Curr Opin Neurobiol. Apr 2013; 23(2): 172–178.
PMCID: PMC3925802

Analysing connectivity with Granger causality and dynamic causal modelling

Highlights

► A brief introduction to the analysis of directed connectivity in brain networks. ► An overview of advances in Granger causality and dynamic causal modelling. ► A comparative evaluation of both approaches in terms of their pros and cons.

Abstract

This review considers state-of-the-art analyses of functional integration in neuronal macrocircuits. We focus on detecting and estimating directed connectivity in neuronal networks using Granger causality (GC) and dynamic causal modelling (DCM). These approaches are considered in the context of functional segregation and integration and — within functional integration — the distinction between functional and effective connectivity. We review recent developments that have enjoyed a rapid uptake in the discovery and quantification of functional brain architectures. GC and DCM have distinct and complementary ambitions that are usefully considered in relation to the detection of functional connectivity and the identification of models of effective connectivity. We highlight the basic ideas upon which they are grounded, provide a comparative evaluation and point to some outstanding issues.

Current Opinion in Neurobiology 2013, 23:172–178

This review comes from a themed issue on Macrocircuits

Edited by Steve Petersen and Wolf Singer

For a complete overview see the Issue and the Editorial

Available online 21st December 2012

0959-4388/$ – see front matter, © 2012 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.conb.2012.11.010

Introduction

Several dichotomies have proved useful in thinking about analytic approaches to functional brain architectures. Perhaps the most fundamental is the distinction between functional segregation and integration. Functional segregation refers to the anatomical segregation of functionally specialised cortical and subcortical systems, while functional integration refers to the coordination and coupling of functionally segregated systems [1••]. Within functional integration, two main classes of connectivity have emerged — functional and effective connectivity. Functional connectivity refers to the statistical dependence or mutual information between two neuronal systems, while effective connectivity refers to the influence that one neural system exerts over another [2•,3]. This distinction is particularly acute when considering the different analyses one might apply to electrophysiological or neuroimaging timeseries.

Functional and effective connectivity

Because functional connectivity is defined in terms of statistical dependencies, it is an operational concept that underlies the detection of (inference about) a functional connection, without any commitment to how that connection was caused. In other words, one tests for dependencies between two or more timeseries, to reject the null hypothesis of statistical independence. This is equivalent to assessing the mutual information and testing for significant departures from zero. At its simplest, this involves assessing (patterns of) correlations — of the sort that define intrinsic brain networks. An important distinction — within functional connectivity — rests on whether dependencies are instantaneous or reflect an underlying dynamical process, in which causes precede consequences. This leads to the distinction between analyses of directed and undirected functional connectivity that do and do not appeal to temporal precedence respectively. Common examples of techniques used to assess undirected functional connectivity (dependencies) include independent components analysis [4] and various measures of synchrony, correlation, or coherence [5]. However, we will focus on analyses of directed functional connectivity — of which the prime example is Granger causality (GC) [6•]. This is because coupling in the brain is both directed and largely reciprocal (producing cyclic graphs or networks with loops that preclude structural causal modelling). As we will see below, GC and related concepts such as transfer entropy (TE) rest on establishing a statistical dependence between a local measurement of neuronal activity and measurements of activity elsewhere in the past.

Functional connectivity considers dependencies between measured neurophysiological responses. In contrast, effective connectivity is between hidden neuronal states generating measurements. Crucially, effective connectivity is always directed and rests on an explicit (parameterised) model of causal influences — usually expressed in terms of difference (discrete time) or differential (continuous time) equations. The most popular approach to effective connectivity is dynamic causal modelling (DCM) [7–10,11••,12••]. In this context, causality is inherent in the form of the model, where fluctuations in hidden neuronal states cause changes in others: for example, changes in postsynaptic potentials in one area are caused by inputs from other areas. The parameters of dynamic causal models correspond to effective connectivity — usually cast as synaptic density or coupling parameters — that are optimised by fitting the model to data. The notion of effective connectivity stems from the pioneering work of Gerstein and Perkel [13] in early attempts to interpret multivariate electrophysiological recordings. At its inception, effective connectivity referred to models; in the sense of the simplest possible circuit diagrams that explain observed responses [14]. In modern parlance, these correspond to dynamic causal models with the greatest evidence: namely, models with the minimum complexity that furnish an accurate explanation for data (see below). In what follows, we review recent developments in the analysis of directed functional connectivity with GC and TE, the analysis of directed effective connectivity with DCM and then consider the approaches in light of each other. Figure 1 provides an overview of recent developments in these techniques.

