Recent advances in supervised learning for brain graph classification

Modelling brain networks as graphs has become a dominant approach in neuroimaging. Substantial recent efforts in this area has led to a large number of new methods for analysing such brain graphs. In this paper, we review recent methods for estimating brain graphs and highlight some recent advances in predictive modelling on graphs. We divide the existing methods into three main categories, namely machine learning approaches, statistical hypothesis testing approaches, and network science approaches, and discuss techniques associated with each approach as well as links between the approaches. Graph-based methods have strong roots in pattern recognition, computer vision, social sciences, and statistical physics, and many methods developed for brain graphs are readily transferable to other fields. We thus foresee this methodological upsurge in brain graph analysis will have a wide impact on applications beyond neuroimaging in years to come.