Dynamic network partition via Bayesian connectivity bi-partition change point model

Dynamic functional interaction has received much attention recently in the field of neuroimaging. Past studies reveal that the dynamics of functional interactions only exists in part of brain. In this paper, a novel Bayesian inference model is developed to bi-partition the brain regions into dynamic/stable sub networks and to simultaneously segment the temporal sequence of dynamic network into several states based on the interaction dynamics among regions. The accuracy of the model has been verified by synthesized data. Also, the model has been applied to a working-memory task-based fMRI dataset and interesting results on both dynamic network and change points were obtained.