Search

Epileptic seizure detection and prediction by using noninvasive measurements such as scalp EEG signals or invasive, intracranial recordings, has been at the heart of epilepsy studies for at least three decades. To this end, the most common approach has been to consider short‐length recordings (several seconds to a few minutes) around a seizure, aiming to identify significant changes that occur before or during seizures. An inherent assumption in this approach is the presence of a relatively constant EEG activity in the interictal period, which is interrupted by seizure occurrence. Here, we examine this assumption by using long‐duration scalp EEG data (21–94 hr) in nine patients with epilepsy, based on which we construct functional brain networks. Our results reveal that these networks vary over time in a periodic fashion, exhibiting multiple peaks at periods ranging between 1 and 24 hr. The effects of seizure onset on the functional brain network properties were found to be considerably smaller in magnitude compared to the changes due to these inherent periodic cycles. Importantly, the properties of the identified network periodic components (instantaneous phase) were found to be strongly correlated to seizure onset, especially for the periodicities around 3 and 5 hr. These correlations were found to be largely absent between EEG signal periodicities and seizure onset, suggesting that higher specificity may be achieved by using network‐based metrics. In turn, this implies that more robust seizure detection and prediction can be achieved if longer term underlying functional brain network periodic variations are taken into account.

Gamma oscillations are thought to play a key role in neuronal network function and neuronal communication, yet the underlying generating mechanisms have not been fully elucidated to date. At least partly, this may be due to the fact that even in simple network models of interconnected inhibitory (I) and excitatory (E) neurons, many parameters remain unknown and are set based on practical considerations or by convention. Here, we mitigate this problem by requiring PING (Pyramidal Interneuron Network Gamma) models to simultaneously satisfy a broad set of criteria for realistic behaviour based on empirical data spanning both the single unit (spikes) and local population (LFP) levels while unknown parameters are varied. By doing so, we were able to constrain the parameter ranges and select empirically valid models. The derived model constraints implied weak rather than strong PING as the generating mechanism for gamma, connectivity between E and I neurons within specific bounds, and variations of the external input to E but not I neurons. Constrained models showed valid behaviours, including gamma frequency increases with contrast and power saturation or decay at high contrasts. Using an empirically-validated model we studied the route to gamma instability at high contrasts. This involved increased heterogeneity of E neurons with increasing input triggering a breakdown of I neuron pacemaker function. Further, we illustrate the model's capacity to resolve disputes in the literature concerning gamma oscillation properties and GABA conductance proxies. We propose that the models derived in our study will be useful for other modelling studies, and that our approach to the empirical constraining of PING models can be expanded when richer empirical datasets become available. As local gamma networks are the building blocks of larger networks that aim to understand complex cognition through their interactions, there is considerable value in improving our models of these building blocks.