Figure 1
A timeline of recent advances in Granger causality (top panel) and dynamic causal modelling (bottom panel). Entries above the time lines pertain to functional magnetic resonance imaging (MRI) and those below the lines report specific developments for ...

Granger causality and transfer entropy

The core idea behind GC is that X ‘Granger causes’ Y if X contains information that helps predict the future of Y better than information already in the past of Y (and in the past of other ‘conditioning’ variables Z). The most common implementation of GC is via linear vector autoregressive (VAR) modelling of timeseries data, enabling both statistical significance testing and estimation of GC magnitudes [6•,15•,16]. However, GC is not limited to this implementation; it can use nonlinear, time-varying, and non-parametric models [17,18]. In particular, TE [19] represents an information-theoretic generalisation of GC that does not require a parameterised model (is model-free). Specifically, the TE from X to Y is zero if, and only if, Y is conditionally independent of X's past, given its own past. Importantly, for Gaussian data, TE is equivalent to GC [20••], furnishing a useful interpretation of GC in terms of information transfer in ‘bits’. Related approaches include partial directed coherence and the directed transfer function; see [21] for a review. Here we focus on the most popular of these techniques, namely GC:

Following its introduction within econometrics [6•,15•], GC has been applied in neuroscience partly because it is simple to estimate, given (stationary stochastic) timeseries. Such data are generated by a wide range of neuroimaging and neurophysiological methods. GC has some useful properties including a decomposition of causal influence by frequency [15•] and formulation in an ‘ensemble’ form, allowing evaluation of GC between multivariate sets of responses [22]. GC has provided useful descriptions of directed functional connectivity in many electrophysiological studies [23–25]. Recently, Bosman et al. [26••] analysed electrocorticographic data from macaque monkeys to show that ‘bottom-up’ signals across multiple cortical regions were most prominent in the gamma band, while ‘top down’ influences dominated at beta frequencies — a finding that is strikingly congruent with neural implementations of predictive coding [27]. GC can also be applied to standard EEG or MEG signals, either at the source or sensor level (following spatial filtering to reduce the impact of volume conduction). For example, Barrett et al. [28•] used source-localised EEG to show that gamma-band GC between posterior and anterior cingulate cortices reliably increased during anaesthetic loss of consciousness, extending previous results obtained using (undirected) phase synchrony [29]. We will turn to this example later in the context of DCM.

The application GC to fMRI is more controversial, given the slow dynamics and regional variability of the haemodynamic response to underlying neuronal activity [30,31]; and see ‘Pros and Cons’ below. While naïve application of GC to fMRI data is unlikely to be informative, careful consideration of the methodological issues has permitted some useful applications that have produced testable hypotheses. For example, Wen et al. [32•] analysed fMRI data obtained from a cued spatial visual attention task; finding that GC from dorsal to ventral frontoparietal regions predicted enhanced performance, while GC in the reciprocal direction was associated with degraded performance. These findings are consistent with the notion that dorsal attentional regions mediate goal-oriented top-down deployment of attention, while ventral regions mediate stimulus-driven bottom-up reorienting. In a similar paradigm, Bressler et al. [33] found that GC from parietal to occipital areas was predictive of behavioural performance. In a final and unusual example, Schippers et al. [34] used GC of fMRI signals to analyse directed interactions between the brains of two subjects engaged in a social game (charades), providing novel evidence for ‘mirror neuron system’ formulations of social interaction. Another promising application of GC is to intracranial local field potentials, which possess high temporal and spatial resolution and which comprise comparatively few variables (as compared to fMRI voxels or EEG sensors). An early application in this area, Gaillard et al. [35] examined directed functional connectivity during supraliminal as compared to subliminal visual word processing.

Dynamic causal modelling

The basic idea behind DCM is that neural activity propagates through brain networks as in an input-state-output system, where causal interactions are mediated by unobservable (hidden) neuronal dynamics. This multi-input multi-output neuronal model is augmented with a forward, or observation model that describes the mapping from neural activity to observed responses. Together neuronal and observation model comprise a full generative model that takes a particular form depending on the data modality. The key outputs of DCM are the evidence for different models and the posterior parameter estimates of the (best) model, particularly those describing the coupling among brain regions. These allow for model and system identification, respectively.

