Electroencephalographic Time Series Research Articles

Training Datasets for Epilepsy Analysis: Preprocessing and Feature Extraction from Electroencephalography Time Series

We describe 20 datasets derived through signal filtering and feature extraction steps applied to the raw time series EEG data of 20 epileptic patients, as well as the methods we used to derive them. Background: Epilepsy is a complex neurological disorder which has seizures as its hallmark. Electroencephalography plays a crucial role in epilepsy assessment, offering insights into the brain’s electrical activity and advancing our understanding of seizures. The availability of tagged training sets covering all seizure phases—inter-ictal, pre-ictal, ictal, and post-ictal—is crucial for data-driven epilepsy analyses. Methods: Using the sliding window technique with a two-second window length and a one-second time slip, we extract multiple features from the preprocessed EEG time series of 20 patients from the Freiburg Seizure Prediction Database. In addition, we assign a class label to each instance to specify its corresponding seizure phase. All these operations are made through a software application we developed, which is named Training Builder. Results: The 20 tagged training datasets each contain 1080 univariate and bivariate features, and are openly and publicly available. Conclusions: The datasets support the training of data-driven models for seizure detection, prediction, and clustering, based on features engineering.

Methodological Issues in Evaluating Machine Learning Models for EEG Seizure Prediction: Good Cross-Validation Accuracy Does Not Guarantee Generalization to New Patients

There is an increasing interest in applying artificial intelligence techniques to forecast epileptic seizures. In particular, machine learning algorithms could extract nonlinear statistical regularities from electroencephalographic (EEG) time series that can anticipate abnormal brain activity. The recent literature reports promising results in seizure detection and prediction tasks using machine and deep learning methods. However, performance evaluation is often based on questionable randomized cross-validation schemes, which can introduce correlated signals (e.g., EEG data recorded from the same patient during nearby periods of the day) into the partitioning of training and test sets. The present study demonstrates that the use of more stringent evaluation strategies, such as those based on leave-one-patient-out partitioning, leads to a drop in accuracy from about 80% to 50% for a standard eXtreme Gradient Boosting (XGBoost) classifier on two different data sets. Our findings suggest that the definition of rigorous evaluation protocols is crucial to ensure the generalizability of predictive models before proceeding to clinical trials.

Fractal Structure of Brain Electrical Activity of Patients With Mental Disorders.

This work was aimed at a comparative analysis of the degree of multifractality of electroencephalographic time series obtained from a group of healthy subjects and from patients with mental disorders. We analyzed long-term records of patients with paranoid schizophrenia and patients with depression. To evaluate the properties of multifractal scaling of various electroencephalographic time series, the method of maximum modulus of the wavelet transform and multifractal analysis of fluctuations without a trend were used. The stability of the width and position of the singularity spectrum for each of the test groups was revealed, and a relationship was established between the correlation and anticorrelation dynamics of successive values of the electroencephalographic time series and the type of mental disorders. It was shown that the main differences between the multifractal properties of brain activity in normal and pathological conditions lie in the different width of the multifractality spectrum and its location associated with the correlated or anticorrelated dynamics of the values of successive time series. It was found that the schizophrenia group is characterized by a greater degree of multifractality compared to the depression group. Thus, the degree of multifractality can be included in a set of tests for differential diagnosis and research of mental disorders.

On the meaning of Hurst entropy applied to EEG data series

The Hurst exponent estimates the degree of self-similarity and predictability of a time series, which, under this nonlinear statistical model, can adopt two opposing tendencies with respect to the way these data series are mobilized over time. Persistent and anti-persistent series are thus described, depending on whether the Hurst value describing them is greater or less than 0.5 in a range of 0 to 1. Those series that show "Hurst effect" (persistent series with H>0.5) are potentially predictable in the short or medium term, which makes them especially interesting for any predictive interest in science or economics. In the brain, when the oscillatory electrical activity of an EEG is analyzed, the time series that make up a complete recording include a variable number of sources, electrodes or channels, which detect the electrical activity of millions of pyramidal neurons at the level of the cerebral cortex. When this signal is decomposed into its frequency components, its spectrum ranges from the slowest waves, or Delta (0.1-4Hz), to the fastest, Gamma (>30 Hz). In the intermediate range, the Theta (4-8Hz); Alpha (8-12Hz); and Beta (13-30Hz) oscillations complete the range of the EEG spectrum. When estimating the Hurst exponent by filtered intervals of the total frequency range, only for the Delta band, the H values exceed H=0.5. The Hurst values for Delta were close to H = 0.85 and all the rest of the spectrum falls to values less than 0.5, in a range from 1.5<H<3.5.In the present work, we extend the interpretation of the Hurst exponent applied to electroencephalographic time series, and we propose to pay special attention to the so-called anti-persistent series (H<0.5) which, overshadowed by the predictive promises of the Hurst effect of persistent series, we believe may hide the very long-term persistence of complex homeostatic processes, stabilized in the human brain over millions of years of evolution.

