Consensus Matching Pursuit for multi-trial EEG signals

An open-source toolbox for Multi-patient Intracranial EEG Analysis (MIA)

Intracranial EEG (iEEG) performed during the pre-surgical evaluation of refractory epilepsy provides a great opportunity to investigate the neurophysiology of human cognitive functions with exceptional spatial and temporal precisions. A difficulty of the iEEG approach for cognitive neuroscience, however, is the potential variability across patients in the anatomical location of implantations and in the functional responses therein recorded. In this context, we designed, implemented, and tested a user-friendly and efficient open-source toolbox for Multi-Patient Intracranial data Analysis (MIA), which can be used as standalone program or as a Brainstorm plugin. MIA helps analyzing event related iEEG signals while following good scientific practice recommendations, such as building reproducible analysis pipelines and applying robust statistics. The signals can be analyzed in the temporal and time-frequency domains, and the similarity of time courses across patients or contacts can be assessed within anatomical regions. MIA allows visualizing all these results in a variety of formats at every step of the analysis. Here, we present the toolbox architecture and illustrate the different steps and features of the analysis pipeline using a group dataset collected during a language task.

Assessment of Effective Network Connectivity among MEG None Contaminated Epileptic Transitory Events

Characterizing epileptogenic zones EZ (sources responsible of excessive discharges) would assist a neurologist during epilepsy diagnosis. Locating efficiently these abnormal sources among magnetoencephalography (MEG) biomarker is obtained by several inverse problem techniques. These techniques present different assumptions and particular epileptic network connectivity. Here, we proposed to evaluate performances of distributed inverse problem in defining EZ. First, we applied an advanced technique based on Singular Value Decomposition (SVD) to recover only pure transitory activities (interictal epileptiform discharges). We evaluated our technique's robustness in separation between transitory and ripples versus frequency range, transitory shapes, and signal to noise ratio on simulated data (depicting both epileptic biomarkers and respecting time series and spectral properties of realistic data). We validated our technique on MEG signal using detector precision on 5 patients. Then, we applied four methods of inverse problem to define cortical areas and neural generators of excessive discharges. We computed network connectivity of each technique. Then, we confronted obtained noninvasive networks to intracerebral EEG transitory network connectivity using nodes in common, connection strength, distance metrics between concordant nodes of MEG and IEEG, and average propagation delay. Coherent Maximum Entropy on the Mean (cMEM) proved a high matching between MEG network connectivity and IEEG based on distance between active sources, followed by Exact low-resolution brain electromagnetic tomography (eLORETA), Dynamical Statistical Parametric Mapping (dSPM), and Minimum norm estimation (MNE). Clinical performance was interesting for entire methods providing in an average of 73.5% of active sources detected in depth and seen in MEG, and vice versa, about 77.15% of active sources were detected from MEG and seen in IEEG. Investigated problem techniques succeed at least in finding one part of seizure onset zone. dSPM and eLORETA depict the highest connection strength among all techniques. Propagation delay varies in this range [18, 25]ms, knowing that eLORETA ensures the lowest propagation delay (18 ms) and the closet one to IEEG propagation delay.

Extracting seizure onset from surface EEG with independent component analysis: Insights from simultaneous scalp and intracerebral EEG

•

Independent component analysis (ICA) is able to identify seizure generators.

•

Simultaneous long-term scalp-SEEG allows validation of the ICA results.

•

Ability to record seizure onset patterns on scalp depends on generator depth.

The success of stereoelectroencephalographic (SEEG) investigations depends crucially on the hypotheses on the putative location of the seizure onset zone. This information is derived from non-invasive data either based on visual analysis or advanced source localization algorithms. While source localization applied to interictal spikes recorded on scalp is the classical method, it does not provide unequivocal information regarding the seizure onset zone. Raw ictal activity contains a mixture of signals originating from several regions of the brain as well as EMG artifacts, hampering direct input to the source localization algorithms. We therefore introduce a methodology that disentangles the various sources contributing to the scalp ictal activity using independent component analysis and uses equivalent current dipole localization as putative locus of ictal sources. We validated the results of our analysis pipeline by performing long-term simultaneous scalp – intracerebral (SEEG) recordings in 14 patients and analyzing the wavelet coherence between the independent component encoding the ictal discharge and the SEEG signals in 8 patients passing the inclusion criteria. Our results show that invasively recorded ictal onset patterns, including low-voltage fast activity, can be captured by the independent component analysis of scalp EEG. The visibility of the ictal activity strongly depends on the depth of the sources. The equivalent current dipole localization can point to the seizure onset zone (SOZ) with an accuracy that can be as high as 10 mm for superficially located sources, that gradually decreases for deeper seizure generators, averaging at 47 mm in the 8 analyzed patients. Independent component analysis is therefore shown to have a promising SOZ localizing value, indicating whether the seizure onset zone is neocortical, and its approximate location, or located in mesial structures. That may contribute to a better crafting of the hypotheses used as basis of the stereo-EEG implantations.