DCM was introduced for fMRI timeseries [36], where the neuronal model comprises one or two hidden (lumped) neuronal states for each region. The neuronal dynamics of each region depend on the strength of connections within that region (parameterised by a self-connection), on the strength of external inputs (experimental input parameters) and on inputs from other regions in the network (the coupling parameters). Neuronal activity is then transformed through a haemodynamic model (with region-specific parameters) to model measured responses [37]. The coupling between brain regions can then be estimated for a particular model architecture using standard variational Bayesian techniques [36]. In practice, it is usual to specify different architectures or hypotheses and formally compare the evidence for these models, before examining parameter estimates [38]. DCM necessarily accounts for directed connections among brain regions and disambiguates the neuronal drivers of a particular event and subsequent signal propagation. Electrophysiological measurements support richer models of neuronal dynamics in DCM that comprise sources with laminar specific mixtures of neuronal populations. These have evolved from kernel-based models [7] that use postsynaptic convolution operators to describe responses at excitatory and inhibitory synapses to conductance-based models, where particular ion channels can be modelled and identified [39]. These neural mass models are accompanied by linear electromagnetic forward models to generate responses at EEG scalp electrodes, at MEG sensors or at intracranial recording sites. Application of DCM to animal local-field potential data has facilitated validation studies, where independent, invasive measurements (e.g. microdialysis or pharmacological perturbations) suggest that DCM can be used to estimate the physiological mechanisms responsible for mediating effective connectivity [40•].

Pros and cons

Clearly, GC and DCM have complementary aims and strengths: GC can be applied directly to any given timeseries to detect the coupling among empirically sampled neuronal systems. This can provide useful insights into the system's dynamical behaviour in different conditions or in spontaneously active ‘resting’ states. One might then proceed to a more mechanistic (model or hypothesis — driven) characterisation using DCM. However, this calls for bespoke models of the system in question [41••]. In other words, GC is a generic inferential procedure characterising directed functional connectivity, while DCM is a framework that enforces (or enables) specific models or hypotheses to be tested. Crucially, both rest on model selection: In DCM this involves comparing the evidence for different models directly [38], while model selection in GC is implicit in the test for the presence of GC — and also arises in the selection of VAR model order, using standard approximations to model evidence, such as the Akaike or Bayesian information criteria [42].

Although GC is generic, its naive application is not always justified. For example, application to fMRI must recognise the indirect relation between neuronal activity and haemodynamic responses. In particular, regional variations in haemodynamic latency could confound the temporal precedence assumptions of GC [30]. While these variations can be partially controlled for by contrasting GC between experimental conditions [17,43,31] false inferences remain possible. Interestingly, recent theory and modelling suggests that GC may be robust to haemodynamic variations but not when combined with down-sampling and measurement noise [44]. In contrast, DCM models haemodynamic variations explicitly and tries to explain the data at the level of hidden neuronal states — in other words, it tries to get beneath the surface structure of the data to explain how they were generated: see [45,46•] for further discussion.

In analysis of electrophysiological timeseries, GC is more widely accepted because there is no temporal lag between the responses recorded and their underlying (neuronal) causes and because the data can be sampled at fast timescales. The advantages of GC in furnishing frequency-dependent and multivariate measures have been clearly demonstrated [22,26••,28•]. However, there is an unresolved issue in this setting — the random fluctuations assumed by GC are serially independent (show no temporal correlations and fluctuate at very fast timescales). This is an issue because neuronal fluctuations in the brain are produced by neuronal systems that have the same time constants as the system studied. While serial independence can be checked for, the nature of neuronal fluctuations may deserve more attention in the future.

A key feature of DCM is that it can include variables that describe dynamics that are hidden from observation. For example, the GC analysis of anaesthetic loss of consciousness by Barrett et al. [28•] mentioned above, was complemented by a mechanistic study by Boly et al. [47•] using DCM. She found that a DCM that included a hidden thalamic source performed better than DCMs based solely on observed cortical timeseries, and established a dissociation between the effects of (measured) cortical and (inferred) subcortical structures on levels of consciousness. In contrast to GC, being able to model hidden sources means the model (hypothesis) space can be very large and calls for a principled approach to Bayesian model comparison of models that are (a priori) considered equally plausible. The specification and interrogation of the model space is an outstanding conceptual issue for DCM.

