Dual Constraints Analysis Of Brain Functional Connectivity Applied To Resting-state FMRI Data

Functional magnetic resonance imaging(fMRI)is a kind of important brain imaging technique.Independent component analysis(ICA)can obtain activated spatial maps in brain and corresponding time courses from collected fMRI data,which can be used for brain functional or effective connectivity analysis and then provide reference for brain function research and the diagnosis of some brain diseases.At present,several methods for brain connectivity analysis are commonly used,such as the Lag-shift for functional connectivity,the Granger causality and Bayesian network(BN)for effective connectivity,and so on.However,a major problem of the existing analytical methods is that these methods are independent of each other,whether for the same or different fMRI data.Therefore it makes the acquired network connections for different methods are independent as well,hard to compare,and furthermore lack of stability evidence.To this end,this thesis analyzed the resting-state fMRI data of 82 subjects by using some typical methods for brain connectivity.The stable network connections were obtained by imposing appropriate parameter constraints and building bridges for different methods.The main work is as follows:(1)The Group-ICA algorithm was applied with high model order to extract brain regions from the 82 subjects’ resting-state fMRI data.The time courses of seven sub-components of default mode network(DMN)were applied to functional connectivity analysis.After the functional connectivity analysis for each subject,single sample t-test was conducted for the results of 82 participants.This thesis chose the significant connection with p < 0.05 for the dual constraints analysis.(2)In order to solve the problem of functional connectivity and effective connectivity mutually independent,seven sub-components of DMN were extracted.Lag-shift analysis and Granger causality analysis were separately conducted for seven sub-networks.This thesis proposed a dual constraints analysis method with delay constraints for Lag-shift and causality coherence constraints for Granger,and successfully established the relationship between functional connectivity and effective connectivity.(3)This thesis then studied structure learning algorithm of the sparse Bayesian network(SBN)and implemented it for resting-state fMRI data.The results of the simulation and actual fMRI data showed that the performance of SBN is superior to the existing BN structure learning algorithm.To solve the problem of different effective connectivity methods mutually independent,a dual constraints analysis method of Granger causality with causality coherence constraints and SBN with regression coefficient constraints were conducted and then verified the conclusion of obtaining the consistent network.(4)Based on the stabilization network connectivity extracted from 82 subjects by the dual constraints analysis of Granger and SBN,effective connectivity analyses were conducted for 40 healthy controls group and 42 patients with schizophrenia group respectively.The results showed that the effective connectivity within the stabilization network connectivity can obtain significant differences between schizophrenia subjects and healthy controls,while the results showed no significant abnormalities within non-stabilization network connectivity.This thus has a broad application prospect in computer aided diagnosis of brain diseases.