Application Research Of Functional Magnetic Resonance Imaging Brain Networks In Schizophrenia

Schizophrenia is a common mental disorder,its main characteristics including thinking,emotion and behavior deficit and dysfunctional mental activity.It has been affected about 1% of the world’s population.At present,there are about 8 million schizophrenia patients in our country,and showed a trend of rising year by year.Individuals with schizophrenia have poorly social function,which brings heavy economic burden to the patients,families and society.Now,the diagnosis method of schizophrenia is mainly based on the patients’ clinical feature,but the method may result in missed diagnosis or misdiagnosis.Therefore,it is a challenge to accurate and fast identify schizophrenia patients.Although the neural mechanism of schizophrenia is unclear,abundant studies indicate that the symptoms of schizophrenia are related to the atypical brain functional network.In recent years,with the rapid development of neuroimaging technology,especially the emergence of functional magnetic resonance imaging,which provides an effective tool to investigate brain functional network.Study of brain functional network has shown great potential in understanding brain functions and identifying biomarkers for neurological and psychiatric disorders.However,most of current functional brain network studies about schizophrenia are based on static brain network using entire low-frequency information,ignoring the fact that different brain networks from distinct frequency bands show different properties and functions,and the time-varying characteristics of brain network.To address these issues,we will studying multi-frequency brain network and dynamic brain network.The main work includes the following several parts in this paper:(1)Most of previous brain network studies of schizophrenia just focus on the correlation between brain regions from the low frequency band,ignoring the specificity of brain networks from different frequencies and the interaction between regions from distinct frequency bands.Therefore,we propose a multiple frequency networks fusion method for schizophrenia classification.At the first,the entire frequency band is divided into four sub-frequency bands,then we construct the sameand cross-frequency brain networks.Next,we use similarity network fusion algorithm to integrate different networks information.Finally,we use the fused network to classify schizophrenia patients.Experimental results show that the proposed method is effective and can be used as aided-diagnosis tool of schizophrenia.(2)Currently,multi-frequency brain network studies are based on static brain network.In this paper,we propose multi-frequency dynamic weighted brain network method.Firstly,the entire frequency band is divided into four non-overlapping sub-frequency bands,then we construct dynamic weighted brain network in each frequency band,and extract the temporal,spatial and spatio-temporal variabilities of dynamic brain networks as the classification features.The experiment results on COBRE dataset show that our proposed method has good classification performance.(3)In this paper,we propose time-varying window length dynamic brain network method.In order to accurate describe the correlation between different brain regions,we introduce the dynamic time warping algorithm,the algorithm can capture the non-stationary time-lags introduced by the dynamic switching of brain states,and can also reduce the effect of noise.For furthermore reduce the effect of noise,we propose orthogonal minimum spanning tree,and use the method into dynamic brain network to get brain network with biological significance.After that,we test our proposed method on COBRE dataset,the experimental results show that our proposed method is effective.(4)Current dynamic brain network studies usually only consider information of each window network,ignoring the relationship among different networks.Therefore,in this paper,we propose dynamic networks fusion method.Firstly,we use distance correlation to describe the correlation between distributed brain regions,because distance correlation can evaluate both the linear and nonlinear dependencies of two signals,and can evaluate both the static and dynamic functional connectivity.Then,improved similarity network fusion is used to integrate different window brain networks.To verify the performance of our proposed method,we conduct experiments on two independent datasets,the experimental results show that the proposed method is robust and can effectively improve the classification performance.