Spectral feature extraction of EEG signals and pattern recognition during mental tasks of 2-D cursor movements for BCI using SVM and ANN

Brain computer interface (BCI) is a new communication way between man and machine. It identifies mental task patterns stored in electroencephalogram (EEG). So, it extracts brain electrical activities recorded by EEG and transforms them machine control commands. The main goal of BCI is to make available assistive environmental devices for paralyzed people such as computers and makes their life easier. This study deals with feature extraction and mental task pattern recognition on 2-D cursor control from EEG as offline analysis approach. The hemispherical power density changes are computed and compared on alpha–beta frequency bands with only mental imagination of cursor movements. First of all, power spectral density (PSD) features of EEG signals are extracted and high dimensional data reduced by principle component analysis (PCA) and independent component analysis (ICA) which are statistical algorithms. In the last stage, all features are classified with two types of support vector machine (SVM) which are linear and least squares (LS-SVM) and three different artificial neural network (ANN) structures which are learning vector quantization (LVQ), multilayer neural network (MLNN) and probabilistic neural network (PNN) and mental task patterns are successfully identified via k-fold cross validation technique.     

Epileptic seizure detection in EEG signal using machine learning techniques

Epilepsy is a well-known nervous system disorder characterized by seizures. Electroencephalograms (EEGs), which capture brain neural activity, can detect epilepsy. Traditional methods for analyzing an EEG signal for epileptic seizure detection are time-consuming. Recently, several automated seizure detection frameworks using machine learning technique have been proposed to replace these traditional methods. The two basic steps involved in machine learning are feature extraction and classification. Feature extraction reduces the input pattern space by keeping informative features and the classifier assigns the appropriate class label. In this paper, we propose two effective approaches involving subpattern based PCA (SpPCA) and cross-subpattern correlation-based PCA (SubXPCA) with Support Vector Machine (SVM) for automated seizure detection in EEG signals. Feature extraction was performed using SpPCA and SubXPCA. Both techniques explore the subpattern correlation of EEG signals, which helps in decision-making process. SVM is used for classification of seizure and non-seizure EEG signals. The SVM was trained with radial basis kernel. All the experiments have been carried out on the benchmark epilepsy EEG dataset. The entire dataset consists of 500 EEG signals recorded under different scenarios. Seven different experimental cases for classification have been conducted. The classification accuracy was evaluated using tenfold cross validation. The classification results of the proposed approaches have been compared with the results of some of existing techniques proposed in the literature to establish the claim.     

Automated sleep staging of OSAs based on ICA preprocessing and consolidation of temporal correlations

An automated sleep staging based on analyzing long-range time correlations in EEG is proposed. These correlations, indicating time-scale invariant property or self-similarity at different time scales, are known to be salient dynamical characteristics of stage succession for a sleeping brain even when the subject suffers a destructive disorder such as Obstructive Sleep Apnea (OSA). The goal is to extract a set of complementary features from cerebral sources mapped onto the scalp electrodes or from a number of denoised EEG channels. For this purpose, source localization/extraction and noise reduction approaches based on Independent Component Analysis were used prior to correlation analysis. Feature extracted segments were then classified in one of the five classes including WAKE, STAGE1, STAGE2, SWS and REM via an ensemble neuro-fuzzy classifier. Some techniques were employed to improve the classifier’s performance including Scaled Conjugate Gradient Method to speed up learning the ANFIS classifiers, a pruning algorithm to eliminate irrelevant fuzzy rules and the 10-fold cross-validation technique to train and test the system more efficiently. The performance of classification for two strategies including (1) feature extraction from effective cerebral sources and (2) feature extraction from selected channels of denoised EEG signals was compared and contrasted in terms of training errors and test accuracies. The first and second strategies achieved 92.23 and 88.74% agreement with human expert respectively which indicates the effectiveness of the staging system based on cerebral sources of activity. Our results further indicate that the misclassification rates were almost below 10%. The proposed automated sleep staging system is reliable due to the fact that it is based on the underlying dynamics of sleep staging which is less likely to be affected by sleep fragmentations occurred repeatedly in OSA.     