Radius selection using kernel density estimation for the computation of nonlinear measures.

When nonlinear measures are estimated from sampled temporal signals with finite-length, a radius parameter must be carefully selected to avoid a poor estimation. These measures are generally derived from the correlation integral, which quantifies the probability of finding neighbors, i.e., pair of points spaced by less than the radius parameter. While each nonlinear measure comes with several specific empirical rules to select a radius value, we provide a systematic selection method. We show that the optimal radius for nonlinear measures can be approximated by the optimal bandwidth of a Kernel Density Estimator (KDE) related to the correlation sum. The KDE framework provides non-parametric tools to approximate a density function from finite samples (e.g., histograms) and optimal methods to select a smoothing parameter, the bandwidth (e.g., bin width in histograms). We use results from KDE to derive a closed-form expression for the optimal radius. The latter is used to compute the correlation dimension and to construct recurrence plots yielding an estimate of Kolmogorov-Sinai entropy. We assess our method through numerical experiments on signals generated by nonlinear systems and experimental electroencephalographic time series.

Driver sleepiness detection with deep neural networks using electrophysiological data

Objective. The objective of this paper is to present a driver sleepiness detection model based on electrophysiological data and a neural network consisting of convolutional neural networks and a long short-term memory architecture. Approach. The model was developed and evaluated on data from 12 different experiments with 269 drivers and 1187 driving sessions during daytime (low sleepiness condition) and night-time (high sleepiness condition), collected during naturalistic driving conditions on real roads in Sweden or in an advanced moving-base driving simulator. Electrooculographic and electroencephalographic time series data, split up in 16 634 2.5 min data segments was used as input to the deep neural network. This probably constitutes the largest labeled driver sleepiness dataset in the world. The model outputs a binary decision as alert (defined as ≤6 on the Karolinska Sleepiness Scale, KSS) or sleepy (KSS ≥ 8) or a regression output corresponding to KSS ϵ [1–5, 6, 7, 8, 9]. Main results. The subject-independent mean absolute error (MAE) was 0.78. Binary classification accuracy for the regression model was 82.6% as compared to 82.0% for a model that was trained specifically for the binary classification task. Data from the eyes were more informative than data from the brain. A combined input improved performance for some models, but the gain was very limited. Significance. Improved classification results were achieved with the regression model compared to the classification model. This suggests that the implicit order of the KSS ratings, i.e. the progression from alert to sleepy, provides important information for robust modelling of driver sleepiness, and that class labels should not simply be aggregated into an alert and a sleepy class. Furthermore, the model consistently showed better results than a model trained on manually extracted features based on expert knowledge, indicating that the model can detect sleepiness that is not covered by traditional algorithms.

Transition-based complexity-entropy causality diagram: A novel method to characterize complex systems

Complexity-entropy causality plane (CECP) and ordinal transition network (OTN) are both crucial tools to reveal the characteristics of time series and distinguish complex systems. However, when the parameters of the system to be distinguished have a wide range of values, the distinguishing function of CECP is weakened. Therefore, we propose a new measure called transition Fisher information (TFI) based on the probability transition matrix in OTN. The TFI is combined with conditional entropy of ordinal patterns and complexity measure to form a novel three-dimensional graph, called transition-based complexity-entropy causality diagram(TB-CECD). These three statistics depict the complex system from different angles. Through simulation experiments, we prove that even if the parameters of complex systems are wide-ranging, the systems of different properties can be assigned to different areas of the graph. Moreover, we find that the trace of the transition probability matrix can be seen as a function of time delay and used to reflect the periodic information of the system. For applications, the proposed methods are applied to vehicle dynamic response data to diagnose periodic short-wave defects such as rail corrugation. The financial time series and Electroencephalographic (EEG) time series are also researched.