DCM posits and identifies neuronal mechanisms responsible for functional integration in the brain. Connectivity in this setting necessitates biologically plausible explanations. In DCM for fMRI, new developments [48,49] enable the incorporation of background or ongoing spontaneous cortical fluctuations, nonlinearities and inhibitory neuronal populations [53]. The addition of spontaneous or stochastic fluctuations enhances the plausibility of the generative model at the neuronal level, where non-Markovian noise processes sit atop experimentally induced brain activations [48]. In DCM for electrophysiological data, the models will potentially allow the characterisation of receptor-specific contributions to brain connectivity, which may be important in a pharmacological and clinical setting [11••].

Conclusion

In conclusion, GC and DCM are complementary: both model neural interactions and both are concerned with directed causal interactions. GC models dependency among observed responses, while DCM models coupling among the hidden states generating observations. Despite this fundamental difference, the two approaches may be converging. On the one hand DCM for stochastic systems [48] can now accommodate the random fluctuations assumed by GC. On the other hand, state-space GC approaches can incorporate modality specific observation equations [50]. The ability to handle large numbers of sources for regions is facilitated by multivariate (ensemble) GC [22] and sparse regression techniques [51], as well as recent developments in post hoc model optimisation for network discovery with DCM [52]. One might hope that both approaches — perhaps GC disclosing candidate models for DCM — will counter the claims that modern brain mapping is a neo-phrenology and provide characterisations of brain circuits that may hold promise for the treatment of neurological and psychiatric disorders.

References and recommended reading

Papers of particular interest, published within the period of review, have been highlighted as:

  • • of special interest
  • •• of outstanding interest

Acknowledgements

KJF and RM are funded by the Wellcome Trust. AKS is funded by the EPSRC (EP/G007543/1) and the Dr. Mortimer and Theresa Sackler Foundation.

