Construction And Characteristic Analysis Of Schizophrenia Brain Network Based On Resting State FMRI

The lack of biomarkers and the unknown pathological mechanism make the diagnosis and treatment of schizophrenia difficult.At present,studying the pathological mechanism of schizophrenia,determining stable and available clinical biological markers,and developing computer-aided diagnosis technology are the most urgent needs in the study of schizophrenia.Resting state functional magnetic resonance imaging has the advantages of high temporal and spatial resolution,non-invasiveness,and convenient acquisition.It has unique advantages in the study of brain function in mental diseases.At the same time,the rapid development of brain network analysis methods provides an effective means for image-based brain disease research.Based on this,we start from the brain function network construction and feature learning methods,and research on the resting function magnetic resonance data of patients with schizophrenia and healthy control.The main contents are as follows:(1)Analysis of group difference between schizophrenia and healthy controls.In order to accurately locate the abnormal brain areas that are significantly associated with schizophrenia,we analyzed the differences between patients and healthy control groups from two aspects of low-frequency spontaneous neural activity and brain networks.In terms of low-frequency spontaneous neural activity,we use three different indicators to conduct research from three different sub-bands;and in the brain network topology,we use six network topology indicators to study the functional separation and integration performance of the patient’s brain network.The results showed that the lowfrequency spontaneous neural activity in the Calcarine fissure and surrounding cortex,thalamus and lingual gyrus of the patient decreased significantly,and the functional separation and integration performance of the brain function network decreased,and the functional connections between the Middle cingulate & paracingulate gyri and other brain regions have changed most significantly.These abnormalities may provide theoretical support for the determination of the pathological mechanism and biomarkers of schizophrenia.(2)Weighted sparse representation brain network construction and classification.In order to improve the classification accuracy of schizophrenia under the traditional brain network model,we adopted a new weighted sparse representation(WSR)brain network method to model the brain networks of schizophrenia patients and healthy controls,and using functional connections as features,using a variety of feature selection methods for feature selection,and using linear support vector machines for classification.The results show that the WSR model can obtain higher classification accuracy than the traditional Pearson’s correlation(PC)and sparse representation(SR)methods under several different feature selection methods.Under the Kendall correlation feature selection method,the WSR model can obtain the highest classification accuracy of 81.82%,which is 11.57% and 6.61% higher than the highest classification accuracy of PC and SR,respectively.(3)Dynamic brain network construction and classification.Static brain networks(such as PC,SR,WSR,etc.)can effectively capture the functional associations of brain regions,but ignore the rich dynamic interaction patterns of the brain during the scan.In order to fully excavate the dynamic change information of the brain in the data,we used the sliding window method to construct the low-order dynamic brain network and the high-order brain network of the sample at the same time.After that,time variability features are extracted from low-order dynamic networks,network topology features are extracted from high-order networks,and the two features are merged for classification.The results show that compared with static PC functional connection features,single time variability features,and single high-order network topology features,the fusion dynamic features can increase the classification accuracy by 13.22%,9.92%,and8.27%,respectively.