Automatic anesthesia depth staging using entropy measures and relative power of electroencephalogram frequency bands

Many of the surgeries performed under general anesthesia are aided by electroencephalogram (EEG) monitoring. With increased focus on detecting the anesthesia states of patients in the course of surgery, more attention has been paid to analyzing the power spectra and entropy measures of EEG signal during anesthesia. In this paper, by using the relative power of EEG frequency bands and the EEG entropy measures, a new method for detecting the depth of anesthesia states has been presented based on the least squares support vector machine (LS-SVM) classifiers. EEG signals were recorded from 20 patients before, during and after general anesthesia in the operating room at a sampling rate of 200 Hz. Then, 12 features were extracted from each EEG segment, 10 s in length, which are used for anesthesia state monitoring. No significant difference was observed (p?>?0.05) between these features and the bispectral index (BIS), which is the commonly used measure of anesthetic effect. The used LS-SVM classifier based method is able to identify the anesthesia states with an accuracy of 80% with reference to the BIS index. Since the underlying equation of the utilized LS-SVM is linear, the computational time of the algorithm is not significant and therefore it can be used for online application in operation rooms.     

Diagnosis of multiple sclerosis from EEG signals using nonlinear methods

EEG signals have essential and important information about the brain and neural diseases. The main purpose of this study is classifying two groups of healthy volunteers and Multiple Sclerosis (MS) patients using nonlinear features of EEG signals while performing cognitive tasks. EEG signals were recorded when users were doing two different attentional tasks. One of the tasks was based on detecting a desired change in color luminance and the other task was based on detecting a desired change in direction of motion. EEG signals were analyzed in two ways: EEG signals analysis without rhythms decomposition and EEG sub-bands analysis. After recording and preprocessing, time delay embedding method was used for state space reconstruction; embedding parameters were determined for original signals and their sub-bands. Afterwards nonlinear methods were used in feature extraction phase. To reduce the feature dimension, scalar feature selections were done by using T-test and Bhattacharyya criteria. Then, the data were classified using linear support vector machines (SVM) and k-nearest neighbor (KNN) method. The best combination of the criteria and classifiers was determined for each task by comparing performances. For both tasks, the best results were achieved by using T-test criterion and SVM classifier. For the direction-based and the color-luminance-based tasks, maximum classification performances were 93.08 and 79.79% respectively which were reached by using optimal set of features. Our results show that the nonlinear dynamic features of EEG signals seem to be useful and effective in MS diseases diagnosis.     

Data selection in EEG signals classification

The alcoholism can be detected by analyzing electroencephalogram (EEG) signals. However, analyzing multi-channel EEG signals is a challenging task, which often requires complicated calculations and long execution time. This paper proposes three data selection methods to extract representative data from the EEG signals of alcoholics. The methods are the principal component analysis based on graph entropy (PCA-GE), the channel selection based on graph entropy (GE) difference, and the mathematic combinations channel selection, respectively. For comparison purposes, the selected data from the three methods are then classified by three classifiers: the J48 decision tree, the K-nearest neighbor and the Kstar, separately. The experimental results show that the proposed methods are successful in selecting data without compromising the classification accuracy in discriminating the EEG signals from alcoholics and non-alcoholics. Among them, the proposed PCA-GE method uses only 29.69 % of the whole data and 29.5 % of the computation time but achieves a 94.5 % classification accuracy. The channel selection method based on the GE difference also gains a 91.67 % classification accuracy by using only 29.69 % of the full size of the original data. Using as little data as possible without sacrificing the final classification accuracy is useful for online EEG analysis and classification application design.     

