Sparse Bayesian network for brain connectivity modeling of Alzheimer's disease

Recent studies have shown that Alzheimer's disease (AD) is related to the alteration in the brain connectivity network. One type of connectivity, called effective connectivity, defined as the causal influence between distinct brain regions, is essential to the brain's functional process. However, little research has been done to model the effective connectivity of AD and characterize its difference from normal controls (NC). We propose sparse Bayesian network (SBN) for effective connectivity modeling. Specifically, we propose a novel formulation in the learning of SBN. Theoretical analysis and simulation studies are performed, both implying that the learning under the proposed formulation is more accurate and efficient than many existing algorithms. We apply the proposed method to the neuroimaging PET data of 42 AD and 67 NC subjects, and identify and compare the effective connectivity models for AD and NC. Our study reveals several connectivity patterns distinctly different between AD and NC, which are consistent with literature findings. New patterns are also discovered which may help the knowledge discovery of AD.