LAUSR

Purpose The use of Multivariate Granger Causality (MVGC) in estimating directed BOLD connectivity is still controversial, mostly because of the intrinsic low temporal resolution of BOLD data and the slow haemodynamic response of the neurovascular coupling. Characterizing MVGC detection performances in silico could allow us to understand the confidence of the results when applying MVGC in vivo. Methods and materials In this contribution we evaluate MVGC on synthetically generated BOLD data from complex network models based on Izhikevich neuronal populations interlinked by realistic neuronal fiber bundles which exert compounded directed influences and cascade into Baloon-like neurovascular coupling models. We use massively parallel computing to generate synthetic BOLD data from a family of 20-node networks. Each node represents a small cortical volume (see Fig. 1 for a graphical summary). MVGC was then assesed on in vivo fMRI data (HCP 100 unrelated subjects). After pre-processing the average BOLD signal was extracted in 116 regions of interest (ROIs) using the AAL atlas. Results In silico, top 1st percentile of a BOLD connectivity matrix has a PPV very close to 1. Fig. 2 shows the elements of the inter-subject median MVGC matrix between the 116 ROIs extracted from the AAL atlas and the inter-subject median of normalized MVGC. Conclusion We have shown in silico that the, and demonstrate in HCP data that functional interactions between parietal and prefrontal cortices are not symmetrical, but consists of directional connectivity from parietal areas to prefrontal cortices rather than vice versa [1] , [2] .