Linking structural and effective brain connectivity: structurally informed Parametric Empirical Bayes (si-PEB)

Arseny A Sokolov,Peter Zeidman,Philippe Ryvlin,Michael Erb,Marina A Pavlova,Karl J Friston

doi:10.1007/s00429-018-1760-8

Abstract

Despite the potential for better understanding functional neuroanatomy, the complex relationship between neuroimaging measures of brain structure and function has confounded integrative, multimodal analyses of brain connectivity. This is particularly true for task-related effective connectivity, which describes the causal influences between neuronal populations. Here, we assess whether measures of structural connectivity may usefully inform estimates of effective connectivity in larger scale brain networks. To this end, we introduce an integrative approach, capitalising on two recent statistical advances: Parametric Empirical Bayes, which provides group-level estimates of effective connectivity, and Bayesian model reduction, which enables rapid comparison of competing models. Crucially, we show that structural priors derived from high angular resolution diffusion imaging on a dynamic causal model of a 12-region network—based on functional MRI data from the same subjects—substantially improve model evidence (posterior probability 1.00). This provides definitive evidence that structural and effective connectivity depend upon each other in mediating distributed, large-scale interactions in the brain. Furthermore, this work offers novel perspectives for understanding normal brain architecture and its disintegration in clinical conditions.

Highlights

Brain connectivity can be measured or inferred at multiple levels, but integrating these levels poses a significant challenge
We introduced an efficient method for integrating structural and effective connectivity at the group level
By operating at the group level, high-quality diffusion data can be introduced into effective connectivity models—either from the same subjects or from an atlas, such as the dense structural connectivity matrix due to be released by the Human Connectome Project (Van Essen et al 2013)—minimizing the potential for local minima caused by noisy individual data

Summary

Introduction

Brain connectivity can be measured or inferred at multiple levels, but integrating these levels poses a significant challenge. Two regions may lack direct structural (axonal) connections, but, communicate through polysynaptic white-matter pathways (Koch et al 2002). Differences in spatial and temporal resolution between MRI, EEG, and MEG represent another significant challenge. It is, unsurprising that there are inconsistent findings across studies seeking to bridge structural and functional brain connectivity. Straightforward associations have been reported between structural connectivity and fMRI- or MEG-based resting-state functional connectivity (rsFC; Garces et al 2016), as well as between white-matter fibre pathway characteristics and functional connection strength (Hermundstad et al 2013), while other work has indicated a rather complex relationship, with existence of rsFC in the absence of detectable direct structural connections (Honey et al 2009)

Methods

Results

Conclusion