A graph theoretical regression model for brain connectivity learning of Alzheimer'S disease

Learning functional brain connectivity is essential to the understanding of neurodegenerative diseases. In this paper, we introduce a novel graph regression model (GRM) which regards the imaging data as signals defined on a graph and optimizes the fitness between the graph and the data, with a sparsity level regularization. The proposed framework features a nice interpretation in terms of low-pass signals on graphs, and is more generic compared with the previous statistical models. Results based on the simulated data illustrates that our approach can obtain a very close reconstruction of the true network. We then apply the GRM to learn the brain connectivity of Alzheimer's disease (AD). Evaluations performed upon PET imaging data of 30 AD patients demonstrate that the connectivity patterns discovered are easy to interpret and consistent with known pathology.