The task of automated epileptic seizure detection and prediction by using non-invasive measurements such as scalp EEG signals or invasive, intracranial recordings, has been at the heart of epilepsy studies for at least three decades. By far, the most common approach for tackling this problem is to examine short-length recordings around the occurrence of a seizure - normally ranging between several seconds and up to a few minutes before and after the epileptic event - and identify any significant changes that occur before or during the event. An inherent assumption in these studies is the presence of a relatively constant EEG activity in the interictal period, which is presumably interrupted by the occurrence of a seizure. Here, we examine this assumption by using long-duration scalp EEG data (ranging between 21 and 94 hours) in patients with epilepsy, based on which we construct functional brain networks. Our results suggest that not only these networks vary over time, but they do so in a periodic fashion, exhibiting multiple peaks at periods ranging between around one and 24 hours. The effects of seizure onset on the functional brain network properties were found to be considerably smaller in magnitude compared to the changes due to the inherent periodic cycles of these networks. Importantly, the properties of the identified network periodic components (instantaneous phase, particularly that of short-term periodicities around 3 and 5 h) were found to be strongly correlated to seizure onset. These correlations were found to be largely absent between EEG signal periodicities and seizure onset, suggesting that higher specificity may be achieved by using network-based metrics. In turn, this suggests that to achieve more robust seizure detection and/or prediction, the evolution of the underlying longer term functional brain network periodic variations should be taken into account.Highlights- We have examined the long-term characteristics of EEG functional brain networks and their correlations to seizure onset- We show periodicities over multiple time scales in network summative properties (degree, efficiency, clustering coefficient)- We also show that, in addition to average network properties, similar periodicities exist in network topology using a novel measure based on the graph edit distance, suggesting that specific connectivity patterns recur over time- These periodic patterns were preserved when we corrected for the effects of volume conduction and were found to be of much larger magnitude compared to seizure-induced modulations- For the first time to our knowledge, we demonstrate that seizure onset occurs preferentially at specific phases of network periodic components that were consistently observed across subjects, particularly for shorter periodicities (around 3 and 5 hours)- These correlations between the phase of network periodic components and seizure onset were nearly absent when examining univariate properties (EEG signal power), suggesting that network-based measures are more tightly coupled with seizure onset compared to EEG signal-based measures- Our findings suggest that seizure detection and prediction algorithms may benefit significantly by taking into account longer-term variations in brain network properties- As we show strong evidence that shorter network-based periodicities (3-5 hours) are tightly coupled with seizure onset, our results pave the way for further investigation into the pathophysiology of seizure generation mechanisms beyond the well-known effects of circadian rhythms

We investigated the correlation of epileptic seizure onset times with long term EEG functional brain network properties. To do so, we constructed binary functional brain networks from long-term, multichannel electroencephalographic data recorded from nine patients with epilepsy. The corresponding network properties were quantified using the average network degree. It was found that the network degree (as well as other network properties such as the network efficiency and clustering coefficient) exhibited large fluctuations over time; however, it also exhibited specific periodic temporal structure over different time scales (1.5hr-24hr periods) that was consistent across subjects. We investigated the correlation of the phases of these network periodicities with the seizure onset by using circular statistics. The results showed that the instantaneous phases of the 3.5hr, 5.5hr, 12hr and 24hr network degree periodic components are not uniformly distributed, suggesting that functional network properties are related to seizure generation and occurrence.

Recent modelling studies (Hadjipapas et al. [2009]: Neuroimage 44:1290-1303) have shown that it may be possible to distinguish between different neuronal populations on the basis of their macroscopically measured (EEG/MEG) mean field. We set out to test whether the different orientation columns contributing to a signal at a specific cortical location could be identified based on the measured MEG signal. We used 1.5deg square, static, obliquely oriented grating stimuli to generate sustained gamma oscillations in a focal region of primary visual cortex. We then used multivariate classifier methods to predict the orientation (left or right oblique) of the stimuli based purely on the time-series data from this one location. Both the single trial evoked response (0–300 ms) and induced post-transient power spectra (300–2,300 ms, 20–70 Hz band) due to the different stimuli were classifiable significantly above chance in 11/12 and 10/12 datasets respectively. Interestingly, stimulus-specific information is preserved in the sustained part of the gamma oscillation, long after perception has occurred and all neuronal transients have decayed. Importantly, the classification of this induced oscillation was still possible even when the power spectra were rank-transformed showing that the different underlying networks give rise to different characteristic temporal signatures. Hum Brain Mapp, 2010. © 2009 Wiley-Liss, Inc.

