Magnetic Resonance Imaging-Based Mechanism Of Schizophrenia: Individualized Brain Age And Large-Scale Brain Networks

BackgroundThe etiology and symptomatology of schizophrenia(SCZ)are significant heterogeneity,resulting in clinical diagnosis,treatment,prognosis affected by many factors.The reason is that the pathogenesis is still unclear,and there is a lack of effective objective biomarkers to further increase the understanding of the etiology and promote the development of therapeutic methods.From a network perspective,as a disconnection disorder,the neurobiological signature of SCZ in the brain is unclear,resulting in a lack of reliable biomarkers and difficulty in identifying therapeutic targets.Therefore,finding effective biomarkers has been a research hotspot in recent years,among which neuroimaging studies based on magnetic resonance imaging(MRI)are widely used.At the same time,the development of technologies such as machine learning,spectral dynamic causal modeling(DCM)technology make it possible for individualized brain age and large-scale brain network research to discover potential biomarkers.In terms of structural network,the SCZ has extensive white matter connectivity abnormalities.Since early studies have observed that the brain and other organs of SCZ patients age to a greater extent than the normal population,scholars have proposed that SCZ is a disease of accelerated aging,that is,SCZ is associated with accelerated brain aging,which may indicate an increased risk of neurodegenerative diseases and death.But how to characterize brain aging in living samples is a crucial question.In recent years,under the rapid development of machine learning strategies,MRI-based image data analysis can describe the characteristics of brain aging in vivo—brain age.However,structural network-based SCZ brain age prediction remains to be studied.In terms of functional networks,the triple network model was proposed by Menon et al.to understand the relationship between brain network connectivity and cognitive and psychopathological phenotypes.The central executive network(CEN)and the default mode network(DMN)are considered to have antagonistic functions,and there is a negative correlation;while the salience network(SN)coordinates the switching between CEN and DMN.Disturbances between these three networks may cause SCZ patients to develop abnormal mappings when faced with internal and external stimuli,leading to specific symptoms.Thus,the triple network model has been proposed as a common framework to understand dysfunctional brain dynamics in neuropsychiatric disorders,but how the triple networks are connected between and within the SCZ is unclear.AimsExperiment 1: To explore whether the prediction of brain age could be used as a biomarker for the diagnosis of SCZ,and whether there was a correlation between clinical features and the predicted age difference(PAD)of the brain.Experiment 2: Based on the brain age prediction model constructed in Experiment 1,a follow-up study was conducted in patients treated with standard antipsychotic drugs to explore the effects of drug treatment on predicted brain age,clinical symptoms,and white matter fiber tracts in patients with first-episode SCZ.Experiment 3: To further explore the dysregulated cross-network interactions among the triple networks and how they contribute to the different symptoms of SCZ patients.MethodsBased on diffusion tensor imaging(DTI)and blood oxygen level-dependent,functional MRI(BOLD-f MRI),patients with SCZ and healthy controls(HC)were included in this study.In Experiment 1,the testing dataset was 523 HCs,the principal dataset was 60 first-episode SCZ patients and 60 matched HCs,and the replication dataset included 40 first-episode SCZ patients and 40 HCs;experiment 2 included 21 first-episode SCZ patients and matched 21 HCs;experiment 3 included 76 SCZ patients and 80 HCs.Experiment 1: A brain age prediction model was established based on machine learning methods using DTI of 523 healthy people.DTI data of patients and HCs and clinical information(severity of symptoms and cognitive status)of patients undergoing MRI scans were collected.The average fractional anisotropy(FA)value of 50 regions of interest(ROI)(ie,50 white matter fiber tracts)was extracted using the template,and a predictive model was introduced to calculate the brain PAD values ??of all subjects.Differences in PAD values ??between groups were compared,and correlations between PAD values ??and clinical measures were analyzed.Using the brain age model,DTI data for patients and HCs and clinical information for patients undergoing MRI scans were acquired in a replication dataset(different from the image acquisition device of the principal dataset).The average FA value of 50 ROIs was extracted,and a predictive model was introduced to calculate the brain PAD value of all subjects.Compare whether there are differences in brain age between patients and HCs,compare whether there are differences in the estimation