Estimation of effective brain connectivity with dual Kalman filter and EEG source localization methods

Effective connectivity is one of the most important considerations in brain functional mapping via EEG. It demonstrates the effects of a particular active brain region on others. In this paper, a new method is proposed which is based on dual Kalman filter. In this method, firstly by using a brain active localization method (standardized low resolution brain electromagnetic tomography) and applying it to EEG signal, active regions are extracted, and appropriate time model (multivariate autoregressive model) is fitted to extracted brain active sources for evaluating the activity and time dependence between sources. Then, dual Kalman filter is used to estimate model parameters or effective connectivity between active regions. The advantage of this method is the estimation of different brain parts activity simultaneously with the calculation of effective connectivity between active regions. By combining dual Kalman filter with brain source localization methods, in addition to the connectivity estimation between parts, source activity is updated during the time. The proposed method performance has been evaluated firstly by applying it to simulated EEG signals with interacting connectivity simulation between active parts. Noisy simulated signals with different signal to noise ratios are used for evaluating method sensitivity to noise and comparing proposed method performance with other methods. Then the method is applied to real signals and the estimation error during a sweeping window is calculated. By comparing proposed method results in different simulation (simulated and real signals), proposed method gives acceptable results with least mean square error in noisy or real conditions.     

An automated microemboli detection and classification system using backscatter RF signals and differential evolution

Embolic phenomena, whether air or particulate emboli, can induce immediate damages like heart attack or ischemic stroke. Embolus composition (gaseous or particulate matter) is vital in predicting clinically significant complications. Embolus detection using Doppler methods have shown their limits to differentiate solid and gaseous embolus. Radio-frequency (RF) ultrasound signals backscattered by the emboli contain additional information on the embolus in comparison to the traditionally used Doppler signals. Gaseous bubbles show a nonlinear behavior under specific conditions of the ultrasound excitation wave, this nonlinear behavior is exploited to differentiate solid from gaseous microemboli. In order to verify the usefulness of RF ultrasound signal processing in the detection and classification of microemboli, an in vitro set-up is developed. Sonovue micro bubbles are exploited to mimic the acoustic behavior of gaseous emboli. They are injected at two different concentrations (0.025 and 0.05 µl/ml) in a nonrecirculating flow phantom containing a tube of 0.8 mm in diameter. The tissue mimicking material surrounding the tube is chosen to imitate the acoustic behavior of solid emboli. Both gaseous and solid emboli are imaged using an Anthares ultrasound scanner with a probe emitting at a transmit frequency of 1.82 MHz and at two mechanical indices (MI) 0.2 and 0.6. We propose in this experimental study to exploit discrete wavelet transform and a dimensionality reduction algorithm based on differential evolution technique in the analysis and the characterization of the backscattered RF ultrasound signals from the emboli. Several features are evaluated from the detail coefficients. It should be noted that the features used in this study are the same used in the paper by Aydin et al. These all features are used as inputs to the classification models without using feature selection method. Then we perform feature selection using differential evolution algorithm with support vector machines classifier. The experimental results show clearly that our proposed method achieves better average classification rates compared to the results obtained in a previous study using also the same backscatter RF signals.     

基于能量法的脑机接口运动想象分类研究

针对脑机接口运动想象脑电信号的分类识别问题,提出了一种基于小波包分解的C3、C4二通道能量特征提取方法。该方法首先采用6阶的巴特沃斯带通滤波对二通道脑电信号进行降噪;然后采用Daubechies类小波函数对其进行5层分解,选择第四层CD4、第五层CD5的小波系数进行重构并抽取其能量特征;最后采用线性距离判别进行分类和使用Kappa系数进行分类衡量。利用BCI2008竞赛的标准数据BCICIV_2b_gdf进行验证,结果表明利用该方法可以较好地反映事件相关同步和事件相关去同步的特征,为BCI研究中事件相关电位的分类识别提供了有效的手段。     

Glossokinetic potential based tongue–machine interface for 1-D extraction

The tongue is an aesthetically useful organ located in the oral cavity. It can move in complex ways with very little fatigue. Many studies on assistive technologies operated by tongue are called tongue–human computer interface or tongue–machine interface (TMI) for paralyzed individuals. However, many of them are obtrusive systems consisting of hardware such as sensors and magnetic tracer placed in the mouth and on the tongue. Hence these approaches could be annoying, aesthetically unappealing and unhygienic. In this study, we aimed to develop a natural and reliable tongue–machine interface using solely glossokinetic potentials via investigation of the success of machine learning algorithms for 1-D tongue-based control or communication on assistive technologies. Glossokinetic potential responses are generated by touching the buccal walls with the tip of the tongue. In this study, eight male and two female naive healthy subjects, aged 22–34 years, participated. Linear discriminant analysis, support vector machine, and the k-nearest neighbor were used as machine learning algorithms. Then the greatest success rate was achieved an accuracy of 99% for the best participant in support vector machine. This study may serve disabled people to control assistive devices in natural, unobtrusive, speedy and reliable manner. Moreover, it is expected that GKP-based TMI could be alternative control and communication channel for traditional electroencephalography (EEG)-based brain–computer interfaces which have significant inadequacies arisen from the EEG signals.     

