Brain Funtional Network Study Of Schizophrenia Based On Resting-state FMRI

Chapter Ⅰ:Canonical correlation analysis between brain functional connectivity and clinical symptoms in drug-na（?）ve patients with schizophreniaObjective:Schizophrenia（SCZ）is a common disabling mental disease,but its pathogenesis is still unclear.Functional magnetic resonance imaging（f MRI）is an important method to study schizophrenia,and the canonical correlation analysis（CCA）is a significant method to analyze the correlation between high-dimensional imaging data and clinical data.In this study,we used CCA to explore canonical correlations between Resting-State Functional Connectivity（rs-FC）and clinical behavioral data of drug-na（?）ve patients with schizophrenia.This study aimed to:1)Search characteristic abnormalities of rs-FC in drug-na（?）ve schizophrenia patients at baseline;2)Investigate canonical variables between the altered rs-FC and clinical characterizations in drug-na（?）ve patients with schizophrenia;3)Explore the influence of age and gender on these canonical variables.Materials and Methods:A total of 150 drug-na（?）ve patients with schizophrenia and 185 healthy subjects were enrolled in this study.Data collection mainly contained three aspects:1)General demographic information and clinical information were collected from all schizophrenia patients and healthy subjects.Clinical information mainly included the History of Present Illness（HPI）,Past History of Disease（PHD）,Positive and Negative Symptom Scale（PANSS）and Global Assessment Function（GAF）;2)Imaging data were collected from all SCZ patients and healthy subjects with High-Resolution Structural Imaging（3D-Spoiled Gradient T1-weighted Sequence）and Blood-Oxygen Level-Dependent Imaging（Echo Planar Imaging Sequence）.The processes of data analysis mainly included:1)MRI data preprocessing;2)rs-FC extraction from all subjects;3)Comparing the differences of rs-FC between patients and healthy controls;4)Analyzing the canonical correlations between rs-FC and clinical behavioral data in the patient group;6)Regression analysis with age and gender as covariates.Results:Through the first part of the study,we found that:1)There were some inter-group differences in rs-FC between the patient group and the control group.In other words,rs-FC of patients with schizophrenia showed characteristic changes compared with healthy controls(P_FDR<0.05);2)Through canonical correlation analysis,12 groups of canonical variables between clinical characterizations and rs-FC were obtained;3)Four groups of these typical variables passed the statistical correction;4)These four groups were called as"Activation","Anergia","Depression"and"Paranoid";5)Distinctive changes of these rs-FC were found in each variable,and the most characteristic brain regions were the frontal lobe for the"Activation"group,the parietal lobe for the"Anergia"group,cingulate gyrus for the"Depression"group,and temporal lobe for the"Paranoid"group;6)Regression analysis showed that there were gender differences in the clinical characteristics of the four groups of canonical variables between male and female patients.The clinical symptoms of female patients were more severe than male patients.The severity of disease in these four canonical variables showed some trends with the increase of age.Conclusion:The results of this study confirmed that there were some altered rs-FC among core brain regions in drug-na（?）ve patients with schizophrenia.The main brain regions with abnormalities had been proved to be the core susceptible brain regions of schizophrenia in previous studies.At the same time,in different clinical dimensions,the brain regions with significant changes in functional connectivity had a good corresponding relationship with the clinical symptoms they represented,suggesting that the characteristic changes in functional connectivity at baseline played an important role in the neurobiological mechanism of schizophrenia and led to different clinical manifestations of SCZ patients.Chapter Ⅱ: Predictive study of SCZ patients and healthy controls based on SVMObjective:In the second part,we imported the rs-FC with significant between-group difference obtained in the first part as input-features into the Support Vector Machine（SVM）model to classify drug-na（?）ve schizophrenia patients and healthy subjects,then evaluated the effect of classification.We aimed at: 1)Exploring the predictive efficiency of SVM in patients and healthy subjects;2)Combing with supervised machine learning meothod to confirm the feasibility and accuracy of CCA adopted in the first part of this study.Materials and Methods:The primary