of PAD values ??by different image acquisition equipment,and analyze whether there is a relationship between brain PAD values ??and clinical characteristics.Experiment 2: Using the brain age prediction model constructed in Experiment 1,the baseline DTI data and the clinical information when they underwent MRI scans of the first-episode SCZ patients were collected.The patients were treated with antipsychotic drugs for about 4.7 months,and the DTI data and clinical information of the patients were collected again.The average FA value of 50 white matter fiber ROIs was extracted,and the prediction model was introduced to calculate the brain PAD value of the subjects.Differences in brain PAD values ??between pre-and post-treatment patients were compared using paired-samples t-tests,and correlations between PAD values ??and clinical measures before and after treatment,as well as the effect of antipsychotic drugs on white matter fiber,were analyzed.Experiment 3: ROIs within DMN,CEN,and SN were determined using independent component analysis;according to the peak coordinates of the core area in each network,11 ROIs were selected;and time series were extracted using a general linear model.DCM was used to investigate effective connectivity between triple networks and further analyze the relationship between network dynamics outcomes and the positive and negative syndrome scale(PANSS)in patients with SCZ.ResultsExperiment 1: In the principal and replication datasets,compared with the HC group,the PAD values of SCZ patients in both groups were significantly increased;the PAD values of SCZ patients in the principal dataset were significantly higher than those of SCZ patients in the replication dataset.Correlation analysis showed that there was a certain correlation between PAD value and age,gender,education level and disease duration;PAD value was negatively correlated with the positive symptom score of PANSS in the principal dataset,but not significantly correlated with cognitive status.Experiment 2: After antipsychotic treatment,the brain PAD values of the 21 patients who were followed up in the principal dataset were significantly decreased compared with the baseline.The FA values of 31(33 in total,defined as "features")white matter tracts correlated with brain PAD values were statistically significantly different before and after treatment,that is,the integrity of white matter fibers was improved.The brain PAD differences of pre-and post-treatment were positively correlated with baseline PANSS negative symptom scores.Experiment 3: The HC group and SCZ patients showed common network connectivity within and between networks of DMN,CEN,and SN in the resting state.By comparing these common network connections,we found that SCZ patients had significantly reduced SN-centric cross-network interactions;The strength of connectivity from CEN subnet 1 to DMN subnet 1 was positively correlated with PANSS positive symptom scores.Connectivity from DMN subnet 2 to CEN subnet 2 was negatively correlated with PANSS negative symptom scores.Conclusions(1)Using machine learning technology to mine structural MRI imaging data,first establish a normal human brain age prediction model,and on this basis,extract the PAD values ??of SCZ patients and HCs,and further conduct longitudinal follow-up of SCZ patients.Combining clinical and imaging data to clarify the evolution law of PAD value,an indicator of brain aging,and a study on the mechanism of SCZ drug treatment based on PAD value was established to provide objective imaging markers for individualized diagnosis and treatment of SCZ.The PAD value,an indicator of brain aging,was introduced to explore its changes in different time courses in SCZ,and to explore the relationship between the PAD value and the outcome after treatment.The brain age of the first-episode SCZ patients is greater than chronological age,i.e.,there is disease-related accelerated brain aging.Antipsychotic treatment is expected to improve the extent of accelerated brain aging in patients with first-episode SCZ by improving white matter microarchitecture.Neuroimaging-based brain age prediction can be used as an individualized biomarker,providing new insights into the neural mechanism of SCZ and the mechanism of antipsychotics.(2)Using the spectral DCM approach,this experiment provides robust evidence for dysregulated brain dynamics between DMN,CEN,and SN in SCZ patients from the triple-network perspective,and provide a theoretical description of abnormal brain dynamics associated with psychiatric disorders,and has the potential ability to predict differences of patients’ clinical symptoms.Abnormal dynamic functional interactions among triple networks are significant and robust features in first-episode SCZ,where the link between DMN and CEN may be a clinically relevant neurobiological signature of SCZ symptoms.The findings suggest that the description of the brain triple network hypothesis may help to discover potential biomarker of SCZ.