Bayesian Inference of Whole-Brain Networks

Abstract: In structural brain networks the connections of interest consist of white-matter fibre bundles between spatially segregated brain regions. The presence, location and orientation of these white matter tracts can be derived using diffusion MRI in combination with probabilistic tractography. Unfortunately, as of yet no approaches have been suggested that provide an undisputed way of inferring brain networks from tractography. In this paper, we provide a computational framework which we refer to as Bayesian connectomics. Rather than applying an arbitrary threshold to obtain a single network, we consider the posterior distribution of networks that are supported by the data, combined with an exponential random graph (ERGM) prior that captures a priori knowledge concerning the graph-theoretical properties of whole-brain networks. We show that, on simulated probabilistic tractography data, our approach is able to reconstruct whole-brain networks. In addition, our approach directly supports multi-model data fusion and group-level network inference.