Demonstrating causal links between fMRI time series using time-lagged correlation

An autoregressive modelling-based technique, Granger analysis, has increasingly been used for measuring causal connections between brain regions in fundamental magnetic resonance imaging brain imaging (fMRI). This paper summarises the possibilities and limitations of applying Granger analysis to fMRI. It also describes a replication of a previous theoretical study, and an application to spatial working memory currently under way. Previous researchers have described methods for detecting time-lagged correlation between neural activations in brain regions of interest (ROI)–often variants of ‘Grangercausality analysis’ (GCA). With appropriate caveats, GCA can draw inferences from time-lagged correlation about effective connectivity between ROIs in a way other popular methods do not. We replicated an existing theoretical model using a different non-linear function, and different method of combining a haemodynamic response function. We then examined whether GCA can estimate the direction of causation in brain regions shown to act together in spatial working memory tasks. Independent Component Analysis (ICA) was used to identify independent spatial components. Then, GCA tested the interactions between those components. The method identified causal relationships on replicated, artificially simulated data and in between extracted components in the data. Work is now underway to determine the implications of these relationships. Keywords—Granger causality analysis; independent component analysis; effective connectivity; fMRI; time-lagged correlation