Modeling And Application Of Directed Dynamic Brain Network Based On Autoregression Model

The brain is a complex,large scale network composed of many interacting regions.Knowledge of spatial and temporal organization of brain network is critical for understanding of the relationship between functional brain communication and cognitive behaviors,and further explorating of the neruo basis of neurological diseases.Functional magnetic resonance imaging(fMRI)is widely used to detect the activation of regions in brain.Interregional correlation between fluctuations of fMRI signals in resting state potentially can reveal information about the degree to which regions of brain are functionally coupled together,thereby quantify brain network.However,current methods cannot quantify spontaneous spatiotmporal dynamics of the functional brain network completely.The current study will utilize GC model and Non-homogeneous Markov model(NMM)to quanify the timing and dynamics of information in the brain.It consists of three parts:In the first part,application of Granger causality model in resting-state fMRI data was discussed.The directed effective networks of the group of developmental prosopagnosia patients and controls were constructed,in which nodes defined by 14 face recognition relative regions,and edges were calculated by Granger causality analysis.Network contingency analysis was then performed to determine whether the two group's network exhibited different GC directions or strengths.It lays foundation for extending regular undirected functional network to directed network.In the second part,we aimed to model the plastic and dynamic nature of large-scale functional brain network,and quantify both stability and flexibility of the brain network by using the NMM framework.Firstly,we demonstrated theoretically the Markov process defined in our model is ergodic;thereby the brain system is convergent.Then the maximum reachable probability quantifying the possibly total maximum flow and the corresponding optimal reachable step from one voxel to the remain brain regions were defined.As a computationally traceable model,the optimal steps can be identified from voxels of visual,auditory,and somatosensory regions to other voxels in the functional brain network.As another application of this model,we explored multiple path dynamics between primary sensory regions.In the last part,we futher extended the model of second part to the directed dynamic network.We defined pathway based on dynamic effective connectivity(dEC)rather than dFC to include spatial direction information to quanify information flow dynamics in the brain.The simulation verified that the recognition of wrong paths can be reduced after adding direction information,and thus can better model the the plastic and dynamic nature of large-scale functional brain network.