Analysis Of The Effective Connectivity Of Human Brain Network Based On Transfer Entropy

The human brain is a complicated neural network system,which is composed of a plurality of interconnected neurons.To explore the connectivity between different brain regions has important significance for revealing the working mechanism of human brain.Functional magnetic resonance imaging(fMRI)provides a new method for the research of brain science.Due to its non-invasive and high spatial resolution,it has become an important technology for the study of the human brain neural network.Research on the effective connectivity is of utmost importance for understanding causal relationships and information flow between regions of the brain,because the interaction of the brain neurons is thought to be internally directional and,hence,quantifying undirected functional connectivity alone can not fully describe brain dynamics.While Transfer Entropy(TE)offers a model-free approach to explore directed information flow between nodes in brain network and is inherently non-linear.In this paper we apply TE to the detection of the effective connectivity of brain network with 998 ROIs(Regions of interest,or nodes).With the fMRI-BOLD time series we obtain the strength of directed effective connection and analyze its relationship with the structural connections.The main researches and contributions are threefold:1.Summarize the theory and application of the model based methods,which are commonly used for detecting the effective connectivity of human brain,and sum up their existing problems and shortcomings systematically.2.The improved Symbolic Transfer Entropy algorithm is applied to the analysis of the fMRI-BOLD time series signal.Traditional static symbolic method has been improved by using dynamic adaptive segmentation method for symbolizing the time series followed by encoding.It remedied the drawback of the loss of detail information,and described the coupling strength and the information transfer between the two systems effectively.3.A parameterized transfer entropy algorithm is proposed,where Takens’ delay embedding is used to map scalar time series into trajectories in a state space of possibly high dimension.The parameters d and τ are optimized by Ragwitz’ criteria method.Then the nearest neighbor algorithm and Kraskov-St?gbauer-Grassberger estimation algorithm are combined to estimate the value of transfer entropy.An iterative algorithm is used to obtain the prediction time u.The algorithm can effectively estimate the coupling strength between the two regions of human brain and optimize relevant parameters.Because of the estimate of the transfer probability will come with statistical bias,a non-parametrical statistical test,namely permutation test is applied to the verification of the efficacy.Finally,the relationship between the effective connectivity and structural connectivity is discussed.