References

Zeki S., Shipp S. The functional logic of cortical connections. Nature. 1988;335:311–317. [PubMed]A short but important conceptual paper highlighting the importance of functional integration among specialised or segregated visual areas — and the hierarchical message passing supported by extrinsic (between area) and intrinsic (within area) connections.
Friston K.J., Frith C.D., Liddle P.F., Frackowiak R.S. Functional connectivity: the principal-component analysis of large (PET) data sets. J Cereb Blood Flow Metab. 1993;13:5–14. [PubMed]This paper introduced the distinction between functional and effective connectivity in brain imaging.
3. Breakspear M. Dynamic connectivity in neural systems — theoretical and empirical considerations. Neuroinformatics. 2004;2:205–225. [PubMed]
4. Bell A.J., Sejnowski T.J. An information maximisation approach to blind separation and blind de-convolution. Neural Comput. 1995;7:1129–1159. [PubMed]
5. Siegel M., Donner T.H., Engel A.K. Spectral fingerprints of large-scale neuronal interactions. Nat Rev Neurosci. 2012;13:131–134. [PubMed]
Granger C.W.J. Investigating causal relations by econometric models and cross-spectral methods. Econometrica. 1969;37:424–438.This paper introduced the operationalization of Granger causality in the form of linear vector autoregressive models, within the field of econometrics.
7. David O., Kiebel S., Harrison L., Mattout J., Kilner J.M., Friston K.J. Dynamic causal modeling of evoked responses in EEG and MEG. Neuroimage. 2006;30:1255–1272. [PubMed]
8. Garrido M.I., Kilner J.M., Kiebel S.J., Friston K.J. Evoked brain responses are generated by feedback loops. Proc Natl Acad Sci USA. 2007;104:20961–20966. [PubMed]
9. Kiebel S.J., Garrido M.I., Friston K.J. Dynamic causal modelling of evoked responses: the role of intrinsic connections. Neuroimage. 2007;36:332–345. [PubMed]
10. Garrido M.I., Kilner J.M., Kiebel S.J., Stephan K.E., Baldeweg T., Friston K.J. Repetition suppression and plasticity in the human brain. Neuroimage. 2009;48:269–279. [PubMed]
Moran R.J., Symmonds M., Stephan K.E., Friston K.J., Dolan R.J. An in vivo assay of synaptic function mediating human cognition. Curr Biol. 2011;21:1320–1325. [PubMed]This study served as a validation of DCM using non-invasive magneto-encephalography in human subjects. The experimental manipulation here was pharmacological, where levodopa was used to boost prefrontal dopamine during a working memory experiment. This induced known changes in the receptor activation in reverberatory networks in PFC. These changes at NMDA, AMPA and GABAa were recovered using a conductance-based DCM. Moreover, estimates from the DCMs were predictive of inter-individual behaviour in the memory task.
Boly M., Garrido M.I., Gosseries O., Bruno M.A., Boveroux P., Schnakers C., Massimini M., Litvak V., Laureys S., Friston K. Preserved feedforward but impaired top-down processes in the vegetative state. Science. 2011;332:858–862. [PubMed]An important example of the characterisation of directed (effective) connectivity in distinguishing between top-down and bottom-up extrinsic connections and their function in supporting different levels of consciousness.
13. Gerstein G.L., Perkel D.H. Simultaneously recorded trains of action potentials: analysis and functional interpretation. Science. 1969;164:828–830. [PubMed]
14. Aertsen A.M., Gerstein G.L., Habib M.K., Palm G. Dynamics of neuronal firing correlation: modulation of “effective connectivity” J Neurophysiol. 1989;61:900–917. [PubMed]
Geweke J. Measures of conditional linear dependence and feedback between time series. J Am Stat Assoc. 1984;79:907–915.Important extension of the basic GC concept to the frequency domain and to multivariate variables (the latter worked out more fully in [22]).
16. Seth A.K. A MATLAB toolbox for Granger causal connectivity analysis. J Neurosci Meth. 2010;186:262–273. [PubMed]
17. Roebroeck A., Formisano E., Goebel R. The identification of interacting networks in the brain using fMRI: model selection, causality and deconvolution. Neuroimage. 2011;58:296–302. [PubMed]
18. Dhamala M., Rangarajan G., Ding M. Analyzing information flow in brain networks with nonparametric Granger causality. Neuroimage. 2008;41:354–362. [PubMed]
19. Schreiber T. Measuring information transfer. Phys Rev Lett. 2000;85:461–464. [PubMed]
Barnett L.C., Barrett A.B., Seth A.K. Granger causality and transfer entropy are equivalent for Gaussian variables. Phys Rev Lett. 2009;103:238701. [PubMed]By showing an equivalence between GC and TE in the Gaussian case, this paper provides an information-theoretic interpretation of GC magnitude and eliminates the need to consider more complicated nonlinear GC measures (given Gaussian data).