Human brain wave activity during exposure to radiofrequency field emissions from mobile phones

The aim of this study was to determine whether there is an effect of mobile phone electromagnetic field emissions on the human electroencephalograph (EEG). EEG recordings from ten awake subjects were taken during exposure to radiofrequency (RF) emissions from a mobile phone positioned behind the head. Two experimental trials were conducted. In the first trial, RF exposures were generated by a GSM mobile phone with the speaker disabled and configured to transmit at full-radiated power. During the second trial, exposures were generated by a non-modified GSM mobile phone in active standby mode. For each trial, subjects were exposed in five minute intervals to a randomized, interrupted sequence of five active and five sham exposures. The experiment was conducted under single-blind conditions. The average EEG band power in active exposure recordings was compared to corresponding sham recordings. Statistical tests indicated significant difference in the full-power mode trial within the EEG alpha (8-13 Hz) and beta (13-32 Hz) bands. A subsequent statistical analysis of median spectral power in discrete EEG rhythms revealed significant differences in 7 of the 32 distinct frequencies overall. In conclusion, the results of this study lend support to EEG effects from mobile phones activated in talk-mode.     

Covariate-adjusted region-referenced generalized functional linear model for EEG data

Electroencephalography (EEG) studies produce region-referenced functional data in the form of EEG signals recorded across electrodes on the scalp. It is of clinical interest to relate the highly structured EEG data to scalar outcomes such as diagnostic status. In our motivating study, resting-state EEG is collected on both typically developing (TD) children and children with autism spectrum disorder (ASD) aged 2 to 12 years old. The peak alpha frequency (PAF), defined as the location of a prominent peak in the alpha frequency band of the spectral density, is an important biomarker linked to neurodevelopment and is known to shift from lower to higher frequencies as children age. To retain the most amount of information from the data, we consider the oscillations in the spectral density within the alpha band, rather than just the peak location, as a functional predictor of diagnostic status (TD vs ASD), adjusted for chronological age. A covariate-adjusted region-referenced generalized functional linear model is proposed for modeling scalar outcomes from region-referenced functional predictors, which utilizes a tensor basis formed from one-dimensional discrete and continuous bases to estimate functional effects across a discrete regional domain while simultaneously adjusting for additional nonfunctional covariates, such as age. The proposed methodology provides novel insights into differences in neural development of TD and ASD children. The efficacy of the proposed methodology is investigated through extensive simulation studies.     

Multi-class EEG classification of motor imagery signal by finding optimal time segments and features using SNR-based mutual information

The electroencephalogram signals are used to distinguish different motor imagery tasks in brain–computer interfaces. In most studies, in order to classify the EEG signals recorded in a cue-guided BCI paradigm, time segments for feature extraction after the onset of the visual cue were selected manually. In addition, in these studies the authors have selected a single identical time segment for different subjects. The present study emphasized on the inter-individual variability and difference between different motor imagery tasks as the potential source of erroneous results and used mutual information and the subject specific time interval to overcome this problem. More specifically, a new method was proposed to automatically find the best subject specific time intervals for the classification of four-class motor imagery tasks by using MI between the BCI input and output. Moreover, the signal-to-noise ratio was used to calculate the MI values, while the MI values were used as feature selection criteria to select the discriminative features. The time segments and the best discriminative features were found by using training data and used to assess the evaluation data. Furthermore, the CSP algorithm was used to extract signal features. The dataset 2A of BCI competition IV used in this study consisted of four different motor imagery signals, which were obtained from nine different subjects. One Vs One decomposition scheme was used to deal with the multi-class nature of the problem. The MI values showed that the obtained time segments not only varied between different subjects but also varied between different classifiers of different pair of classes. Finally, the results suggested that the proposed method was efficient in classifying multi-class motor imagery signals as compared to other classification strategies proposed by the other studies.     

