Neuroimaging Analysis And Diagnostic Research Of Schizophrenia Based On Complex Network And Deep Learning

Schizophrenia is a severe psychiatric disorder with complex neural mechanisms,and the cause of it has not yet been determined.Meanwhile,due to that its clinical symptoms involve many aspects such as thinking,emotion and behavior,and the obvious differences between individuals,make it difficult to diagnose.As schizophrenia is a neurodevelopmental disorder,it is of great significance to study the changes inside the brain of the patient to reveal its pathology and help treat the patient.Resting-state functional magnetic resonance imaging(fMRI)is a non-invasive brain image acquisition technology,which can record the baseline spontaneous activity of cerebral nerves.And it is widely used in the study of various brain diseases.In the past decades,complex network theory has been developed rapidly,showing its great potential for characterizing the complex systems and suitable for the multivariate time series analysis.Deep learning technology has a powerful ability to learn effective features from data.It has been widely used in image processing field,but few researches based on fMRI signal.Based on the resting state fMRI data from 128 patients with schizophrenia and 103 matched healthy controls,a series of studies have been carried out in this dissertation.The main works are as follow:(1)By using time series analysis methods,include Pearson correlation coefficient,mutual information and Granger causality test,the model of brain functional network is constructed and used to explore the abnormal changes of brain regions in patients with schizophrenia in resting state.By calculating the network measures,the topological characteristics of the network are quantitatively described.The results show that significant nodes differences at several regions including the prefrontal cortex,hippocampus,temporal gyrus between patients and controls.Furthermore,we construct a pathological subnetwork based on the above brain regions to reveal the changes of functional connections between brain regions.The results showed intra-subnetwork edge strength differences located at the bilateral orbital gyrus and right hippocampus and the connection within temporal gyrus.(2)On the basis of analyzing the changes of brain network nodes and edge strength,the brain changes in patients are further analyzed from the perspective of network structure.We adopt three analysis methods: motif,rich-club and community.The motif fingerprint reveals the changes of the local connection pattern in brain network,and the results showed the topological rearrangement of the network mainly occurs in the frontal lobe,anterior cingulate gyrus and the parahippocampal gyrus.Rich-club property reveals the trend of connectivity between network nodes,and the analysis results show a decrease level of connectivity between hub nodes in patients.The community division of the network reveals the tendency of brain areas to interact with the others.And the results show that hippocampus,parietal hippocampal gyrus and amygdala form a relatively independent functional connection.(3)A convolutional neural network model combined with complex network method is proposed for the diagnosis of schizophrenic patients.Its main idea can be summarized as constructing a network based on mutual information and extracting its adjacency matrix as the input of deep learning model.Then employ the convolutional neural network to extract the features and output the classification results.Our study is based on the 231 subjects’ fMRI data and the verification results showed that the model achieved an average accuracy of 85.2% in the diagnostic tasks of patients,which is improved compared with the previous studies.