21. Gourévitch B., Bouquin-Jeannès R.L., Faucon G. Linear and nonlinear causality between signals: methods, examples, and neurophysiological applications. Biol Cybern. 2006;95:349–369. [PubMed]
22. Barrett A.B., Barnett L.C., Seth A.K. Multivariate Granger causality and generalized variance. Phys Rev E. 2010;81:041907. [PubMed]
23. Bernasconi C., Konig P. On the directionality of cortical interactions studies by structural analysis of electrophysiological recordings. Biol Cybern. 1999;81:199–210. [PubMed]
24. Ding M., Bressler S.L., Yang W., Liang H. Short-window spectral analysis of cortical event-related potentials by adaptive multivariate autoregressive modelling: data preprocessing, model validation, and variability assessment. Biol Cybern. 2000;83:35–45. [PubMed]
25. Brovelli A., Ding M., Ledberg A., Chen Y., Nakamura R., Bressler S.L. Beta oscillations in a large-scale sensorimotor cortical network: directional influences revealed by Granger causality. Proc Natl Acad Sci USA. 2004;101:9849–9854. [PubMed]
Bosman CA, Schoffelen J-M, Brunet N, Oostenveld R, Bastos AM, Womelsdorf T, Rubehn B, Stieglitz T, de Weerd P, Fries P: Attentional stimulus selection through selective synchronization between monkey visual areas. Neuron, in press [PMC free article] [PubMed]This recent paper applied GC to electrocorticographic data, showing differential top-down and bottom-up directed functional interactions in the beta and gamma frequency bands respectively, across widely distributed cortical areas. These data-driven results align well with influential predictive coding models of cortical processing (e.g. [27]).
27. Friston K.J. The free-energy principle: a rough guide to the brain? Trends Cogn Sci. 2009;13:239–301. [PubMed]
Barrett A.B., Murphy M., Bruno M.-A., Noirhomme Q., Boly M., Laureys S., Seth A.K. Granger causality analysis of steady-state electroencephalographic signals during propofol-induced anaesthesia. PLoS One. 2012;7:e29072. [PubMed]A rigorous application of GC to source-localised EEG data showing increases in GC between cortical midline areas under anaesthesia, particularly in the beta and gamma range. The same data were analysed using DCM in [47•]. Importantly, issues of bias in the GC statistic are explicitly recognized and accounted for.
29. Murphy M., Bruno M.-A., Riedner B., Boveroux P., Noirhomme Q., Landsness E.C., Brichant J.F., Phillips C., Massimini M., Laureys S. Propofol anesthesia and sleep: a high-density EEG study. Sleep. 2011;34:283–291. [PubMed]
30. David O., Guillemain I., Saillet S., Reyt S., Deransart C., Segebarth C., Depaulis A. Identifying neural drivers with functional MRI: an electrophysiological validation. PLoS Biol. 2008;6:2683–2697. [PubMed]
31. Roebroeck A., Formisano E., Goebel R. Mapping directed influences over the brain using Granger causality and fMRI. Neuroimage. 2005;25:230–242. [PubMed]
Wen X., Lau L., Liu Y., Ding M Causal interactions in attention networks predict behavioral performance. J Neurosci. 2012;32:1284–1292. [PubMed]This paper describes a detailed GC analysis of fMRI data in a cued spatial visual attention task, correlating GC from dorsal to ventral frontoparietal regions with behavioural performance.
33. Bressler S., Tang W., Sylvester C.M., Shulman G.S., Corbetta M. Top-down control of human visual cortex by frontal and parietal cortex in anticipatory visual spatial attention. J Neurosci. 2008;28:10056–10061. [PubMed]
34. Schippers M.B., Roebroeck A., Renken R., Nanetti L., Keysers C. Mapping the information flow from one brain to another during gestural communication. Proc Natl Acad Sci USA. 2010;107:9388–9393. [PubMed]
35. Gaillard R., Dehaene S., Adam C., Clemenceau S., Hasboun D., Baulac M., Cohen L., Naccache L Convering intracranial markers of conscious access. PLoS Biol. 2009;7:e61. [PubMed]
36. Friston K.J., Harrison L., Penny W. Dynamic causal modelling. Neuroimage. 2003;19:1273–1302. [PubMed]
37. Stephan K.E., Weiskopf N., Dyrsdale P.M., Robinson P.A., Friston K.J. Comparing hemodynamic models with DCM. Neuroimage. 2007;39:387–401. [PubMed]
38. Penny W.D. Comparing dynamic causal models using AIC, BIC, and free energy. Neuroimage. 2012;59:319–330. [PubMed]
39. Marreiros A.C., Kiebel S.J., Daunizeau J., Harrison L.M., Friston K.J. Population dynamics under the Laplace assumption. Neuroimage. 2009;44:701–714. [PubMed]
