Deep learning approach to detect seizure using reconstructed phase space images

Epilepsy is a chronic neurological disorder that affects the function of the brain in people of all ages. It manifests in the electroencephalogram (EEG) signal which records the electrical activity of the brain. Various image processing, signal processing, and machine-learning based techniques are employed to analyze epilepsy, using spatial and temporal features. The nervous system that generates the EEG signal is considered nonlinear and the EEG signals exhibit chaotic behavior. In order to capture these nonlinear dynamics, we use reconstructed phase space (RPS) representation of the signal. Earlier studies have primarily addressed seizure detection as a binary classification (normal vs. ictal) problem and rarely as a ternary class (normal vs. interictal vs. ictal) problem. We employ transfer learning on a pre-trained deep neural network model and retrain it using RPS images of the EEG signal. The classification accuracy of the model for the binary classes is (98.5±1.5)% and (95±2)% for the ternary classes. The performance of the convolution neural network (CNN) model is better than the other existing statistical approach for all performance indicators such as accuracy, sensitivity, and specificity. The result of the proposed approach shows the prospect of employing RPS images with CNN for predicting epileptic seizures.     

基于迁移学习和空洞卷积的癫痫状态识别方法

癫痫患者脑电信号的自动检测和发作诊断对临床治疗癫痫具有重要意义。针对训练数据有限及训练与测试数据分布不一致的难点,采用领域间联合知识迁移学习方法,实现小训练数据量下的癫痫状态识别。首先对脑电信号进行4层小波包分解,提取小波包分解系数作为特征,通过边缘分布和联合分布迭代调整,完成源域和目标域特征之间的知识迁移,训练空洞卷积神经网络作为分类器,完成目标域癫痫状态识别。分别在波士顿儿童医院CHB-MIT脑电数据集(22 例被试,共计790 h)和波恩大学癫痫脑电数据集(5 组,每组100 个片段,每段23.6 s)上进行算法验证,实验结果表明,所提出的方法对复杂癫痫状态的平均识别准确度、敏感性、特异性在CHB-MIT数据集上达到96.8％、96.1％、96.4％;在波恩数据集上,平均识别准确率为96.9%,有效提高了癫痫状态识别综合性能,实现了癫痫发作稳定可靠检测。     

Electroencephalogram signals processing for topographic brain mapping and epilepsies classification

In this study, topographic brain mapping and wavelet transform-neural network method are used for the classification of grand mal (clonic stage) and petit mal (absence) epilepsies into healthy, ictal and interictal (EEGs). Preprocessing is included to remove artifacts occurred by blinking, wandering baseline (electrodes movement) and eyeball movement using the Discrete Wavelet Transformation (DWT). De-noising EEG signals from the AC power supply frequency with a suitable notch filter is another job of preprocessing. In experimental data, the preprocessing enhanced speed and accuracy of the processing stage (wavelet transform and neural network). The EEGs signals are categorized to normal and petit mal and clonic epilepsy by an expert neurologist. The categorization is confirmed by Fast Fourier Transform (FFT) analysis and brain mapping. The dataset includes waves such as sharp, spike and spike-slow wave. Through the Counties Wavelet Transform (CWT) of EEG records, transient features are accurately captured and separated and used as classifier input. We introduce a two-stage classifier based on the Learning Vector Quantization (LVQ) neural network location in both time and frequency contexts. The brain mapping used for finding the epilepsy locates in the brain. The simulation results are very promising and the accuracy of the proposed classifier in experimental clinical data is ∼80%.     

