Dynamic Brain Functional Network Analysis Of Rs-fMRI Based On Tensor Decomposition

The development of imaging technologies such as functional magnetic resonance imaging(fMRI)provides an effective way of exploring brain functional activities with high temporal and spatial resolution.It becomes a hot spot in brain imaging research that through brain imaging data to further construct brain networks and analyzing brain functions based the brain connection in recent years.Considering most of the previous methods based on the stationary assumption,which cannot reflect the time-varying characteristics of the brain functional network.Therefore,we propose an orthogonal sparse tensor decomposition method to analyze dynamic functional network based on the hypothesis that resting-state functional brain network states are formed by overlapping different functional subnetwork patterns.By factorizing the roi×time×subject tensor into several rank-one components,we utilize the variation of components’ weights to detect the change points and further estimate dynamic functional network.Finally,the validity of the model is verified on the simulated and PNC datasets,respectively.The main work of this paper includes:1.We designed an orthogonal sparse tensor decomposition model.Considering the high spatial-temporal resolution and high dimension of fMRI data,the tensor CP decomposition model with sparse constraints is introduced.The L2,1-norm regularizer(i.e.,group sparsity)is enforced to select a few common features among multiple subjects.The L1-norm regularizer is enforced to obtain sparse representation of time matrix.The orthogonal constraint is enforced to ensure the linear independence among ROI nodes.2.We analyzed the dynamic characteristics of brain functional network.After tensor decomposition,we obtain different component matrices.According to the change of the weight of the time matrix,the change points are detected and different functional network states are divided.The dynamic brain functional network is constructed according to ROI matrix.Compared with the experimental results of other methods,the results show that the relative error of the proposed algorithm is smaller,and the more significant dynamic characteristics of the functional network can be captured.Compared the proposed algorithm with other method,the experimental results show that our method has smaller relative error and can capture more significant dynamic characteristics.3.We analyzed age-related brain functional network characteristics.Considering the influence of age,we divided participants into children group and young adults group,and analyzed the characteristics of brain functional network at different scales with age.The results showed that the connection patterns within and between different brain regions have undergone subtle changes with age.Generally speaking,the brain tends to be more integrated with age.