Research paper
Predicting seizure by modeling synaptic plasticity based on EEG signals - a case study of inherited epilepsy

https://doi.org/10.1016/j.cnsns.2017.08.020Get rights and content

Highlights

  • We make data processing of a clinical EEG record from an inherited epilepsy patient and analyze the correlation of each EEG channels for some episodes.

  • We modify a reasonable epilepsy model based on the clinical EEG data.

  • The behavior of model EEG and the clinical EEG match well, such as synchronization and frequency.

  • Robustness of our model is verified, including different connected structures.

  • Synaptic plasticity and excitatory signals are found to affect the seizure duration.

Abstract

This paper explores the internal dynamical mechanisms of epileptic seizures through quantitative modeling based on full brain electroencephalogram (EEG) signals. Our goal is to provide seizure prediction and facilitate treatment for epileptic patients. Motivated by an earlier mathematical model with incorporated synaptic plasticity, we studied the nonlinear dynamics of inherited seizures through a differential equation model. First, driven by a set of clinical inherited electroencephalogram data recorded from a patient with diagnosed Glucose Transporter Deficiency, we developed a dynamic seizure model on a system of ordinary differential equations. The model was reduced in complexity after considering and removing redundancy of each EEG channel. Then we verified that the proposed model produces qualitatively relevant behavior which matches the basic experimental observations of inherited seizure, including synchronization index and frequency. Meanwhile, the rationality of the connectivity structure hypothesis in the modeling process was verified. Further, through varying the threshold condition and excitation strength of synaptic plasticity, we elucidated the effect of synaptic plasticity to our seizure model. Results suggest that synaptic plasticity has great effect on the duration of seizure activities, which support the plausibility of therapeutic interventions for seizure control.

Introduction

EEG is considered as the gold standard for inherited seizure detection among other seizure detection modalities [1], [2], [3], since it is non-invasive and can repeatedly record the brain activities of patients for a long duration for analyzing patients’ conditions and monitoring treatments [4], [5], [6], including evaluating the most suitable anti-epileptic medicine, determining the seizure focus, finding out the causes for impaired cognitive function. Clinical diagnosis and treatment of patients have driven recent research on epileptic seizure prediction based on EEG data [7], including both physiological experiments [8], [9], [10], [11] and model analysis [12], [13], [14].

Epileptic seizure prediction remains a challenging problem. Considerable efforts have been made to predict seizures focused on several types of features that discriminate between interictal and preictal states [15]. A number of statistical features of seizures and non-seizures can be extracted based on continuously recorded EEG data which can improve prediction accuracy. These include univariate features, such as the power spectral density or autoregressive modeling coefficients of single EEG channels, as well as bivariate features that measure pairwise correlations between EEG channels, such as maximum cross correlation or phase synchrony [16], [17]. Signal analysis techniques may transform ictal EEG signals in a way that inherent hidden structures are revealed. Due to the nonstationary and nonlinear nature of EEG signals, time domain analysis, frequency domain analysis, time-frequency domain analysis, wavelet transform analysis and other nonlinear methods have been used to distinguish EEG signals in normal period and during seizures [18], [19], [20]. These models usually reflect statistical features of seizures.

Alternatively, dynamical seizure model established by electrical phenomenon can also contribute to seizure prediction. Through dynamics modeling analysis, we can obtain insight about the mechanics and symptoms of disease, and then to improve the prediction. Deriving a simple and functional dynamic model that represents the onset of epileptic seizures is attractive [21], [22], [23]. Taylor et al. demonstrated epileptic spike-wave discharges in an extended version of Amari’s neural field model, and used a computational model of epilepsy spike-wave dynamics to evaluate the effectiveness of a pseudospectral method to simulate absence seizures [24], [25]. The model leads to a prototypic equation of clinical epileptic dynamics in time and space. By using a biophysically based model, Chen et al. pointed out that the typical absence seizure activities can be controlled and modulated by direct GABAergic projections from substantia nigra pars reticulata (SNr) to either either the thalamic reticular nucleus (TRN) or the specific relay nuclei (SRN) of thalamus [26]. A number of studies have shown that synaptic plasticity contributes to the pathophysiology of epilepsy and other neurological and psychiatric disorders [27], [28]. Alamir et. al proposed a dynamical seizure model which demonstrated that epileptic activity is induced by a high level of synchrony that is due to a disturbed balance between two opposing mechanisms: A basic functional desynchronization mechanism and a synaptic based synchronization mechanism [29].

The mapping of human brain function in real time has suffered from a lack of innovation in the past decade. Even though brain-imaging tools, e.g. functional magnetic resonance imaging (fMRI), positron emission computed tomography (PET), magnetic encephalography (MEG) are widely used, they are limited by low spatial and temporal resolution, cost, mobility and suitability for long-term monitoring. For example, fMRI has the advantage of providing spatially-resolved data, but suffers from an ill-posed temporal inverse problem, i.e., a map with regional activations does not contain information about when and in which order these activations have occurred [30]. In contrast, EEG signals have been successfully used to obtain useful diagnostic information in clinical contexts. Further, they present the advantage to be highly portable, inexpensive, and can be acquired at the bedside or in real-life environments with a high temporal resolution. EEG offers the possibility of measuring the electrical activity of neuronal cell assemblies on the sub millisecond time scale [31], [32], [33].