基于自适应多尺度脑功能连接的局灶性癫痫发作检测方法研究

利用长时程脑电图检测癫痫发作是临床中较为广泛的应用,然而这项工作乏味、耗时,且很大程度上依赖于临床医生的自身经验和主观判断,准确性和可重复性也较低。针对长时程脑电图检测癫痫中存在的问题,提出一种基于自适应多尺度脑功能连接的癫痫发作检测方法(AMBFC),并选取10例局灶性癫痫患者的发作期和非发作期的样本作为研究对象。首先在一个滑动时间窗内,通过多元经验模态分解(MEMD)提取19通道脑电信号的7个本征模函数(IMF)分量及残差;然后建立多变量自回归(MVAR)模型,利用有向传递函数(DTF)提取流出信息强度,进行特征组合,并通过主成分分析(PCA)降维保留原始特征数目的85%;最后经代价敏感支持向量机(CSVM)分类区分发作期和非发作期脑电,并通过五重交叉验证进行癫痫发作检测算法的效果评价。结果表明,AMBFC算法检测脑电癫痫发作得到的平均准确率为98.6%,精确率为81.9%,召回率为81.4%,F2值为0.80。与各IMF分量、DTF-CSVM算法等检测结果相比,AMBFC算法更具有优越性。有望应用于长时程脑电的实时监测。     

An optimized design of seizure detection system using joint feature extraction of multichannel EEG signals

The detection of seizure onset and events using electroencephalogram (EEG) signals are important tasks in epilepsy research. The literature available on seizure detection has discussed the implementation of advanced signal processing algorithms using tools accessed over the cloud. However, seizure monitoring application needs near sensor processing due to privacy and latency issues. In this paper, a real time seizure detection system has been implemented using an embedded system. The proposed system is based on ensemble empirical mode decomposition (EEMD) and tunable-Q wavelet transform (TQWT) algorithms. The analysis and classification of non-stationary EEG signals require the wavelet transform with high Q-factor. However, direct use of TQWT increases the computational complexity of feature extraction from multivariate EEG signals. In this paper, the first step is to process the signal by using EEMD to obtain 8 intrinsic mode functions (IMFs). The Kraskov (KraEn), sample (SampEn), and permutation (PermEn) entropy features of IMFs are extracted and based on optimum values, and 4 IMFs are decomposed using TQWT. Secondly, centered correntropy (CenCorrEn) features of the 1^st and 16^th sub-band of TQWT have been used as classifier inputs. The performance of multilayer perceptron neural networks (MLPNN), least squares support vector machine (LSSVM), and random forest (RF) classifiers has been tested on the multichannel EEG data recorded from a local hospital. The RF classifier has produced the highest accuracy of 96.2% in classifying the signals. The proposed scheme has been employed in developing an embedded seizure detection system to assist neurologists in making seizure diagnostic decisions.     

High-Frequency Oscillations Detected in Epileptic Networks Using Swarmed Neural-Network Features

Localizing epileptic networks is a central challenge in guiding epilepsy surgery, deploying antiepileptic devices, and elucidating mechanisms underlying seizure generation. Recent work from our group and others suggests that high-frequency epileptic oscillations (HFEOs) arise from brain regions constituting epileptic networks, and may be important to seizure generation. HFEOs are brief 50–500 Hz pathologic events measured in intracranial field and unit recordings in patients with refractory epilepsy. They are challenging to detect due to low signal to noise ratio, and because they occur in multiple channels with great frequency. Their morphology is also variable and changes with distance from intracranial electrode contacts, which are sparsely placed for patient safety. Thus reliable, automated methods to detect HFEOs are required to localize and track seizure generation in epileptic networks. We present a novel method for mapping the temporal evolution of these oscillations in human epileptic networks. The technique combines a particle swarm optimization algorithm with a neural network to create features that robustly detect and track HFEOs in human intracranial EEG (IEEG) recordings. We demonstrate the algorithm’s performance on IEEG data from six patients, one pediatric and five adult, and compare it to an existing method for detecting high-frequency oscillations.     

Early detection of epileptic seizures based on parameter identification of neural mass model

Physiologically based models are attractive for seizure detection, as their parameters can be explicitly related to neurological mechanisms. We propose an early seizure detection algorithm based on parameter identification of a neural mass model. The occurrence of a seizure is detected by analysing the time shift of key model parameters. The algorithm was evaluated against the manual scoring of a human expert on intracranial EEG samples from 16 patients suffering from different types of epilepsy. Results suggest that the algorithm is best suited for patients suffering from temporal lobe epilepsy (sensitivity was 95.0%±10.0% and false positive rate was 0.20±0.22 per hour).     

