Estimation of Functional Connectivity in Cognitive Impaired Brain Using Non-Homogeneous Dynamic Bayesian Model

Probabilistic graph models have been viewed as an essential tool for learning networks from data. The dynamic functional connectivity using graph models depends on time-varying networks of Functional MRI data. The structure of the brain organization has been studied by exploring functional network analysis. Network-oriented studies on whole-brain connectivity analysis are helpful to infer the dynamics of the brain. Estimation of functional connectivity is a challenging task for large dimensional data. Understanding and discovering the arrangement of network data about the complicated human brain is fundamentally significant for diagnosis. The template-based graph models such as Dynamic Bayesian Networks (DBN) provides a single compact model which captures characteristics of a system dynamics and distributions over time. In this work, non-homogeneous DBNs have been constructed for Default Mode Network (DMN) of three category subjects such as Alzheimer's Disease (AD), Early Mild Cognitive Impairment (EMCI) and Control Normal (CN) subjects. The connectivity interactions and weights between the eleven Region of Interests (ROI) of DMN for three category subjects show a significant difference in the form of information flow between the ROIs.