A study of decoding human brain activities from simultaneous data of EEG and fMRI using MVPA

Neuroscientists have investigated the functionality of the brain in detail and achieved remarkable results but this area still need further research. Functional magnetic resonance imaging (fMRI) is considered as the most reliable and accurate technique to decode the human brain activity, on the other hand electroencephalography (EEG) is a portable and low cost solution in brain research. The purpose of this study is to find whether EEG can be used to decode the brain activity patterns like fMRI. In fMRI, data from a very specific brain region is enough to decode the brain activity patterns due to the quality of data. On the other hand, EEG can measure the rapid changes in neuronal activity patterns due to its higher temporal resolution i.e., in msec. These rapid changes mostly occur in different brain regions. In this study, multivariate pattern analysis (MVPA) is used both for EEG and fMRI data analysis and the information is extracted from distributed activation patterns of the brain. The significant information among different classes is extracted using two sample t test in both data sets. Finally, the classification analysis is done using the support vector machine. A fair comparison of both data sets is done using the same analysis techniques, moreover simultaneously collected data of EEG and fMRI is used for this comparison. The final analysis is done with the data of eight participants; the average result of all conditions are found which is 65.7% for EEG data set and 64.1% for fMRI data set. It concludes that EEG is capable of doing brain decoding with the data from multiple brain regions. In other words, decoding accuracy with EEG MVPA is as good as fMRI MVPA and is above chance level.     

Interference between mobile phones and pacemakers: a look inside

In this study we analyzed the problem of electromagnetic interference (EMI) between mobile telephones and cardiac pacemakers (PM), by looking at the mechanisms by which the radiated radio frequency (RF) GSM signal may affect the pacemaker function. From a literature review on this topic, we noticed that older pacemakers had a higher rate of being affected by mobile phones when compared to newer ones. This is probably due to the fact that new generation of PM are more protected against electromagnetic field, being equipped with RF feedthrough filters incorporated to the internal PM circuitry. In some experiments conducted by our group, we found that modulated RF signals are somehow demodulated by the PM internal non-linear circuit elements, if no feedthrough assembly is incorporated inside the PM. Such demodulation phenomenon poses a critical problem because digital cellular phones use extremely low-frequency modulation (as low as 2 Hz), that can be mistaken for normal heartbeat. The feedthrough assembly seems instead to prevents the RF signals from accessing the PM enclosure, thus attenuating EMI signals over a broad range of frequencies.     

Fusion of heart rate variability and pulse rate variability for emotion recognition using lagged poincare plots

Designing an efficient automatic emotion recognition system based on physiological signals has attracted great interests within the research of human–machine interactions. This study was aimed to classify emotional responses by means of a simple dynamic signal processing technique and fusion frameworks. The electrocardiogram and finger pulse activity of 35 participants were recorded during rest condition and when subjects were listening to music intended to stimulate certain emotions. Four emotion categories, including happiness, sadness, peacefulness, and fear were chosen. Estimating heart rate variability (HRV) and pulse rate variability (PRV), 4 Poincare indices in 10 lags were extracted. The support vector machine classifier was used for emotion classification. Both feature level (FL) and decision level (DL) fusion schemes were examined. Significant differences have been observed between lag 1 Poincare plot indices and the other lagged measures. The mean accuracies of 84.1, 82.9, 79.68, and 76.05% were obtained for PRV, DL, FL, and HRV measures, respectively. However, DL outperformed others in discriminating sadness and peacefulness, using SD₁ and total features, correspondingly. In both cases, the classification rates improved up to 92% (with the sensitivity of 95% and specificity of 83.33%). Totally, DL resulted in better performances compared to FL. In addition, the impact of the fusion rules on the classification performances has been confirmed.     