Multivariate Matching Pursuit Decomposition and Normalized Gabor Entropy for Quantification of Preictal Trends in Epilepsy

Quantification of the complexity of signals recorded concurrently from multivariate systems, such as the brain, plays an important role in the study and characterization of their state and state transitions. Multivariate analysis of the electroencephalographic signals (EEG) over time is conceptually most promising in unveiling the global dynamics of dynamical brain disorders such as epilepsy. We employed a novel methodology to study the global complexity of the epileptic brain en route to seizures. The developed measures of complexity were based on Multivariate Matching Pursuit (MMP) decomposition of signals in terms of time–frequency Gabor functions (atoms) and Shannon entropy. The measures were first validated on simulation data (Lorenz system) and then applied to EEGs from preictal (before seizure onsets) periods, recorded by intracranial electrodes from eight patients with temporal lobe epilepsy and a total of 42 seizures, in search of global trends of complexity before seizures onset. Out of five Gabor measures of complexity we tested, we found that our newly defined measure, the normalized Gabor entropy (NGE), was able to detect statistically significant (p < 0.05) nonlinear trends of the mean global complexity across all patients over 1 h periods prior to seizures’ onset. These trends pointed to a slow decrease of the epileptic brain’s global complexity over time accompanied by an increase of the variance of complexity closer to seizure onsets. These results show that the global complexity of the epileptic brain decreases at least 1 h prior to seizures and imply that the employed methodology and measures could be useful in identifying different brain states, monitoring of seizure susceptibility over time, and potentially in seizure prediction.

Multi-Regional Adaptation in Human Auditory Association Cortex

In auditory cortex, neural responses decrease with stimulus repetition, known as adaptation. Adaptation is thought to facilitate detection of novel sounds and improve perception in noisy environments. Although it is well established that adaptation occurs in primary auditory cortex, it is not known whether adaptation also occurs in higher auditory areas involved in processing complex sounds, such as speech. Resolving this issue is important for understanding the neural bases of adaptation and to avoid potential post-operative deficits after temporal lobe surgery for treatment of focal epilepsy. Intracranial electrocorticographic recordings were acquired simultaneously from electrodes implanted in primary and association auditory areas of the right (non-dominant) temporal lobe in a patient with complex partial seizures originating from the inferior parietal lobe. Simple and complex sounds were presented in a passive oddball paradigm. We measured changes in single-trial high-gamma power (70–150 Hz) and in regional and inter-regional network-level activity indexed by cross-frequency coupling. Repetitive tones elicited the greatest adaptation and corresponding increases in cross-frequency coupling in primary auditory cortex. Conversely, auditory association cortex showed stronger adaptation for complex sounds, including speech. This first report of multi-regional adaptation in human auditory cortex highlights the role of the non-dominant temporal lobe in suppressing neural responses to repetitive background sounds (noise). These results underscore the clinical utility of functional mapping to avoid potential post-operative deficits including increased listening difficulties in noisy, real-world environments.

Conundrums of high-frequency oscillations (80-800 Hz) in the epileptic brain

Pathological high-frequency oscillations (HFOs) (80-800 Hz) are considered biomarkers of epileptogenic tissue, but the underlying complex neuronal events are not well understood. Here, we identify and discuss several outstanding issues or conundrums in regards to the recording, analysis, and interpretation of HFOs in the epileptic brain to critically highlight what is known and what is not about these enigmatic events. High-frequency oscillations reflect a range of neuronal processes contributing to overlapping frequencies from the lower 80 Hz to the very fast spectral frequency bands. Given their complex neuronal nature, HFOs are extremely sensitive to recording conditions and analytical approaches. We provide a list of recommendations that could help to obtain comparable HFO signals in clinical and basic epilepsy research. Adopting basic standards will facilitate data sharing and interpretation that collectively will aid in understanding the role of HFOs in health and disease for translational purpose.