Log Energy Entropy-Based EEG Classification with Multilayer Neural Networks in Seizure

In this study, normal EEG series recorded from healthy volunteers and epileptic EEG series recorded from patients within and without seizure are classified by using Multilayer Neural Network (MLNN) architectures with respect to several time domain entropy measures such as Shannon Entropy (ShanEn), Log Energy Entropy (LogEn), and Sample Entropy (Sampen). In tests, the MLNN is performed with several numbers of neurons for both one hidden layer and two hidden layers. The results show that segments in seizure have significantly lower entropy values than normal EEG series. This result indicates an important increase of EEG regularity in epilepsy patients. The LogEn approach, which has not been experienced in EEG classification yet, provides the most reliable features into the EEG classification with very low absolute error as 0.01. In particular, the MLNN can be proposed to distinguish the seizure activity from the seizure-free epileptic series where the LogEn values are considered as signal features that characterize the degree of EEG complexity. The highest classification accuracy is obtained for one hidden layer architecture.     

Predicting movement during anaesthesia by complexity analysis of electroencephalograms

A new approach to predicting movement during anaesthesia by using complexity analysis of electroencephalograms (EEG) signals is presented. The raw EEG signal is first decomposed into six consecutive different scaling components by wavelet transform on the basis of its self-similarity. The Lempel-Ziv complexity measures C(n) are extracted from the raw EEG and its corresponding components by complexity analysis. Prediction of movement during anaesthesia is then made by a four-layer artificial neural network (ANN) using the C(n)s. The combination of these three different approaches enables the system to address the non-analytical, nonstationary, non-linear and dynamical properties of the EEG. From 20 dog experiments, 109 distinct EEG recordings are collected under isoflurane anaesthesia. Testing the ANN using the ‘drop one dog’ method, the performance obtained for the system in detecting movement is: sensitivity 88%, specificity 97% and accuracy 92%. Comparisons with other methods, such as spectral edge frequency, median frequency and principal component analysis, show that the proposed system has a certain advantage. This new method is computationally fast and well suited for realtime clinical implementation.     

Wavelet-based feature extraction for classification of epileptic seizure EEG signal

Electroencephalogram (EEG) signal-processing techniques are the prominent role in the detection and prediction of epileptic seizures. The detection of epileptic activity is cumbersome and needs a detailed analysis of the EEG data. Therefore, an efficient method for classifying EEG data is required. In this work, a constructive pattern recognition strategy for analysing EEG data as normal and epileptic seizure has been proposed. With this strategy, the signals were decomposed into frequency sub-bands using discrete wavelet transform (DWT). principal component analysis (PCA) and linear discriminant analysis (LDA) are applied to reduce the dimensionality of EEG data. These reduced features were used as input to Naïve Bayes and K-Nearest Neighbour Classifier to classify normal or epileptic seizure signal. The performance of classifier was evaluated in terms of accuracy, sensitivity and specificity. The experimental results show that PCA with Naïve Bayes classifier provides 98.6% accuracy and LDA with Naïve Bayes classifier attains improved result of 99.8% accuracy. Also, the result shows that PCA, LDA with K-NN achieves 98.5% and 100% accuracy. This evaluation is used to propose a reliable, practical epilepsy detection method to enhance the patient’s care and quality of life.     