Higher hypnotic suggestibility is associated with the lower EEG signal variability in theta, alpha, and beta frequency bands.

Variation of information in the firing rate of neural population, as reflected in different frequency bands of electroencephalographic (EEG) time series, provides direct evidence for change in neural responses of the brain to hypnotic suggestibility. However, realization of an effective biomarker for spiking behaviour of neural population proves to be an elusive subject matter with its impact evident in highly contrasting results in the literature. In this article, we took an information-theoretic stance on analysis of the EEG time series of the brain activity during hypnotic suggestions, thereby capturing the variability in pattern of brain neural activity in terms of its information content. For this purpose, we utilized differential entropy (DE, i.e., the average information content in a continuous time series) of theta, alpha, and beta frequency bands of fourteen-channel EEG time series recordings that pertain to the brain neural responses of twelve carefully selected high and low hypnotically suggestible individuals. Our results show that the higher hypnotic suggestibility is associated with a significantly lower variability in information content of theta, alpha, and beta frequencies. Moreover, they indicate that such a lower variability is accompanied by a significantly higher functional connectivity (FC, a measure of spatiotemporal synchronization) in the parietal and the parieto-occipital regions in the case of theta and alpha frequency bands and a non-significantly lower FC in the central region's beta frequency band. Our results contribute to the field in two ways. First, they identify the applicability of DE as a unifying measure to reproduce the similar observations that are separately reported through adaptation of different hypnotic biomarkers in the literature. Second, they extend these previous findings that were based on neutral hypnosis (i.e., a hypnotic procedure that involves no specific suggestions other than those for becoming hypnotized) to the case of hypnotic suggestions, thereby identifying their presence as a potential signature of hypnotic experience.

Long-range linear correlations and nonlinear chaos estimation differentially characterizes functional connectivity and organization of the Brain EEG

We obtained and compared linear and nonlinear estimation of synchronicity and internal organization of a set of electroencephalographic (EEG) time series recorded during the realization of cognitive tasks. Linear estimation consisted of the assessment of Spearman R correlations between pairs of EEG electrodes, andnonlinear estimation consisted in apply rescaled (R/S) analysis on the EEG signal to obtain the Hurst exponent which estimates the degree of predictability (order) of the signal in an unpredictable (chaotic) ongoing background activity. We found adifferential characterization of the coupled synchronicity between pairs of scalp electrodes and the estimation of chaos/no-chaos global balance of the signal. While H increases indicating a tendency towards persistent self-similarity (affinity) and self-organizing processes, the amount of pairs of electrodes highly correlated (R>0.85) diminish, suggesting a sort of counteracting dynamic of balancebetween spontaneously driven homogenizing long-range linear correlation tendency againstan inner brain mechanisms leading to persistent self-organization, information richness and heterogeneity.

Two step transfer entropy – An estimator of delayed directional couplings between multivariate EEG time series

Quantifying delayed directional couplings between electroencephalographic (EEG) time series requires an efficient method of causal network inference. This is especially due to the limited knowledge about the underlying dynamics of the brain activity. Recent methods based on information theoretic measures such as Transfer Entropy (TE) made significant progress on this issue by providing a model-free framework for causality detection. However, TE estimation from observed data is not a trivial task, especially when the number of variables is large which is the case in a highly complex system like human brain. Here we propose a computationally efficient procedure for TE estimation based on using sets of the Most Informative Variables that effectively contribute to resolving the uncertainty of the destination. In the first step of this method, some conditioning sets are determined through a nonlinear state space reconstruction; then in the second step, optimal estimation of TE is done based on these sets. Validation of the proposed method using synthetic data and neurophysiological signals demonstrates computational efficiency in quantifying delayed directional couplings compared with the common TE analysis.

