Multimodal Mri Fusion Technique And Its Application In Schizophrenia

Multimodal MRI can depict human brain from multiple aspects,e.g.,function connectivity(FC),gray matter(GM)volume and anatomical connectivity,etc.In recent years,multiple types of neuroimaging data(e.g.,fMRI,sMRI and dMRI)collected from a single subject have been widely used to investigate brain disorders,such as schizophrenia(SZ),depression and so on.However,most of recent studies often analyze each modality separately and then simply compare them,resulting in the mutual information among multimodal data not fully utilized.Due to complexity and diversity of brain disorders,the results from unimodal analysis may be relatively high inconsistent,low repeatable,and not comprehensively reflect inter-modal co-alterations.Hence,multimodal data analysis has received an increasing attention.Based on multimodal neuroimaging data fusion technique,this study jointly analyzes the interaction among different modes to help to search more neuroimaging biomarkers,and thereby deepen our understanding of the pathophysiological mechanisms of brain disorders.Clinically,schizophrenia is a severe mental disorders often influenced by a genetic factor,which is clinically complicated and difficult to cure.The main symptoms of schizophrenia patients include hallucination,paranoia,delusion and serious violence,of which the patients themselves are even not aware.Therefore,we expect to take an advantage of a large sample size from multiple collection site and then utilize an advanced multimodal data fusion method to organically integrate multiple types of data,such as brain structure,connections and functions,to mine the complementary and interactive information among different types of data resulting in a deeper understanding of the pathophysiological mechanisms of schizophrenia in order to search imaging biomarkers of mental illness.Based on 4-way multi-set canonical correlation analysis(mCCA)+ joint independent component analysis(j ICA)multimodal fusion method,for the first time,we combine the four MRI features of 605 subjects(307 SZs and 298 healthy controlsubjects,HC)from a large Chinese Han sample collected from 4 sites(hospitals)in China,including regional homogeneity(ReHo)and functional network connectivity(FNC)from fMRI,gray matter(GM)from sMRI,and fractional anisotropy(FA)from dMRI.This study analyzes multimodal co-alterations of schizophrenia from three aspects as below:(1)By performing a two sample t-test on the mixing profiles of each feature,output of multimodal fusion method,we analyze modality-common significant abnormalities between patients with schizophrenia and healthy controls(HC).Our results show that,compared to healthy controls,patients with schizophrenia exhibit higher ReHo,lower GM volume,reduced functional connectivity strength in basal ganglia network(BGN),hippocampus and salience network(SAN),and lower white matter(WM)integrity in tracts of anterior thalamic radiation(ATR)and superior longitudinal fasciculus(SLF).In addition,all above regions were closely “connected”and co-altered in schizophrenia in all of the modalities.(2)To study single auditory hallucination symptom of schizophrenia,we compute the correlation between the mixing profiles of multimodal feature components and auditory hallucination rating scale(AHRS),finding that a thalamo-cortical percept circuit including prefrontal cortex,anterior cingulate,insular,thalamus and superior temporal gyrus was correlated with auditory verbal hallucination(AVH)of SZ patients,and these brain areas were highly overlapped and co-altered in ReHo,GM and FNC component.(3)Given the fact that many studies have demonstrated schizophrenia patients are commonly impaired in cognitive function especially working memory.We also used the digit span task including two modalities: digit forward(DF)and digit backward(DB)to measure working memory capacity.By computing the correlation between digit span scores and the mixing profiles of multimodal feature components,we found bigger GM volume in dorsolateral prefrontal cortex(DLPFC)and medial prefrontal cortex(m PFC)and better white matter integrity in tracts of corticospinal tract(CST),SLF and ATR,corresponding to bigger DB numbers,i.e.,better working memory performance.