Detecting single-trial EEG evoked potential using a wavelet domain linear mixed model: application to error potentials classification

OBJECTIVE

The main goal of this work is to develop a model for multisensor signals, such as magnetoencephalography or electroencephalography (EEG) signals that account for inter-trial variability, suitable for corresponding binary classification problems. An important constraint is that the model be simple enough to handle small size and unbalanced datasets, as often encountered in BCI-type experiments.

APPROACH

The method involves the linear mixed effects statistical model, wavelet transform, and spatial filtering, and aims at the characterization of localized discriminant features in multisensor signals. After discrete wavelet transform and spatial filtering, a projection onto the relevant wavelet and spatial channels subspaces is used for dimension reduction. The projected signals are then decomposed as the sum of a signal of interest (i.e., discriminant) and background noise, using a very simple Gaussian linear mixed model.

MAIN RESULTS

Thanks to the simplicity of the model, the corresponding parameter estimation problem is simplified. Robust estimates of class-covariance matrices are obtained from small sample sizes and an effective Bayes plug-in classifier is derived. The approach is applied to the detection of error potentials in multichannel EEG data in a very unbalanced situation (detection of rare events). Classification results prove the relevance of the proposed approach in such a context.

SIGNIFICANCE

The combination of the linear mixed model, wavelet transform and spatial filtering for EEG classification is, to the best of our knowledge, an original approach, which is proven to be effective. This paper improves upon earlier results on similar problems, and the three main ingredients all play an important role.

AnyWave: a cross-platform and modular software for visualizing and processing electrophysiological signals

BACKGROUND

The importance of digital signal processing in clinical neurophysiology is growing steadily, involving clinical researchers and methodologists. There is a need for crossing the gap between these communities by providing efficient delivery of newly designed algorithms to end users. We have developed such a tool which both visualizes and processes data and, additionally, acts as a software development platform.

NEW METHOD

AnyWave was designed to run on all common operating systems. It provides access to a variety of data formats and it employs high fidelity visualization techniques. It also allows using external tools as plug-ins, which can be developed in languages including C++, MATLAB and Python.

RESULTS

In the current version, plug-ins allow computation of connectivity graphs (non-linear correlation h2) and time-frequency representation (Morlet wavelets). The software is freely available under the LGPL3 license.

COMPARISON WITH EXISTING METHODS

AnyWave is designed as an open, highly extensible solution, with an architecture that permits rapid delivery of new techniques to end users.

CONCLUSIONS

We have developed AnyWave software as an efficient neurophysiological data visualizer able to integrate state of the art techniques. AnyWave offers an interface well suited to the needs of clinical research and an architecture designed for integrating new tools. We expect this software to strengthen the collaboration between clinical neurophysiologists and researchers in biomedical engineering and signal processing.

Maximum-likelihood estimation of channel-dependent trial-to-trial variability of auditory evoked brain responses in MEG

Background

We propose a mathematical model for multichannel assessment of the trial-to-trial variability of auditory evoked brain responses in magnetoencephalography (MEG).

Methods

Following the work of de Munck et al., our approach is based on the maximum likelihood estimation and involves an approximation of the spatio-temporal covariance of the contaminating background noise by means of the Kronecker product of its spatial and temporal covariance matrices. Extending the work of de Munck et al., where the trial-to-trial variability of the responses was considered identical to all channels, we evaluate it for each individual channel.

Results

Simulations with two equivalent current dipoles (ECDs) with different trial-to-trial variability, one seeded in each of the auditory cortices, were used to study the applicability of the proposed methodology on the sensor level and revealed spatial selectivity of the trial-to-trial estimates. In addition, we simulated a scenario with neighboring ECDs, to show limitations of the method. We also present an illustrative example of the application of this methodology to real MEG data taken from an auditory experimental paradigm, where we found hemispheric lateralization of the habituation effect to multiple stimulus presentation.

Conclusions

The proposed algorithm is capable of reconstructing lateralization effects of the trial-to-trial variability of evoked responses, i.e. when an ECD of only one hemisphere habituates, whereas the activity of the other hemisphere is not subject to habituation. Hence, it may be a useful tool in paradigms that assume lateralization effects, like, e.g., those involving language processing.