Moran R.J., Jung F., Kumagai T., Endepols H., Graf R., Dolan R.J., Friston K.J., Stephan K.E., Tittgemeyer M. Dynamic causal models and physiological inference a validation study using isoflurane anaesthesia in rodents. PLoS One. 2011;6:e22790. [PubMed]A validation of DCMs’ estimates of synaptic coupling using intracranial EEG from rodents was performed using an independent pharmacological intervention. A two-region DCM (comprising for each region a three layer cortical model with pyramidal cells and inhibitory interneurons in supra and infra granular layers and spiny stellate cells in layer IV) was applied to the steady-state summary of a timeseries from auditory cortex under 4 different doses of isoflurane. In line with the hierarchical configuration of electrode placement, DCM revealed a forward-backward architecture that best described interactions between primary auditory cortex and secondary regions, respectively. Moreover a dose effect of isoflurane was observed at excitatory post-synaptic responses that corresponded to a linear decrease, while a nonlinear increase in inhibitory responses was commensurately observed. These measurements conformed to the known physiological effectors of the anaesthetic. These results provide a validation of DCM's ability to reveal the synaptic mechanisms underlying inter-regional brain connectivity.
Daunizeau J., David O., Stephan K.E. Dynamic causal modelling: a critical review of the biophysical and statistical foundations. Neuroimage. 2011;58:312–322. [PubMed]In this recent review the authors outline the commonalities of established DCMs, including those for fMRI and M/EEG. They describe the importance of considering major components to DCM together namely: first, the neuronal model, formulated as a set of ordinary differential equations; second, the static observation model that transforms neural activity to a haemodynamic response, third, the set of priors on model parameters that enable in-built biological plausibility (such as negative self-connections to preclude runaway — epileptogenic activity) and fourth, the variational Bayesian inversion scheme that furnishes both the evidence for a particular model and its optimal parameters after observing the data.
42. Bressler S., Seth A.K. Wiener–Granger causality: a well established methodology. Neuroimage. 2011;58:323–329. [PubMed]
43. Goebel R., Roebroeck A., Kim D.S., Formisano E. Investigating directed cortical interactions in time-resolved fMRI data using vector autoregressive modelling and Granger causality mapping. Magn Reson Imaging. 2003;21:1251–1261. [PubMed]
44. Seth A.K., Chorley P., Barnett L.C. Granger causality analysis of fMRI BOLD signals is invariant to hemodynamic convolution but not downsampling. Neuroimage. 2013;65:540–555. [PubMed]
45. Smith S.M., Miller K.L., Salimi-Khorshidi G., Webster M., Beckmann C.F., Nichols T.E., Ramsey J.D., Woolrich M.W. Network modelling methods for fMRI. Neuroimage. 2011;54:875–891. [PubMed]
Valdes-Sosa P.A., Roebroeck A., Daunizeau J., Friston K. Effective connectivity: influence, causality and biophysical modeling. Neuroimage. 2011;58:339–361. [PubMed]A comprehensive overview of the concepts and mathematics that underlie analyses of directed connectivity. This review surveys the background and differences between structural and DCM, offering a comparative evaluation of GC and DCM.
Boly M., Moran R., Murphy M., Boveroux P., Bruno M.-A., Noirhomme Q., Ledoux D., Bonhomme V., Brichant J.-F., Tononi G. Connectivity changes underlying spectral EEG changes during propofol-induced loss of consciousness. J Neurosci. 2012;32:7082–7090. [PubMed]In fitting a wideband spectrum of responses from source-localised EEG (incorporating slow delta and theta up to gamma oscillations), this report used DCM to test a set of hypothesised model architectures that could explain spectral differences between healthy human subjects during awake, propofol and loss-of-consciousness states. The authors used a “hidden-node” to represent thalamus and test whether it takes part in generating the data from all three states. While it was found that this hidden node was required for an optimal model of the data, the cortico-cortical connections were shown to change with respect to consciousness per se.
48. Li B., Daunizeau J., Stephan K.E., Penny W.D., Hu D., Friston K. Generalised filtering and stochastic DCM for fMRI. Neuroimage. 2011;58:442–457. [PubMed]
49. Stephan K.E., Kasper L., Harrison L., Daunizeau J., Den Ouden H., Breakspear M., Friston K.J. Nonlinear dynamic causal models for fMRI. Neuroimage. 2008;42:649–662. [PubMed]
50. Ryali S., Supekar K., Chen T., Menon V. Multivariate dynamical systems models for estimating causal interactions in fMRI. Neuroimage. 2011;54:807–823. [PubMed]
51. Valdes-Sosa P.A., Sanchez-Bornot J.M., Lage-Castellos A., Vega-Hernandez M., Bosch-Bayard J., Melie-Garcia L., Canales-Rodrigues E. Estimating brain functional connectivity with sparse multivariate regression. Philos Trans R Soc B. 2005;360:969–981. [PMC free article] [PubMed]
52. Friston K.J., Penny W. Post hoc Bayesian model selection. Neuroimage. 2011;56:2089–2099. [PubMed]
53. Marreiros A.C., Kiebel S.J., Friston K.J. Dynamic causal model for fMRI: a two-state model. Neuroimage. 2008;39:269–278. [PubMed]

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • PubMed
    PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...