Salient networks: A novel application to study brain connectivity

Abstract

Extracting meaningful structures and data, thus unveiling the underlying base of knowledge, is a common challenging task in social, physical and life sciences. In this paper we apply a novel complex network approach based on the detection of salient links to reveal the effect of atrophy on brain connectivity. Starting from structural Magnetic Resonance Imaging (MRI) data, we firstly define a complex network model of brain connectivity, then we show how salient networks extracted from the original ones can emphasize the presence of the disease significantly reducing data complexity and computational requirements. As a proof of concept, we discuss the experimental results on a mixed cohort of 29 normal controls (NC) and 38 Alzheimer disease (AD) patients from the Alzheimer Disease Neuroimaging Initiative (ADNI). In particular, the proposed framework can reach state-of-the-art classification performances with an area under the curve AUC = 0.93 ± 0.01 for the NC-AD classification.