Clustering Analysis Of Functional Brain Network Based On FMRI Data

The human brain consists of many neurons.The interaction between neurons has completed the various functions of the human brain.Because the number of neurons in the human brain is a lot,it is inconvenient to study them individually.The most widely used method is functional magnetic resonance imaging(fMRI),which is based on BOLD signal.It can study not only the functional brain network of the subjects in resting state,but also the dynamic functional brain network.However,fMRI involves more complex pattern,so model-based statistical method is not suitable.On the other hand,the clustering analysis based on data-driven can analyze fVMRI data more scientifically and objectively.In this thesis,the aim is to find the more suitable clustering method,which can better apply to the analysis of fMRI data,and then promote the understanding of brain function connection.First,we propose a DBI based Hierarchical Initialization K-means(DHIKM),which can choose a good initial clustering center and automatic determine the number of clusters.DHIKM gets good experimental results.Since the sparse representation of the signal can be used to express the fMRI signal very well,many researchers have used sparse representation to study the fMRI,and have achieved good results in recent years.Therefore,sparse representation and DHIKM have been used to study the functional brain network.It is found that the brain images have high-dimensional data structure,and subspace learning can reduce the data dimension and can effectively represent the fMRI signal.Sparse subspace clustering(SSC)algorithm is used to cluster the brain functional network.Meanwhile,the neurons closer to each other in spatial distance have higher probability that they are in one brain functional network.In this thesis,we propose a neighboring adaptive local scale based sparse subspace clustering algorithm and the better results are obtained than SSC.