Table 1.

Glossary and outline of methods discussed.

DCM	Dynamic causal modeling	A deterministic approach within a generative model that characterizes neural activity in terms of driving inputs to a distributed neural network, intrinsic connections, and linear or non-linear modulations of connectivity arising from tasks or neural activity (Friston et al., 2003). Critical features of DCM as implemented by SPM software are the simultaneous estimation of a forward model of neurovascular coupling and the interactions among network regions at the level of neuronal activity. These are estimated to optimize a free energy estimate of the log-evidence of a model in a Bayesian framework. DCM is currently applicable to single subject and group studies of fMRI and M/EEG data, with extensions able to incorporate multiple state representations at each region and stochastic or spontaneous activations. See www.fil.ion.ucl.ac.uk/spm
GCM	Granger causality modeling and Granger causality mapping	These methods examine connectivity in terms of “Granger causality” (Roebroeck et al., 2005, 2009a), emphasizing the role of temporal precedence in the inference of causality. They are closely related to multivariate autoregressive modeling, which like SEM has its roots in econometrics, and can be applied to test anatomically defined neural network models (Granger causality modeling), or explore the interactions between a source region all other regions (Granger causality mapping). An invaluable discussion of issues related to GCM for fMRI data is contained in the exchange between Friston and Roebroeck (see Friston, 2009; Roebroeck et al., 2009a,b). See www.brainvoyager.com
ICA	Independent component analysis	Model-free fMRI analysis which may in some packages also estimate the number of interesting noise and signal sources in the data (McKeown et al., 1998; Beckmann et al., 2005). This approach does not assume anatomical connectivity or directionality of influences within the networks, but component networks can be mapped to task events or contexts. See www.fmrib.ox.ac.uk/fsl/melodic or afni.nimh.nih.gov/sscc/gangc/ica
PLS	Partial least squares	Related to principal components analysis, PLS identifies functionally connected brain networks and can identify subject- or experimental-variables associated with them (McIntosh et al., 1996; McIntosh and Lobaugh, 2004) as well as identifying psychophysiological interactions. See www.rotman-baycrest.on.ca/
PPI	Psycho–physiological interactions	A general conceptual framework in which physiological interactions between regions are modulated by psychological or physiological contexts. It can be used to test hypotheses of effective connectivity (Friston et al., 1997), or explore functional connectivity. However, the term PPI is also used to refer to a specific implementation within general linear models (PPI-GLMs). These PPI-GLMs use moderator variables that express the interactions between regional activations and contexts (and higher order interactions with between-subjects factors like age or disease risk factors) (Buchel and Friston, 1997; Rowe et al., 2006; Passamonti et al., 2009). See www.fil.ion.ucl.ac.uk/spm
RSN	Resting state networks	ICA of fMRI data acquired at rest identifies a small number (∼10) of consistent spatially distributed covarying brain networks. One of these is also commonly identified by the brain state when not engaged in typical experimental tasks, known as the default mode network (DMN)
SEM	Structural equation modeling	Introduced into neuroimaging from econometrics and social sciences for the analysis of brain effective connectivity analysis (McIntosh and Gonzalez-Lima, 1994; McIntosh et al., 1994), to determine task-dependent (McIntosh et al., 1994; Buchel and Friston, 1997; Honey et al., 2002) or group-dependent (Grafton et al., 1994; Horwitz et al., 1995; Rowe et al., 2002b) changes in a hypothesized causal structure formalized in a path model. Commonly implemented for fMRI data by LISREL or SPM toolbox software.