A Study Of Transdiagnostic Time-varying Functional Connectivity And Individual Level Machine Learning Diagnosis In Major Psychiatric Disorders

Objective: Major psychiatric disorders such as schizophrenia,bipolar disorder and major depressive disorder have a high incidence rate,disability and mortality rate,which seriously affect human physical and mental health.For a long time,the diagnosis and evaluation of psychiatric disorders mainly rely on clinical symptoms and signs.At present,there is a lack of objective neurobiological markers to be used in clinical diagnosis,treatment evaluation,prognosis evaluation of major psychiatric disorders.Resting-state functional connectivity has been proved to be an objective and reliable marker of brain network activity.The newly proposed dynamic functional connectivity is a dynamic form of functional connectivity,which reflects more abundant psychopathological information of brain network.The primary purpose of this study is to explore the shared dysconnectivity across major psychiatric disorders by using dynamic functional connectivity.Secondly,we used machine learning method to further explore whether functional connectivity is an effective neurobiological marker for the accurate diagnosis of patients with schizophrenia spectrum disorders,and investigated whether the classification model trained using chronic medicated schizophrenia spectrum disorders identified a trait biomarker that can be used to diagnose early-stage schizophrenia spectrum disorders.Methods: Patients with schizophrenia spectrum disorders,bipolar disorder,major depressive disorder and healthy controls were collected in this study.The subjects were scanned by 3.0T MRI scanner,and their demographic data and various scales were collected,including Hamilton Depression Scale,Hamilton Anxiety Scale,Brief Psychiatric Rating Scale,Young Mania Rating Scale and Wisconsin Card Sorting Test.1.Dynamic functional connectivity analysis: we used sliding-window and Kmeans clustering method to analyze the functional connectivity states of the above three major psychiatric disorders,and investigate shared dysconnectivity across major psychiatric disorders in each state.2.Analysis of functional connectivity in the diagnosis of schizophrenia spectrum disorders at the individual level: using machine learning algorithms such as support vector machine,and taking the functional connectivity of chronic medicated schizophrenia spectrum disorders as the feature to train the classification model.The overall classification performance of the model was evaluated in an independent test set,and the classification performance of the model in each schizophrenia spectrum subgroup was also tested.Results: 1.Dynamic functional connectivity analysis: dynamic functional connectivity can be clustered into three states: one was a more frequent state with moderate positive and negative connectivity(state 1)and the other two were less frequent states with stronger positive and negative connectivity(state 2 and state 3).Significant 4-group differences(schizophrenia,bipolar disorder and major depressive disorder,and healthy control groups;q < 0.05,false-discovery rate [FDR]-corrected)in dynamic functional connectivity were nearly only in state 1.Post hoc analyses(q <0.05,FDR-corrected)in state 1 showed that transdiagnostic dysconnectivity patterns among schizophrenia,bipolar disorder and major depressive disorder featured consistently decreased connectivity within most networks(the visual,somatomotor,salience and frontoparietal networks),which was most obvious in both range and extent for schizophrenia.In addition,we found that some characteristics of dynamic functional connectivity were significantly associated with cognitive function.2.Analysis of functional connectivity in the diagnosis of patients with schizophrenia spectrum at the individual level: We found that although the classification model trained using chronic medicated schizophrenia spectrum disorders from datasets 2,3and 4 identified individuals with chronic medicated schizophrenia spectrum disorders in dataset 1,it did not generalize to first-episode unmediated schizophrenia spectrum disorders.Univariable analysis indicated that medication usage had a significant effect on functional connectivity,but illness duration had no significant effect on functional connectivity.In addition,when subgroup analyses were not used,we also found that functional connectivity combined with support vector machine identified those with schizophrenia spectrum disorders with satisfactory classification performances using other three machine learning strategies: the 5-fold crossvalidation that pooled all datasets,the leave-one-site-out cross-validation and the fivefold cross-validation that included only those with first-episode unmedicated schizophrenia spectrum disorders.Conclusion: 1.The results of dynamic functional connectivity analysis suggest that there are more common dysconnectivity across schizophrenia,bipolar disorder and major depressive disorder than we previous expected,and that such dysconnectivity is state-dependent,which provides new insights of the pathophysiological mechanism of major psychiatric disorders.2.The results of machine learning suggest that the classification model trained using chronic medicated schizophrenia spectrum disorders may mainly identify the state of chronic medication usage rather than the trait biomarker of schizophrenia spectrum disorders.Therefore,we should reconsider the current machine learning studies in chronic medicated schizophrenia spectrum disorders more cautiously in terms of the clinical application and highlight the effect of medication usage on functional connectivity.