Multivariate matching pursuit in optimal Gabor dictionaries: theory and software with interface for EEG/MEG via Svarog

BACKGROUND

Matching pursuit algorithm (MP), especially with recent multivariate extensions, offers unique advantages in analysis of EEG and MEG.

METHODS

We propose a novel construction of an optimal Gabor dictionary, based upon the metrics introduced in this paper. We implement this construction in a freely available software for MP decomposition of multivariate time series, with a user friendly interface via the Svarog package (Signal Viewer, Analyzer and Recorder On GPL, http://braintech.pl/svarog), and provide a hands-on introduction to its application to EEG. Finally, we describe numerical and mathematical optimizations used in this implementation.

RESULTS

Optimal Gabor dictionaries, based on the metric introduced in this paper, for the first time allowed for a priori assessment of maximum one-step error of the MP algorithm. Variants of multivariate MP, implemented in the accompanying software, are organized according to the mathematical properties of the algorithms, relevant in the light of EEG/MEG analysis. Some of these variants have been successfully applied to both multichannel and multitrial EEG and MEG in previous studies, improving preprocessing for EEG/MEG inverse solutions and parameterization of evoked potentials in single trials; we mention also ongoing work and possible novel applications.

CONCLUSIONS

Mathematical results presented in this paper improve our understanding of the basics of the MP algorithm. Simple introduction of its properties and advantages, together with the accompanying stable and user-friendly Open Source software package, pave the way for a widespread and reproducible analysis of multivariate EEG and MEG time series and novel applications, while retaining a high degree of compatibility with the traditional, visual analysis of EEG.

Stationarity stopping criterion for matching pursuit-framework and encephalographic illustration

We present a new stopping criterion for the matching pursuit (MP) algorithm, based on evaluating stationarity of the residua of the consecutive MP iterations. The new stopping criterion is based on a model of a nonstationary signal, which assumes that the part of the signal that is of interest is nonstationary and contaminated by a weakly stationary noise. Mean- and variance-stationarity of the residua obtained from each step of MP is evaluated by means of dedicated statistical tests-the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test and the White test, respectively. We illustrate the proposed concept by an example in which we analyse magnetoencephalographic (MEG) data.

A comparison of methods for separation of transient and oscillatory signals in EEG

Brain oscillations constitute a prominent feature of electroencephalography (EEG), in both physiological and pathological states. An efficient separation of oscillation from transient signals in EEG is important not only for detection of oscillations, but also for advanced signal processing such as source localization. A major difficulty lies in the fact that filtering transient phenomena can lead to spurious oscillatory activity. Therefore, in the presence of a mixture of transient and oscillatory events, it is not clear to which extent filtering methods are able to separate them efficiently. The objective of this study was to evaluate methods for separating oscillations from transients. We compared three methods: finite impulse response (FIR) filtering, wavelet analysis with stationary wavelet transform (SWT), time-frequency sparse decomposition with Matching Pursuit (MP). We evaluated the quality of reconstruction and the results of automatic detection of oscillations intermingled with transients. The emphasis of our study was on epileptic signals and single channel processing. In both simulations and on real data, FIR performed generally worse than the time-frequency methods. Both SWT and MP showed good results in separation and detection, each method having its advantages and its limitations. The SWT had good results in separation and detection of transients due to the time invariance property, but still did not completely resolve the frequency overlap for the oscillation during the time-frequency thresholding. The MP provides a sparse representation, and gave good results for simulated data. However, in the real data, we observed distortions introduced by the subtractive approach, and departure from dictionary waveforms. Future directions are proposed for overcoming these limitations.

A comparison of methods for assessing alpha phase resetting in electrophysiology, with application to intracerebral EEG in visual areas

