Exploiting hierarchy in structural brain networks

Whole-brain structural connectivity matrices extracted from Diffusion Weighted Images (DWI) provide a systematic way of representing anatomical brain networks. They are equivalent to weighted graphs that encode both the topology of the network as well as the strength of connection between each pair of region of interest (ROIs). Here, we exploit their hierarchical organization to infer probability of connection between pairs of ROIs. Firstly, we extract hierarchical graphs that best fit the data and we sample across them with a Markov Chain Monte Carlo (MCMC) algorithm to produce a consensus probability map of whether or not there is a connection. We apply our technique in a gender classification paradigm and we explore its effectiveness under different parcellation scenarios. Our results demonstrate that the proposed methodology improves classification when connectivity matrices are based on parcellations that do not confound their hierarchical structure.