Item does not contain fulltext Gamma oscillations contribute significantly to the manner in which neural activity is bound into functional assemblies. The mechanisms that underlie the human gamma response, however, are poorly understood. Previous computational models of gamma rely heavily on the results of invasive recordings in animals, and it is difficult to assess whether these models hold in humans. Computational models of gamma predict specific changes in gamma spectral response with increased excitatory drive. Hence, differences and commonalities between spikes, LFPs and MEG in the spectral responses to changes in excitatory drive can lead to a refinement of existing gamma models. We compared gamma spectral responses to varying contrasts in a monkey dataset acquired previously (Roberts et al., 2013) with spectral responses to similar contrast variations in a new human MEG dataset. We found parametric frequency shifts with increasing contrast in human MEG at the single-subject and the single-trial level, analogous to those observed in the monkey. Additionally, we observed parametric modulations of spectral asymmetry, consistent across spikes, LFP and MEG. However, while gamma power scaled linearly with contrast in MEG, it saturated at high contrasts in both the LFP and spiking data. Thus, while gamma frequency changes to varying contrasts were comparable across spikes, LFP and MEG, gamma power changes were not. This indicates that gamma frequency may be a more stable parameter across scales of measurements and species than gamma power. The comparative approach undertaken here represents a fruitful path towards a better understanding of gamma oscillations in the human brain.

Fine-scale temporal organization of cortical activity in the gamma range (∼25–80Hz) may play a significant role in information processing, for example by neural grouping (‘binding’) and phase coding. Recent experimental studies have shown that the precise frequency of gamma oscillations varies with input drive (e.g. visual contrast) and that it can differ among nearby cortical locations. This has challenged theories assuming widespread gamma synchronization at a fixed common frequency. In the present study, we investigated which principles govern gamma synchronization in the presence of input-dependent frequency modulations and whether they are detrimental for meaningful input-dependent gamma-mediated temporal organization. To this aim, we constructed a biophysically realistic excitatory-inhibitory network able to express different oscillation frequencies at nearby spatial locations. Similarly to cortical networks, the model was topographically organized with spatially local connectivity and spatially-varying input drive. We analyzed gamma synchronization with respect to phase-locking, phase-relations and frequency differences, and quantified the stimulus-related information represented by gamma phase and frequency. By stepwise simplification of our models, we found that the gamma-mediated temporal organization could be reduced to basic synchronization principles of weakly coupled oscillators, where input drive determines the intrinsic (natural) frequency of oscillators. The gamma phase-locking, the precise phase relation and the emergent (measurable) frequencies were determined by two principal factors: the detuning (intrinsic frequency difference, i.e. local input difference) and the coupling strength. In addition to frequency coding, gamma phase contained complementary stimulus information. Crucially, the phase code reflected input differences, but not the absolute input level. This property of relative input-to-phase conversion, contrasting with latency codes or slower oscillation phase codes, may resolve conflicting experimental observations on gamma phase coding. Our modeling results offer clear testable experimental predictions. We conclude that input-dependency of gamma frequencies could be essential rather than detrimental for meaningful gamma-mediated temporal organization of cortical activity., Author Summary Almost 350 years ago the physicist and polymath Christiaan Huygens first observed the synchronization between two pendulum clocks attached to a common support. Since then synchronization has been recognized as a universal phenomenon from astronomy to biology. The phase-locking (synchrony) and the phase-relation between the two pendulums are determined by two principal forces: the synchronization force exerted over the connection and the tendency to desynchronize due to frequency (speed) differences. We propose that gamma synchronization (25–80Hz) among oscillating cortical neurons in the brain can be understood according to the same principles—like a field of many connected pendula—with the critical addition that input changes the frequency of gamma oscillations, as shown by recent experimental studies. It has been assumed that input-dependent changes in oscillation frequency are detrimental for a meaningful role of gamma synchronization in neural processing. To the contrary, our theoretical analysis demonstrates that because input can change the frequency of the oscillation, phase-locking and phase-relations among neurons relate systematically to input. By analogy, it is because a local push to a pendulum will change its frequency, that resulting changes in phase-locking and phase-relation among the pendula can be used to derive the external force applied.