Pathway aggregation for survival prediction via multiple kernel learning

Attempts to predict prognosis in cancer patients using high‐dimensional genomic data such as gene expression in tumor tissue can be made difficult by the large number of features and the potential complexity of the relationship between features and the outcome. Integrating prior biological knowledge into risk prediction with such data by grouping genomic features into pathways and networks reduces the dimensionality of the problem and could improve prediction accuracy. Additionally, such knowledge‐based models may be more biologically grounded and interpretable. Prediction could potentially be further improved by allowing for complex nonlinear pathway effects. The kernel machine framework has been proposed as an effective approach for modeling the nonlinear and interactive effects of genes in pathways for both censored and noncensored outcomes. When multiple pathways are under consideration, one may efficiently select informative pathways and aggregate their signals via multiple kernel learning (MKL), which has been proposed for prediction of noncensored outcomes. In this paper, we propose MKL methods for censored survival outcomes. We derive our approach for a general survival modeling framework with a convex objective function and illustrate its application under the Cox proportional hazards and semiparametric accelerated failure time models. Numerical studies demonstrate that the proposed MKL‐based prediction methods work well in finite sample and can potentially outperform models constructed assuming linear effects or ignoring the group knowledge. The methods are illustrated with an application to 2 cancer data sets.     

Multichannel interictal spike activity detection using time–frequency entropy measure

Localization of interictal spikes is an important clinical step in the pre-surgical assessment of pharmacoresistant epileptic patients. The manual selection of interictal spike periods is cumbersome and involves a considerable amount of analysis workload for the physician. The primary focus of this paper is to automate the detection of interictal spikes for clinical applications in epilepsy localization. The epilepsy localization procedure involves detection of spikes in a multichannel EEG epoch. Therefore, a multichannel Time–Frequency (T–F) entropy measure is proposed to extract features related to the interictal spike activity. Least squares support vector machine is used to train the proposed feature to classify the EEG epochs as either normal or interictal spike period. The proposed T–F entropy measure, when validated with epilepsy dataset of 15 patients, shows an interictal spike classification accuracy of 91.20%, sensitivity of 100% and specificity of 84.23%. Moreover, the area under the curve of Receiver Operating Characteristics plot of 0.9339 shows the superior classification performance of the proposed T–F entropy measure. The results of this paper show a good spike detection accuracy without any prior information about the spike morphology.     

Oscillatory spatial profile of alcohol's effects on the resting state: Anatomically-constrained MEG

It has been firmly established that opening and closing the eyes strongly modulate the electro- and magnetoencephalography (EEG and MEG) signals acquired during wakeful rest. Certain features of the resting EEG are altered in chronic alcoholics and their offspring, and have been proposed as biomarkers for alcoholism. Spontaneous brain oscillations are also affected by pharmacological manipulations, but the spectral and spatial characteristics of these changes are not clear. This study examined effects of the eyes-open (EO) and eyes-closed (EC) resting paradigm and alcohol challenge on the spatial profile of spontaneous MEG and EEG oscillations. Whole-head MEG and scalp EEG signals were acquired simultaneously from healthy social drinkers (n = 17) who participated in both alcohol (0.6 g/kg ethanol for men, 0.55 g/kg for women) and placebo conditions in a counterbalanced design. Power of the signal was calculated with Fast Fourier Transform and was decomposed into its constituent theta (4–7 Hz), alpha (8–12 Hz), and beta (15–20 Hz) frequency bands. High-resolution structural MRI images were additionally obtained from all participants and used to constrain distributed minimum norm inverse source power estimates. The spatial estimates of the main generator nodes were in agreement with studies using a combined fMRI-EEG approach. Alpha band oscillations dominated the spectral profile and their source was estimated to the medial parieto-occipital area. Power in theta and beta bands was weaker overall and their sources were estimated to a more focal medial prefrontal area. EO and EC manipulation most strongly modulated power in the alpha band, but a wide-band power increase was observed during the EC condition. Alcohol intoxication increased alpha power, particularly during the EC condition. Application of this methodology to cohorts of chronic alcoholics or individuals at risk could potentially provide insight into the neural basis of oscillatory differences that may be predictive of the vulnerability to alcoholism.