Statistical Kernel-based Modeling of Connectomes

Comprehensive maps of brain connectivity, known as connectomes, have recently emerged as a powerful way to describe and understand global neurological mechanisms. Nevertheless, connectomes suffer from the curse of dimensionality and its well-known consequences. We present here a novel statistical analysis framework for connectomes: machine learning techniques and kernel principal component analysis in order to model a healthy population of reference. This approach enables to analyze global structures and coupled phenomena inside connectomes, contrary to usual and less powerful independent multivariate analysis approaches. Our framework is tested on synthetic data as well as on real connectomes.