Inferring Human Brain Structural Connectivity By Matrix Decomposition

The complex relationship between human brain functional network and structural network remains a central focus of the emerging field of neuroscience.Structural network is composed of the anatomical connectivity of human brain nerve cells.At present,structural network is usually measured by diffusion tensor/spectrum imaging(DTI/DSI),but it has the problems of time consuming and inaccuracy in the imaging process,especially,it is hard to measure the connectivity between two brain hemispheres.While functional network describes the statistical relationship between the functional signals of nodes in a given period of time.At present,resting state functional magnetic resonance imaging(rs-fMRI)has been the main approach of measuring functional network because of its non-invasion,high speed,high precision and high spatial resolution.Many studies indicate that structural network and functional network relate with each other.Besides,although it's hard to get information about structural connectivity between two brain hemispheres by DTI/DSI,we can obtain rich information about functional connectivity between two brain hemispheres by rs-fMRI.Therefore,this thesis attempts to infer structural network from functional network and recover the missing links between two brain hemispheres in the DTI/DSI measurement data.The main work of this thesis is as follows:1.A summary of the main measuring methods of human brain functional network and structural network,and the relationship between them is analyzed from the perspective of matrix;2.A model based on network deconvolution is proposed to infer human brain structural connectivity.The model assumes that observed functional network consists of direct network and indirect network,represented by one-step structural connectivity(SC)matrix and two-step structural connectivity(SC)matrix respectively,and then the method of eigenvalue decomposition is used to solve the model.Experimental results show that the proposed model performs well to infer human brain structural connectivity;3.An algorithm based on spectral mapping is proposed to infer human brain structural connectivity.The algorithm learns the connectivity information of left and right hemispheres to establish a multivariable linear regression model between structural connectivity(SC)matrix and the functional connectivity(FC)matrix and then infers structural connectivity between two brain hemispheres by use of the functional connectivity(FC)matrix of the whole brain.Experimental results,examined on 998-ROI database and 66-ROI database,show that the algorithm can accurately recover the missing links between two brain hemispheres in the DTI/DSI measurement data and the results are verified by the dynamic mean field(DMF)neural mass model.