiple smaller scale kernels could increase nonlinear expression compared to a larger kernel,this article proposes a model named the three-dimensional multiscale kernels convolutional neural network model for the depression disorder recognition(3DMKDR),which is a three-dimensional convolutional neural network model with multiscale convolutional kernels for depression recognition based on EEG signals.A three-dimensional structure of the EEG is built by extending one-dimensional feature sequences into a two-dimensional electrode matrix to excavate the related spatiotemporal information among electrodes and the collected electrode matrix.By the major depressive disorder(MDD)and the multi-modal open dataset for mental-disorder analysis(MODMA)datasets,the experiment shows that the accuracies of depression recognition are up to99.86%and 98.01%in the subject-dependent experiment,and 95.80%and 82.27%in the subjectindependent experiment,which are higher than alternative competitive methods.The experimental results demonstrate that the proposed 3DMKDR is potentially useful for depression recognition in older persons in the future.