Abstract

I will review recent methods for inferring changes in effective connectivity within small network models (typically a handful of interconnected regions) using human functional neuroimaging data from fMRI and EEG /MEG. These methods are based on Dynamic Causal Modelling (DCM), in which the parameters of a dynamic input-state-output system are identified within a Bayesian framework, given known deterministic inputs and observed outputs. The outputs (data) are derived from the (hidden) neural variables via modality-specific observer models (i.e, the “balloon model” of haemodynamics for fMRI, or equivalent current dipole models for EEG /MEG). The inputs represent experimental perturbations, with bilinear parameters in particular reflecting changes in connectivity induced by inputs. Inferences based on the posterior density of the connectivity parameters therefore reflect more than just functional connectivity. While those inferences are model-dependent, the Bayesian model evidence allows different models to be compared (particularly useful for EEG /MEG, where the data are fixed). While fMRI data provide greater spatial resolution, EEG /MEG data offer the exciting opportunity to explore regional dynamics and coupling at the millisecond level.