11.07.2015 Views

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Roebroeck Seth Valdes-Sosa<strong>in</strong>creas<strong>in</strong>gly used not only to localize structures <strong>in</strong>volved <strong>in</strong> cognitive and perceptualprocesses but also to study the connectivity <strong>in</strong> large-scale bra<strong>in</strong> networks that supportthese functions.Generally a dist<strong>in</strong>ction is made between three types of bra<strong>in</strong> connectivity. Anatomicalconnectivity refers to the physical presence of an axonal projection from one bra<strong>in</strong>area to another. Identification of large axon bundles connect<strong>in</strong>g remote regions <strong>in</strong> thebra<strong>in</strong> has recently become possible non-<strong>in</strong>vasively <strong>in</strong> vivo by diffusion weighted Magneticresonance imag<strong>in</strong>g (DWMRI) and fiber tractography analysis (Johansen-Berg andBehrens, 2009; Jones, 2010). Functional connectivity refers to the correlation structure(or more generally: any order of statistical dependency) <strong>in</strong> the data such that bra<strong>in</strong> areascan be grouped <strong>in</strong>to <strong>in</strong>teract<strong>in</strong>g networks. F<strong>in</strong>ally, effective connectivity model<strong>in</strong>gmoves beyond statistical dependency to measures of directed <strong>in</strong>fluence and causalitywith<strong>in</strong> the networks constra<strong>in</strong>ed by further assumptions (Friston, 1994).Recently, effective connectivity techniques that make use of the temporal dynamics<strong>in</strong> the fMRI signal and employ time series analysis and systems identification theoryhave become popular. With<strong>in</strong> this class of techniques two separate developments havebeen most used: Granger causality analysis (GCA; Goebel et al., 2003; Roebroecket al., 2005; Valdes-Sosa, 2004) and Dynamic Causal Model<strong>in</strong>g (DCM; Friston et al.,2003). Despite the common goal, there seem to be differences between the two methods.Whereas GCA explicitly models temporal precedence and uses the concept ofGranger causality (or G-causality) mostly formulated <strong>in</strong> a discrete time-series analysisframework, DCM employs a biophysically motivated generative model formulated<strong>in</strong> a cont<strong>in</strong>uous time dynamic system framework. In this chapter we will give a generalcausal time-series analysis perspective onto both developments from what we havecalled the Wiener-Akaike-Granger-Schweder (WAGS) <strong>in</strong>fluence formalism (Valdes-Sosa et al., <strong>in</strong> press).Effective connectivity model<strong>in</strong>g of neuroimag<strong>in</strong>g data entails the estimation of multivariatemathematical models that benefits from a state space formulation, as we willdiscuss below. Statistical <strong>in</strong>ference on estimated parameters that quantify the directed<strong>in</strong>fluence between bra<strong>in</strong> structures, either <strong>in</strong>dividually or <strong>in</strong> groups (model comparison)then provides <strong>in</strong>formation on directed connectivity. In such models, bra<strong>in</strong> structures aredef<strong>in</strong>ed from at least two viewpo<strong>in</strong>ts. From a structural viewpo<strong>in</strong>t they correspond to aset of “nodes" that comprise a graph, the purpose of causal discovery be<strong>in</strong>g the identificationof active l<strong>in</strong>ks <strong>in</strong> the graph. The structural model conta<strong>in</strong>s i) a selection of thestructures <strong>in</strong> the bra<strong>in</strong> that are assumed to be of importance <strong>in</strong> the cognitive process ortask under <strong>in</strong>vestigation, ii) the possible <strong>in</strong>teractions between those structures and iii)the possible effects of exogenous <strong>in</strong>puts onto the network. The exogenous <strong>in</strong>puts maybe under control of the experimenter and often have the form of a simple <strong>in</strong>dicator functionthat can represent, for <strong>in</strong>stance, the presence or absence of a visual stimulus <strong>in</strong> thesubject’s view. From a dynamical viewpo<strong>in</strong>t bra<strong>in</strong> structures are represented by statesor variables that describe time vary<strong>in</strong>g neural activity with<strong>in</strong> a time-series model of themeasured fMRI time-series data. The functional form of the model equations can em-74

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!