Parcellation-independent multi-scale framework for brain network analysis

Abstract

Structural brain connectivity can be characterised by studies employing diffusion MR, tractography and the derivation of network measures. However, in some subject populations, such as neonates, the lack of a generally accepted paradigm for how the brain should be segmented or parcellated leads to the application of a variety of atlas- and random-based parcellation methods. The resulting challenge of comparing graphs with differing numbers of nodes and uncertain node correspondences has yet to be resolved, in order to enable more meaningful intraand inter-subject comparisons. This work proposes a parcellation-independent multi-scale analysis of commonly used network measures to describe changes in the brain. As an illustration, we apply our framework to a neonatal serial diffusion MRI data set and show its potential in characterising developmental changes. Furthermore, we use the measures provided by the framework to investigate the inter-dependence between network measures and apply an hierarchical clustering algorithm to determine a subset of measures for characterising the brain.