91 Journal Article 103 130

The Electroencephalogram (EEG) is often contaminated by muscle artifacts. EEG is a widely used recording technique for the study of many brain related diseases such as epilepsy. The detection and removal of muscle artifacts from the EEG signal poses a real challenge and is crucial for the reliable interpretation of EEG-based quantitative measures. In this paper, an automatic method for detection and removal of muscle artifacts from scalp EEG recordings, based on canonical correlation analysis (CCA), is introduced. To this end we exploit the fact that the EEG signal may exhibit altered autocorrelation structure and spectral characteristics during periods when it is contaminated by muscle activity. Therefore, we design classifiers in order to automatically discriminate between contaminated and non-contaminated EEG epochs using features based on the aforementioned quantities and examine their performance on simulated data and in scalp EEG recordings obtained from patients with epilepsy.

Seizure detection and prediction studies using scalp- or intracranial-EEG measurements often focus on short-length recordings around the occurrence of the seizure, normally ranging between several seconds and up to a few minutes before and after the event. The underlying assumption in these studies is the presence of a relatively constant EEG activity in the interictal period, that is presumably interrupted by the occurrence of a seizure, at the time the seizure starts or slightly earlier. In this study, we put this assumption under test, by examining long-duration scalp EEG recordings, ranging between 22 and 72 hours, of five patients with epilepsy. For each patient, we construct functional brain networks, by calculating correlations between the scalp electrodes, and examine how these networks vary in time. The results suggest not only that the network varies over time, but it does so in a periodic fashion, with periods ranging between 11 and 25 hours.

Aims: The relationship between collective population activity (LFP) and spikes underpins network computation, yet it remains poorly understood. Previous studies utilized pre-defined LFP features to predict spiking from simultaneously recorded LFP, and have reported good prediction of spike bursts but only moderate accuracies for individual spikes. Our aim was to utilize a data-driven approach, without relying on feature selection, to predict individual spike times. Methods: The relationship between LFPs and multi-unit spike trains in monkey early visual cortex during passive viewing of grating stimuli was analyzed using a variant of the general Volterra approach (Laguerre-Volterra network). Network parameters were trained based on a hybrid Genetic Algorithm ? Interior Point optimization method, and model selection was achieved via cross-validation. The Matthews Correlation Coefficient (-1 Results: Single trial MCCs ranged from 0.45 to 0.66 (median=0.60). Superior performance of 2nd order relative to 1st order models indicated a nonlinear relationship between LFPs and spikes in visual cortex. Consistent with other studies, the PDMs of the identified system exhibited low-pass (theta frequency) and high-pass (gamma frequency) characteristics. Conclusions: We successfully predicted multi-unit spike times from local LFPs with reasonable accuracy and without selection of a-priori features. Our approach enhances our understanding of spike precision and spike timing, and of the network principles underlying the neural code

This study used magnetoencephalography (MEG) to examine the dynamic patterns of neural activity underlying the auditory steady-state response. We examined the continuous time-series of responses to a 32-Hz amplitude modulation. Fluctuations in the amplitude of the evoked response were found to be mediated by non-linear interactions with oscillatory processes both at the same source, in the alpha and beta frequency bands, and in the opposite hemisphere.

It is well established that both the choice of recording reference (montage) and volume conduction affect the connectivity measures obtained from scalp EEG. Our purpose in this work is to establish the extent to which they influence the graph theoretic measures of brain networks in epilepsy obtained from scalp EEG. We evaluate and compare two commonly used linear connectivity measures - cross-correlation and coherence - with measures that account for volume conduction, namely corrected cross-correlation, imaginary coherence, phase lag index and weighted phase lag index. We show that the graphs constructed with cross-correlation and coherence are the most affected by volume conduction and montage; however, they demonstrate the same trend - decreasing connectivity at seizure onset, which continues decreasing in the ictal and early post-ictal period, increasing again several minutes after the seizure has ended-with all other measures except imaginary coherence. In particular, networks constructed using cross-correlation yield better discrimination between the pre-ictal and ictal periods than the measures less sensitive to volume conduction. Thus, somewhat paradoxically, although removing effects of volume conduction allows for a more accurate reconstruction of the true underlying networks this may come at the cost of discrimination ability with respect to brain state.