While there is successful seizure research in animal model and seizure patients, seizure modeling based on the dynamic features of EEG information is not common. Motivated by previous work [29], we will investigate the dynamics of inherited seizure based on clinical EEG of one patient recordings as a case study. We will follow later with additional data to further modify and validate the modeling strategy. We will describe the methodology in Section 2, including clinical neonatal EEG signals, dynamical seizure model and synchronization measurement. Then in Section 3, simulation results will be provided under different situations to validate our model. Then, Section 4 compares the effects of synaptic plasticity on the dynamics of the seizure network. Conclusions and discussion are in Section 5.

Section snippets

Methods

EEG records a weighted average of local field potential at various brain areas by a group of electrodes placed on the scalp. EEG data can be used to diagnose neural diseases, such as epilepsy, schizophrenia, manic depression and mental disorders. Particularly, neonatal seizures result from synchronous discharges from groups of neurons and manifest as periods of heightened periodicity in the EEG lasting for more than 10 s [34]. Thus EEG has been used as a detection modality for seizure patients.

Numerical simulation

Regardless of seizure period and non-seizure period, the EEG signals of neonatal seizures show a train of burst and suppression waves. A valid model should satisfy these characters that the high amplitude spike and slow wave of irregular mixed outbreak always alternate with nearly flat suppression phase with statistics matching with clinical data. In the following, we give two simulation cases to test the validity of our modeling strategy.

Network connection

Next, to test the relevance of the synaptic connection in our seizure model, we test two extreme connection situations for brain regions to understand the range and limitations of our model.

Synaptic plasticity

There is plenty of experimental evidences demonstrating that synaptic plasticity plays an essential role in epileptic seizures. From the modeling of synaptic plasticity, we know both hth and Uexc are two key elements. Hence in this section, these parameters relevant to synaptic plasticity are analyzed in the epilepsy network from simulated data, using the EEG model of 400–440 s as an example.

Conclusion

In this study we characterize the dynamic evolution of EEG activity during seizures through a modeling study. Our research enriches a simple dynamical model describing the epileptic seizure initiation through transition from interictal to ictal state in a brain predisposed to epilepsy. By analyzing the correlation connection of neuronal circuits during seizure of multi-channels, the subregions of brains in seizure patients are classified. As a result, our dynamical model adequately reproduces

Acknowledgments

This research was supported by the Start-up Scientific Research Foundation of Northwestern Ploytechnical University: No. G2016KY0301 and the National Natural Science Foundation of China: Grant Nos. 11602192, 11325208, and 11772019.

Research reported in this publication was also supported in part by the National Center for Advancing Translational Sciences of the National Institutes of Health under award number UL1TR001105. The content is solely the responsibility of the authors and does not

References (41)

  • D.M. Feeney et al.

    The prediction of posttraumatic epilepsy a mathematical approach

    Formaerly Arch Neurol

    (1979)
  • R. Yadav et al.

    Model-based seizure detection for intracranial EEG recordings

    IEEE Trans Biomed Eng

    (2012)
  • A. Yang et al.

    Development and validation of a seizure prediction model in critically ill children

    Seizure

    (2015)
  • R.R. Clancy et al.

    Epilepsia

    Occult Neonatal Seizures

    (1988)
  • M.J. Painter et al.

    Compared with phenytoin for the treatment of neonatal seizures

    N Engl J Med

    (1999)
  • L. Gergory et al.

    Consequences of neonatal seizures in the rat: morphological and behavioral effects

    Ann Neurol

    (1998)
  • A.B. Gardner et al.

    One-class novelty detection for seizure analysis from intracranial EEG

    J Mach Learn Res

    (2006)
  • R. James et al.

    Seizure prediction using EEG spatiotemporal correlation structure

    Epilepsy Behav

    (2012)
  • F. Mormann et al.

    Seizure prediction: the long and winding road

    Brain

    (2007)
  • P. Celka et al.

    Processing and time-frequency analysis of newborn EEG seizures

    IEEE Eng Med Biol Mag

    (2001)
  • Cited by (20)

    • Design and implementation of a multi-sensor newborn EEG seizure and background model with inter-channel field characterization

      2019, Digital Signal Processing: A Review Journal
      Citation Excerpt :

      Newborn electroencephalogram (EEG) can be used to detect seizures [1–3], predict neurological outcomes [4–6], and assess risk and prognosis [7–10].

    • Automated seizure prediction

      2018, Epilepsy and Behavior
    • Multilevel Mathematical Model of Epileptic Seizures

      2023, Mathematical Biology and Bioinformatics
    View all citing articles on Scopus
    View full text