procedures in the second part including the following: 1)Constructing training set and test set;2)Building SVM prediction model;3)Importing the functional connectivities obtained in the first step as features to classify for prediction;4)Evaluating the effect of SVM classfication.Results:The second part successfully classified SCZ patients and healthy subjects using SVM.Excellent classified-effect was achieved,of which the accuracy was 71.94%,the sensitivity was 69.33%,and the specificity was 74.05%.Conclusion:In this part,SVM showed an outstanding effect,it is indeed the most excellent model in supervised machine learning.Meanwhile,these results confirmed the powerful performance and stability of canonical correlation analysis.Chapter Ⅲ: Study of functional brain networks in drug-na（?）ve schizophrenia patientsObjective:In the third part,we matched those significantly altered rs-FC with significant between-group difference obtained in the first part to functional brain networks.Then we analyzed attributes of drug-na（?）ve patients with schizophrenia in the dimension of "Activation","Anergia","Depression" and "Paranoid" seperately,as well as the whole attribute of four dimensions.The third part of this study aimed at: 1)Looking for changed attributes within and between brain networks;2)Trying to uncover the brain network biomarkers at baseline in drug-na（?）ve patients with schizophrenia;3)Exploring the association between functional brain network abnormalities and different clinical dimensions in drug-na（?）ve patients with schizophrenia.Materials and Methods:The third part of this study focused on functional brain networks,including the following processes: 1)Network construction by mapping altered rs-FC obtained in the first part into twelve funtional brain networks using the Power-264 template;2)Analyzing changes within each functional brain network（within-network）in each clinical dimension of "Activation","Anergia","Depression" and "Paranoid";3)Analyzing the changes between each pair of networks in each clinical dimension（oneto-one network）;4)Calculating the changes between each functional brain network to all other functional networks in each clinical dimension（one-versus-all-othersnetworks）;5)Holistic analysis for all clinical dimensions: the brain networks with significant changes at the overall level of the four clinical dimensions were calculated as the same as step-2 to step-4,including attribute analysis within a single brain network,between one-to-one networks and one-versus-all-others-networks;6)Statistical analysis.Results:Through the third part of this study,we found that: 1)There were characteristic changes within these networks in each clinical dimension,and there were significant differences among different clinical dimensions.These networks mainly included: The（1）"Activation" dimension contained Visual Network（VN）,Salience Network（SN）,Sensorimotor Network（SMN）and Default Mode Network（DMN）;（2）The "Anergia" dimension contained VN and SMN;（3）The "Depression" dimension contained VN,SMN,DMN and SN;（4）The "Paranoid" dimension contained VN.2)In the one-toone network analysis,there were remarkable changes between networks in each clinical dimension.The（1）"Activation" dimension contained four pairs of altered networks,including Cingulo-Opercular Network（CON）and SN,Subcortical Network（SCN）and Auditory Network（AN）,AN and DAN,DMN and Dorsal Attention Network（DAN）.（2）In the "Anergia" dimension,abnormaility occurred between SMN and AN.（3）In the "Depression" dimension,the connectivities between AN and DAN were changed.（4）In the "Paranoid" dimension,five pairs of networks were altered,inculding SN and SCN,DMN and DAN,DMN and VN,SMN and AN,AN and SCN.3)Among step-1 to step-2,all results passed the statistical correction（PFDR < 0.05）,and the change of Visual Network was the most significant.4)No obvious abnormality was found in one-versus-all-others-networks analysis.Conclusion:In the third part of the study,we found several abnormalities in multiple functional brain networks in drug-na（?）ve patients with schizophrenia.These changes existed in intra-brain networks and also extensively occurred between different networks.Moreover,in different clinical dimensions,the function of altered brain networks corresponded well with the typical symptoms represented by the dimension.These results proved that impairments of functional brain networks are closely related to the pathogenesis of schizophrenia,and that the dysfunction of Visual Network may be a characteristic change in drug-na（?）ve schizophrenia patients and play a key role in the psychopathology mechanism of early schizophrenia.