Graph Learning Based Functional Connectivity Estimation with Application to Brain Age Prediction

Recently, functional magnetic resonance imaging (fMRI) has enabled the derivation of brain functional connectivity (FC) patterns that serve as unique fingerprints for predicting individual variations in phenotypic characteristics and cognitive impairments. To estimate FC, we present a kernel distance based graph learning (KD-GL) method in this study. Unlike correlation-based FC computing, it learns the graph by taking signals of brain neuroactivity at each point in time as graph signals. We apply the calculated FC to brain age prediction, and the experimental results show that this method is superior to other competing FC estimation methods. In addition, we also observed an interesting biological phenomenon: 8–14 years old females tend to be more precocious than males, that is, the brain age is older. We also looked at brain regions that are associated with brain age or brain maturity.