Functional Brain Network Learning Combining With Structure Information And Its Applications In The Diagnosis Of Brain Diseases

Functional brain networks analysis from functional magnetic resonance imaging data has become a useful way for diagnosis of neurological disorders,such as the prediction of mild cognitive impairment,a prodromal stage of Alzheimer’s Disease,ect.In the estimation of functional brain networks,Pearson’s correlation is one of the most widely-used method for constructing functional brain networks.Despite its effectiveness and simplicity,the conventional Pearson’s correlation-based method usually results in dense brain networks where regions-of-interests are densely connected.This is not in accordance with the biological prior that regions-of-interests may be sparsely connected in the brain.To address this issue,previous studies proposed to employ a threshold orl₁-regularizer to construct sparse functional brain networks.However,these methods usually ignore rich topology structures,such as low-rank structure which can reduce the noise and modularity which can improving the of the stability of brain.Therefore,in this thesis,we mainly focus on incorporating the structure information of brain into the functional brain network estimation model,including the following two works:（1）Low-rank functional brain network learning based on matrix factorization.Low-rank structure can reduce the noise and improve modularity to enhance the stability of networks.However,most of the methods proposed in the existing work ignore this point.To address this problem,we propose a novel and universally applicable functional brain network estimation method utilizing matrix factorization.More specifically,we firstly construct functional brain networks based on traditional methods.Then,updating the rank of these functional brain networks via matrix factorization model empirically.（2）Accurate module induced brain network construction.Modular structure that has been proven to be an important property for improving the information processing ability of the brain.To this end,in this thesis,we propose an accurate module induced PC model to estimate functional brain networks with a clear modular structure,by including sparse and low-rank constraints on the Laplacian matrix of the network.Based on the property that zero eigenvalues of graph Laplacian matrix indicate the connected components,the proposed method can reduce the rank of the Laplacian matrix to a pre-defined number and obtain functional brain networks with an accurate number of modules.To validate the effectiveness of the proposed methods,we use the estimated functional brain networks to identify subjects with mild cognitive impairment and autism spectrum disorder from healthy controls.Experimental results on subjects from the public dataset with resting-state functional MRIs show that the proposed method achieves better classification performance than conventional methods.