Regular articleA neural mass model for MEG/EEG:: coupling and neuronal dynamics
Introduction
It is generally assumed that the signals measured in magnetoencephalography (MEG) and electroencephalography (EEG) can be decomposed into distinct frequency bands (delta: 1–4 Hz, theta: 4–8 Hz, alpha: 8–12 Hz, beta: 12–30 Hz, gamma: 30–70 Hz) (Nunez, 1981). These rhythms sometimes exhibit robust correlates of behavioural states but often with no obvious functional role. It is clear that MEG/EEG signals result mainly from extracellular current flow, associated with massively summed postsynaptic potentials in synchronously activated and vertically oriented neurons (dendritic activity of macro-columns of pyramidal cells in the cortical sheet). The exact neurophysiological mechanisms, which constrain this synchronisation to a given frequency band, remain obscure. However, the generation of oscillations appears to depend on interactions between inhibitory and excitatory populations, whose kinetics determine their oscillation frequency. This dependency suggests a modelling strategy could help to disclose the causes of different MEG/EEG rhythms and to characterise the neuronal processes underlying MEG/EEG activity.
There are several ways to model neural signals (Whittington et al., 2000a): either using a detailed model, in which it is difficult to determine the influence of each model parameter, or a simplified one, in which realism is sacrificed for a more parsimonious description of key mechanisms. The complexity of neural networks generating MEG/EEG signals is such that the second approach is usually more viable. This involves modelling neuronal activity with simplifying assumptions and empirical priors to emulate realistic signals. Neural mass models Freeman, 1978, Lopes da Silva et al., 1974, Robinson et al., 2001, Stam et al., 1999, Valdes et al., 1999, Wendling et al., 2000 are based upon this approach. These models comprise macro-columns, or even cortical areas, using only one or two state variables to represent the mean activity of the whole population. This procedure, sometimes referred to as a “mean-field approximation,” is very efficient for determining the steady-state behaviour of neuronal systems, but its utility in a dynamic or nonstationary context is less established (Haskell et al., 2001). The majority of neural mass models of EEG responses have been designed to model alpha rhythms Jansen and Rit, 1995, Lopes da Silva et al., 1974, Stam et al., 1999. Recent studies have emphasised that the kinetics of inhibitory populations have a key influence on the signals generated (Wendling et al., 2002). Specifically, it has been suggested that fast inhibition kinetics are needed to produce gamma-like activity (Jefferys et al., 1996). In fact, MEG/EEG signals depend upon the kinetics of both inhibitory and excitatory neural populations, and exhibit very complex dynamics because of the huge diversity and connectivity of cortical areas.
In this study we describe a simple neural mass model that can produce various rhythms ranging from delta to gamma, depending on the kinetics of the populations modelled. We start from the model of Jansen and Rit, (1995) using standard parameters to produce alpha activity in a single area. We then show that variation of excitatory and inhibitory kinetics, within a physiologically plausible range, can generate oscillatory activity in the delta, theta, alpha, beta, and gamma bands using the same model. Next, we assume that a cortical area comprises several resonant neuronal populations, characterised by different kinetics. We describe the particular case of two populations that underlie intrinsic alpha and gamma rhythms. In this dual-kinetic model, a single parameter controls the relative contribution of fast and slow populations, leading to a modulation of the rhythms produced. Finally, we address the coupling of two areas, with a particular focus on the dependence of cortical rhythms upon the strength of the coupling and upon the distance between cortical areas (modelled as a propagation delay). Using several sets of parameters, we show that this model can generate a wide variety of oscillations in the alpha, beta, and gamma bands that are characteristic of MEG/EEG signals.
The goal of this study is to introduce and characterise the behaviour of the model. This is the first component of a broader program that aims to (1) quantify the sensitivity of linear and nonlinear methods for the detection of long-range cortical interactions using MEG/EEG signals, and (2) model event-related dynamics and derive basis functions for statistical models of averaged evoked potentials/fields.
Section snippets
Neural mass models
MEG/EEG signals are generated by the massively synchronous dendritic activity of pyramidal cells. Modelling MEG/EEG signals is seldom tractable using realistic models, at the neuronal level, because of the complexity of real neural networks. Since the 1970s Freeman, 1978, Lopes da Silva et al., 1974, Nunez, 1974, Wilson and Cowan, 1972 the preferred approach has been the use of neural mass models, i.e., models that describe the average activity with a small number of state variables. These
Dynamic properties of simulated signals
For each simulation described below, the differential equations were solved numerically using a second order Runge-Kutta algorithm (a fourth-fifth order Runge-Kutta method gave similar results). As it is known that this type of algorithm is not optimal for stochastic differential equations, we compared the simulated signals with those generated using a second order stochastic Runge-Kutta algorithm (Honeycutt, 1992). There were no qualitative differences between the two integration schemes. The
Discussion
Neural mass models afford a straightforward approach to modelling the activity of populations of neurons. Their main assumption is that the state of the population can be approximated using very few state variables (generally limited to mean membrane currents, potentials, and firing rates). Given a macroscopic architecture, describing the overall connectivity (1) between populations of a given cortical area, and (2) between different cortical areas, it is possible to simulate the steady-state
Acknowledgements
This work was supported by The Wellcome Trust.
References (46)
- et al.
Modeling the link between functional imaging and neuronal activitysynaptic metabolic demand and spike rates
NeuroImage
(2002) - et al.
A robust method for detecting interdependencesapplication to intracranially recorded EEG
Physica D
(1999) - et al.
A model of the coupling between brain electrical activity, metabolism, and hemodynamicsapplication to the interpretation of functional neuroimaging
NeuroImage
(2002) - et al.
Zero-lag synchronous dynamics in triplets of interconnected cortical areas
Neural Netw.
(2001) Bayesian estimation of dynamical systemsan application to fMRI
NeuroImage
(2002)- et al.
Psychophysiological and modulatory interactions in neuroimaging
NeuroImage
(1997) - et al.
Classical and Bayesian inference in neuroimagingtheory
NeuroImage
(2002) - et al.
Neuronal networks for induced “40 Hz” rhythms
Trends Neurosci.
(1996) - et al.
A derivation of a macroscopic field theory of the brain from the quasi-microscopic neural dynamics
Physica D
(1997) - et al.
Alpha rhythmsnoise, dynamics and models
Int. J. Psychophysiol.
(1997)