There are two competing views on the mechanisms underlying the generation of visual evoked potentials/fields in EEG/MEG. The classical hypothesis assumes an additive wave on top of background noise. Another hypothesis states that the evoked activity can totally or partially arise from a phase resetting of the ongoing alpha rhythm. There is no consensus however, on the best tools for distinguishing between these two hypotheses. In this study, we have tested different measures on a large series of simulations under a variety of scenarios, involving in particular trial-to-trial variability and different dynamics of ongoing alpha rhythm. No single measure or set of measures was found to be necessary or sufficient for defining phase resetting in the context of our simulations. Still, simulations permitted to define criteria that were the most reliable in practice for distinguishing additive and phase resetting hypotheses. We have then applied these criteria on intracerebral EEG data recordings in the visual areas during a visual discrimination task. We investigated the intracerebral channels that presented both ERP and ongoing alpha oscillations (n=37). Within these channels, a total of 30% fulfilled phase resetting criteria during the generation of the visual evoked potential, based on criteria derived from simulations. Moreover, 19% of the 37 channels presented dependence of the ERP on the level of pre-stimulus alpha. Only 5% of channels fulfilled both the simulation-related criteria and dependence on baseline alpha level. Our simulation study points out to the difficulty of clearly assessing phase resetting based on observed macroscopic electrophysiological signals. Still, some channels presented an indication of phase resetting in the context of our simulations. This needs to be confirmed by further work, in particular at a smaller recording scale.

Graph-based variability estimation in single-trial event-related neural responses

Extracting information from multitrial magnetoencephalography or electroencephalography (EEG) recordings is challenging because of the very low SNR, and because of the inherent variability of brain responses. The problem of low SNR is commonly tackled by averaging multiple repetitions of the recordings, also called trials, but the variability of response across trials leads to biased results and limits interpretability. This paper proposes to decode the variability of neural responses by making use of graph representations. Our approach has several advantages compared to other existing methods that process single-trial data: first, it avoids the a priori definition of a model for the waveform of the neural response; second, it does not make use of the average data for parameter estimation; third, it does not suffer from initialization problems by providing solutions that are global optimum of cost functions; and last, it is fast. We proceed in two steps. First, a manifold learning algorithm, based on a graph Laplacian, offers an efficient way of ordering trials with respect to the response variability, under the condition that this variability itself depends on a single parameter. Second, the estimation of the variability is formulated as a combinatorial optimization that can be solved very efficiently using graph cuts. Details and validation of this second step are provided for latency estimation. Performance and robustness experiments are conducted on synthetic data, and results are presented on EEG data from a P300 oddball experiment.

Time-Frequency Data Reduction for Event Related Potentials: Combining Principal Component Analysis and Matching Pursuit

Joint time-frequency representations offer a rich representation of event related potentials (ERPs) that cannot be obtained through individual time or frequency domain analysis. This representation, however, comes at the expense of increased data volume and the difficulty of interpreting the resulting representations. Therefore, methods that can reduce the large amount of time-frequency data to experimentally relevant components are essential. In this paper, we present a method that reduces the large volume of ERP time-frequency data into a few significant time-frequency parameters. The proposed method is based on applying the widely-used matching pursuit (MP) approach, with a Gabor dictionary, to principal components extracted from the time-frequency domain. The proposed PCA-Gabor decomposition is compared with other time-frequency data reduction methods such as the time-frequency PCA approach alone and standard matching pursuit methods using a Gabor dictionary for both simulated and biological data. The results show that the proposed PCA-Gabor approach performs better than either the PCA alone or the standard MP data reduction methods, by using the smallest amount of ERP data variance to produce the strongest statistical separation between experimental conditions.

Pitfalls of high-pass filtering for detecting epileptic oscillations: a technical note on "false" ripples

OBJECTIVES

To analyze interictal High frequency oscillations (HFOs) as observed in the medial temporal lobe of epileptic patients and animals (ripples, 80-200Hz and fast ripples, 250-600Hz). To show that the identification of interictal HFOs raises some methodological issues, as the filtering of sharp transients (e.g., epileptic spikes or artefacts) or signals with harmonics can result in "false" ripples. To illustrate and quantify the occurrence of false ripples on filtered EEG traces.

METHODS

We have performed high-pass filtering on both simulated and real data. We have also used two alternate methods: time-frequency analysis and matching pursuit.

RESULTS

Two types of events were shown to produce oscillations after filtering that could be confounded with actual oscillatory activity: sharp transients and harmonics of non-sinusoidal signals.

CONCLUSIONS

High-pass filtering of EEG traces for detection of oscillatory activity should be performed with great care. Filtered traces should be compared to original traces for verification of presence of transients. Additional techniques such as time-frequency transforms or sparse decompositions are highly beneficial.

SIGNIFICANCE

Our study draws the attention on an issue of great importance in the marking of HFOs on EEG traces. We illustrate complementary methods that can help both researchers and clinicians.