Dynamical intrinsic functional architecture of the brain during absence seizures

Epilepsy is characterized by recurrent and temporary brain dysfunction due to discharges of interconnected groups of neurons. The brain of epilepsy patients has a dynamic bifurcation that switches between epileptic and normal states. The dysfunctional state involves large-scale brain networks. It is very important to understand the network mechanisms of seizure initiation, maintenance, and termination in epilepsy. Absence epilepsy provides a unique model for neuroimaging investigation on dynamic evolutions of brain networks over seizure repertoire. By using a dynamic functional connectivity and graph theoretical analyses to study absence seizures (AS), we aimed to obtain transition of network properties that account for seizure onset and offset. We measured resting-state functional magnetic resonance imaging and simultaneous electroencephalography (EEG) from children with AS. We used simultaneous EEG to define the preictal, ictal and postictal intervals of seizures. We measured dynamic connectivity maps of the thalamus network and the default mode network (DMN), as well as functional connectome topologies, during the three different seizure intervals. The analysis of dynamic changes of anti-correlation between the thalamus and the DMN is consistent with an inhibitory effect of seizures on the default mode of brain function, which gradually fades out after seizure onset. Also, we observed complex transitions of functional network topology, implicating adaptive reconfiguration of functional brain networks. In conclusion, our work revealed novel insights into modifications in large-scale functional connectome during AS, which may contribute to a better understanding the network mechanisms of state bifurcations in epileptogenesis.     

Recognition of Motor Imagery Electroencephalography Using Independent Component Analysis and Machine Classifiers

Motor imagery electroencephalography (EEG), which embodies cortical potentials during mental simulation of left or right finger lifting tasks, can be used to provide neural input signals to activate a brain computer interface (BCI). The effectiveness of such an EEG-based BCI system relies on two indispensable components: distinguishable patterns of brain signals and accurate classifiers. This work aims to extract two reliable neural features, termed contralateral and ipsilateral rebound maps, by removing artifacts from motor imagery EEG based on independent component analysis (ICA), and to employ four classifiers to investigate the efficacy of rebound maps. Results demonstrate that, with the use of ICA, recognition rates for four classifiers (fisher linear discriminant (FLD), back-propagation neural network (BP-NN), radial-basis function neural network (RBF-NN), and support vector machine (SVM)) improved significantly, from 54%, 54%, 57% and 55% to 70.5%, 75.5%, 76.5% and 77.3%, respectively. In addition, the areas under the receiver operating characteristics (ROC) curve, which assess the quality of classification over a wide range of misclassification costs, also improved from .65, .60, .62, and .64 to .74, .76, .80 and .81, respectively.     

一种基于EEG特征提取的癫痫棘波综合检测判决方法

癫痫特征的自动检测在临床应用上具有重要的意义。本研究综合小波变换、非线性能量算子、特征提取和神经网络等技术，提出了一种癫痫棘波检测系统，充分发挥各技术的优点，在对真实脑电数据的处理中，表现出良好的性能。     

Predicting movement during anaesthesia by complexity analysis of electroencephalograms

A new approach to predicting movement during anaesthesia by using complexity analysis of electroencephalograms (EEG) signals is presented. The raw EEG signal is first decomposed into six consecutive different scaling components by wavelet transform on the basis of its self-similarity. The Lempel-Ziv complexity measures C(n) are extracted from the raw EEG and its corresponding components by complexity analysis. Prediction of movement during anaesthesia is then made by a four-layer artificial neural network (ANN) using the C(n)s. The combination of these three different approaches enables the system to address the non-analytical, non-stationary, non-linear and dynamical properties of the EEG. From 20 dog experiments, 109 distinct EEG recordings are collected under isoflurane anaesthesia. Testing the ANN using the 'drop one dog' method, the performance obtained for the system in detecting movement is: sensitivity 88%, specificity 97% and accuracy 92%. Comparisons with other methods, such as spectral edge frequency, median frequency and principal component analysis, show that the proposed system has a certain advantage. This new method is computationally fast and well suited for realtime clinical implementation.     

全面性棘慢复合波相关神经网络的EEG-fMRI研究

全面性癫痫的脑电图(EEG)呈全面性棘慢复合波放电(GSWDs),近来研究表明GSWDs并非"全面地"累及整个大脑,而是与特异的神经网络相关。同步脑电图联合功能磁共振(EEG-fMRI)技术可无创、高时空分辨率地了解痫性电活动时各脑区的代谢功能变化。EEG-fMRI研究已发现丘脑-基底节-皮质网络在GSWDs的产生与维持中起重要作用。与GSWDs相关的功能磁共振(fMRI)信号变化规律:前额叶、额顶叶、后扣带回、楔前叶出现相似的负激活(deactivation)信号,丘脑出现较为一致的激活(activation)信号,基底节出现负激活信号。进一步研究发现,fMRI信号随GSWDs出现的时程动态变化,皮质神经元活动可能先于丘脑出现,但不同EEG-fMRI研究的皮质激活部位存在个体化表现,散在地分布于额叶、顶叶等皮质区域。进一步发展更加复杂的分析方法揭示fMRI信号的时程变化规律,研究GSWDs产生的起始、传导和维持涉及的神经网络,揭示全面性癫痫的行为表现、药物治疗反应与神经网络的关系是未来EEG-fMRI研究的方向。     

