FMRI Functional Connectivity Analysis Based On Sparse Representation

The human brain is the most wonderful existence,promoting us to explore it.The development of brain science deepened our understanding of brain functions and brain diseases.Functional Magnetic Resonance Imaging(fMRI)has been widely used in recent years because of its high temporal and spatial resolution.fMRI functional connectivity is an important method in brain imaging analysis.And it has made great progress in recent years.With the studies of functional connectivity,some novel theories and problems have also been proposed The first one is the recently proposed dynamic functional connectivity,which think that dynamic characteristics of the functional connectivity contain more abundant information.The second one is multi-center fusion problem.A proper multi-center data fusion can help enlarge the sample size as well as make the research more credible.Although many methods have been used to study these theories and problems,they did not consider the sparsity characteristics of brain signals.So we introduced Sparse Representation into the field of functional connectivity analysis.With brain disease classification problem,we completed the following research:We proposed a dynamic functional connectivity analysis algorithm based on sparse representation.We assume that functional connectivity of brain is a linear combination of basic patterns.Sparse representation is used to learn these brain patterns and the linear combination coefficients.We discriminated depression from normal people,with an accuracy of 94.34%.Besides,we extracted specific patterns of functional connectivity networks in depression,which mainly distributed in the default network,affective network and visual cortical area.We proposed a multi-center fusion classification algorithm based on multi-task sparse representation SVM.We calculated the feature weight matrix of multi-center data with the help of both sparse and group sparse.We completed the classification of schizophrenia from normal people with an accuracy of 79.23%.Furthermore,we extracted 14 schizophrenia relevant multi-center shared features.In this pater,the sparse representation is successfully applied to the analysis of functional connectivity.With sparse representation,we solved the problem of dynamic functional connectivity classification and multi-center fusion classification.Moreover,we extracted valuable patterns and features of mental disorders,which if important for the classification and research of mental disorders.