The relationship between somatic and cognitive-affective depression symptoms and error-related ERPs

BackgroundThe symptoms that contribute to the clinical diagnosis of depression likely emerge from, or are related to, underlying cognitive deficits. To understand this relationship further, we examined the relationship between self-reported somatic and cognitive-affective Beck׳s Depression Inventory-II (BDI-II) symptoms and aspects of cognitive control reflected in error event-related potential (ERP) responses. MethodsTask and assessment data were analyzed within 51 individuals. The group contained a broad distribution of depressive symptoms, as assessed by BDI-II scores. ERPs were collected following error responses within a go/no-go task. Individual error ERP amplitudes were estimated by conducting group independent component analysis (ICA) on the electroencephalographic (EEG) time series and analyzing the individual reconstructed source epochs. Source error amplitudes were correlated with the subset of BDI-II scores representing somatic and cognitive-affective symptoms. ResultsWe demonstrate a negative relationship between somatic depression symptoms (i.e. fatigue or loss of energy) (after regressing out cognitive-affective scores, age and IQ) and the central–parietal ERP response that peaks at 359ms. The peak amplitudes within this ERP response were not significantly related to cognitive-affective symptom severity (after regressing out the somatic symptom scores, age, and IQ). LimitationsThese findings were obtained within a population of female adults from a maximum-security correctional facility. Thus, additional research is required to verify that they generalize to the broad population. ConclusionsThese results suggest that individuals with greater somatic depression symptoms demonstrate a reduced awareness of behavioral errors, and help clarify the relationship between clinical measures of self-reported depression symptoms and cognitive control.

Explicit Recursive and Adaptive Filtering in Reproducing Kernel Hilbert Spaces

This brief presents a methodology to develop recursive filters in reproducing kernel Hilbert spaces. Unlike previous approaches that exploit the kernel trick on filtered and then mapped samples, we explicitly define the model recursivity in the Hilbert space. For that, we exploit some properties of functional analysis and recursive computation of dot products without the need of preimaging or a training dataset. We illustrate the feasibility of the methodology in the particular case of the γ-filter, which is an infinite impulse response filter with controlled stability and memory depth. Different algorithmic formulations emerge from the signal model. Experiments in chaotic and electroencephalographic time series prediction, complex nonlinear system identification, and adaptive antenna array processing demonstrate the potential of the approach for scenarios where recursivity and nonlinearity have to be readily combined.

A general framework for time series data mining based on event analysis: Application to the medical domains of electroencephalography and stabilometry

There are now domains where information is recorded over a period of time, leading to sequences of data known as time series. In many domains, like medicine, time series analysis requires to focus on certain regions of interest, known as events, rather than analyzing the whole time series.In this paper, we propose a framework for knowledge discovery in both one-dimensional and multidimensional time series containing events. We show how our approach can be used to classify medical time series by means of a process that identifies events in time series, generates time series reference models of representative events and compares two time series by analyzing the events they have in common.We have applied our framework on time series generated in the areas of electroencephalography (EEG) and stabilometry. Framework performance was evaluated in terms of classification accuracy, and the results confirmed that the proposed schema has potential for classifying EEG and stabilometric signals.The proposed framework is useful for discovering knowledge from medical time series containing events, such as stabilometric and electroencephalographic time series. These results would be equally applicable to other medical domains generating iconographic time series, such as, for example, electrocardiography (ECG).

Rejection of pulse related artefact (PRA) from continuous electroencephalographic (EEG) time series recorded during functional magnetic resonance imaging (fMRI) using constraint independent component analysis (cICA)

Rejection of the pulse related artefact (PRA) from electroencephalographic (EEG) time series recorded simultaneously with fMRI data is difficult, particularly during NREM sleep because of the similarities between sleep slow waves and PRA, in both temporal and frequency domains and the need to work with non-averaged data. Here we introduce an algorithm based on constrained independent component analysis (cICA) for PRA removal. This method has several advantages: (1) automatic detection of the components corresponding to the PRA; (2) stability of the solution and (3) computational treatability. Using multichannel EEG recordings obtained in a 3 T MR scanner, with and without concomitant fMRI acquisition, we provide evidence for the sensitivity and specificity of the method in rejecting PRA in various sleep and waking conditions.