Single-trial classification of elementary comparison processes on the basis of instantaneous EEG and MEG coherences

The present study is focused on the evidence of possible single-trial EEG/MEG analysis of information processing. The discrimination between thinking modalities of concept activation and pattern comparison for single tasks of elementary comparison procedures is investigated. A neural network classifier with backpropagation learning algorithm is used. The input vector is constructed by parameters of instantaneous coherence (13-20 Hz) between several channel pairs of the EEG and/or of the MEG. Thereby, the strength of synchronization and the time location of synchronization phenomena are taken into consideration. The combination of EEG and MEG coherence parameters led to a classification accuracy of 85-94% for single subjects. Generally, results reached by neural network classifier show a better generalization than linear discriminant analysis.     

Wavelet-based sparse functional linear model with applications to EEGs seizure detection and epilepsy diagnosis

In epilepsy diagnosis or epileptic seizure detection, much effort has been focused on finding effective combination of feature extraction and classification methods. In this paper, we develop a wavelet-based sparse functional linear model for representation of EEG signals. The aim of this modeling approach is to capture discriminative random components of EEG signals using wavelet variances. To achieve this goal, a forward search algorithm is proposed for determination of an appropriate wavelet decomposition level. Two EEG databases from University of Bonn and University of Freiburg are used for illustration of applicability of the proposed method to both epilepsy diagnosis and epileptic seizure detection problems. For this data considered, we show that wavelet-based sparse functional linear model with a simple classifier such as 1-NN classification method leads to higher classification results than those obtained using other complicated methods such as support vector machine. This approach produces a 100 % classification accuracy for various classification tasks using the EEG database from University of Bonn, and outperforms many other state-of-the-art techniques. The proposed classification scheme leads to 99 % overall classification accuracy for the EEG data from University of Freiburg.     

癫痫发作时的脑电图研究

脑电图已成为癫痫诊断中不可缺少的手段,特别是发作时的脑电图为癫痫诊断的基本要素,本文对22例癫痫发作时的脑电图进行了研究,分析了发作时的癫痫放电及发作前后的波形演变,对癫痫的分类及治疗有着重要的意义。     

动态因果模型在脑网络研究中的应用进展

大脑各功能区之间的有效连接是脑科学研究领域的一个重要内容.研究在不同情形下相关脑区之间有效连接所构成的大脑网络,对于全面理解大脑的功能机制,治疗各种与大脑相关疾病,开发脑功能具有重要意义.动态因果模型是一种分析大脑有效连接的优势方法.结合功能性磁共振成像、脑电、近红外脑功能成像等3种检测技术,综述动态因果模型的相关研究...     

基于多特征多关系图卷积神经网络的癫痫脑电分类

目的:提出一种基于多特征多关系图卷积神经网络的癫痫脑电分类方法,改进图卷积神经网络在癫痫脑电分类领域的应用,提升分类准确率。方法:分别提取癫痫脑电信号的1个频域特征、3个时频域特征和2个非线性动力学特征作为模型节点的特征。提取脑电通道之间的空间相似性和频谱相似性,融合两种通道相似性作为整体图节点之间的边关系矩阵。结果:在TUEP数据集上进行实验,准确率、精确率、召回率、F1分数、AUC结果分别为:0.87±0.02、0.91±0.04、0.82±0.04、0.86±0.02、0.90±0.03。结论:提出的模型与单特征和单关系的图卷积神经网络相比,对于癫痫脑电分类的提升效果明显。