Cerebral Plasticity After Subcortical Stroke as Revealed by Cortico-Muscular Coherence

The latency estimation of cortico-muscular coherence (CMCoh) could provide valuable information, especially for the pathological study. However, the conduction time from the central cortical rhythm to peripheral oscillations has not been explored for stroke patients. In this study one recently proposed method, maximizing coherence, was applied into the coherence analysis to estimate the latency by which the extensor carpi radialis electromyographic signals lagged behind the electroencephalographic time series with seven subcortical stroke subjects. Significantly prolonged conduction time was found in affected sides compared with the unaffected sides. The interhemispheric spatial displacement was also calculated using electrodes projection optimization and spherical surface laplacian. The results showed that the CMCoh could help investigate the cerebral reorganization after stroke.

Estimating nonlinear interdependences in dynamical systems using cellular nonlinear networks

We propose a method for estimating nonlinear interdependences between time series using cellular nonlinear networks. Our approach is based on the nonlinear dynamics of interacting nonlinear elements. We apply it to time series of coupled nonlinear model systems and to electroencephalographic time series from an epilepsy patient, and we show that an accurate approximation of symmetric and asymmetric realizations of a nonlinear interdependence measure can be achieved, thus allowing one to detect the strength and direction of couplings.

ROBUST TRAINING AND CONTROL STRATEGIES FOR THE GRASP TYPE SELECTION OF HAND PROSTHESES

Man/machine interfaces are often designed on patient training data (e.g. electromyographic or electroencephalographic time series) from one session. The design process adapts parameters for signal processing and classification. The automatically adapted classification routine delivers good results on this data set but the man/machine interface might show a lack of classification accuracy (robustness) at following sessions and at activities of daily living. This article discusses the underlying effects and presents a method for robust design. A comparison with a common design delivers conclusions about the accuracy of validation techniques. The new methods are applied to electromyographic patient data for the control of a hand prosthesis.

Estimating phase synchronization in dynamical systems using cellular nonlinear networks

We propose a method for estimating phase synchronization between time series using the parallel computing architecture of cellular nonlinear networks (CNN's). Applying this method to time series of coupled nonlinear model systems and to electroencephalographic time series from epilepsy patients, we show that an accurate approximation of the mean phase coherence R--a bivariate measure for phase synchronization--can be achieved with CNN's using polynomial-type templates.

20.24 A new appraisal of the biomechanics of gait forimproved rehabilitation outcomes

There are now domains where information is recorded over a period of time, leading to sequences of data known as time series. In many domains, like medicine, time series analysis requires to focus on certain regions of interest, known as events, rather than analyzing the whole time series.In this paper, we propose a framework for knowledge discovery in both one-dimensional and multidimensional time series containing events. We show how our approach can be used to classify medical time series by means of a process that identifies events in time series, generates time series reference models of representative events and compares two time series by analyzing the events they have in common.We have applied our framework on time series generated in the areas of electroencephalography (EEG) and stabilometry. Framework performance was evaluated in terms of classification accuracy, and the results confirmed that the proposed schema has potential for classifying EEG and stabilometric signals.The proposed framework is useful for discovering knowledge from medical time series containing events, such as stabilometric and electroencephalographic time series. These results would be equally applicable to other medical domains generating iconographic time series, such as, for example, electrocardiography (ECG).

Multichannel matching pursuit and EEG inverse solutions

We present a new approach to the preprocessing of the electroencephalographic time series for EEG inverse solutions. As the first step, EEG recordings are decomposed by multichannel matching pursuit algorithm—in this study we introduce a computationally efficient, suboptimal solution. Then, based upon the parameters of the waveforms fitted to the EEG (frequency, amplitude and duration), we choose those corresponding to the the phenomena of interest, like e.g. sleep spindles. For each structure, the corresponding weights of each channel define a topographic signature, which can be subject to an inverse solution procedure, like e.g. Loreta, used in this work. As an example, we present an automatic detection and parameterization of sleep spindles, appearing in overnight polysomnographic recordings. Inverse solutions obtained for single sleep spindles are coherent with the averages obtained for 20 overnight EEG recordings analyzed in this study, as well as with the results reported previously in literature as inter-subject averages of solutions for spectral integrals, computed on visually selected spindles.