Functional Network Based On FMRI And Connectivity Association Pattern

MRI promotes the computability of neurophysiology and neuroanatomy and provides a simple and reliable method to diagnose several neuropsychiatric diseases.MRI assists us to analyze brain function and morphology alterations in terms of brain issue,connectivity and circuit,and investigate mechanisms of brain functions based on multiple scales,different fields and multiple modalities analysis.In addition,it is possible to investigate the dynamic changes of brain activity during data acquisition based on fMRI.In this thesis,the main work consists of three aspects and the details are as follows:(1)Independent component analysis,time-varying connectivity analysis and discretization method were applied to the study of the brain activity sub-state changes in autism spectrum disorder(ASD).First of all,fMRI data was subjected to standard preprocessing.Then,brain regions were identified by independent component analysis after taking into account the limitations of brain atlases.Afterwards,time series were segmented by sliding windows and functional networks were constructed with segmented time series.Finally,after the network construction,Graphical LASSO algorithm was used to control the sparseness of the functional network,and K-means clustering algorithm was used to determine the sub-state of brain activity.The results showed that the mean dwell times of some sub-states in brain activity were significantly different between ASD patients and normal subjects,and the percentage of connectivities with different strength in some brain states of ASD patients fluctuated more violently than that of normal subjects.Similar to previous studies,the results showed that some functional connectivities in ASD patients were abnormal and these abnormal connectivities were primarily associated with cognitive control,visual,and default mode networks.(2)The abnormal connectivities in the functional network were found by two sample t-test and leave-one-out cross validation,and Bayesian network,Logistics regression and KNN with ten-fold cross validation method were used to explore the reliability of stationary function connectivity to distinguish the ASD and the normal.The results indicated that the inter-group different connectivity has a better ability to distinguish ASD and normal subjects and classification accuracy can reach 90.7%.In addtiton,the influence of discretization and reduction algorithm on classification results is also explored(3)The frequent itemsets mining algorithm was used to analyze the association of functional connectivities on brain region scale.The results revealed coexisted connectivities among left inferior frontal gyrus: orbital part,right inferior frontal gyrus: orbital part,left hippocampus,left parahippocampal gyrus,right parahippocampal gyrus,left olfactory cortex,right olfactory cortex,left temporal pole: superior temporal gyrus,left inferior occipital gyrus.To some extent,the results might indicate that the brain spontaneous activity was induced by memory rather than random.This study might help to understand the pathology of neuropsychiatric disorders and contribute to the automated diagnosis of neuropsychiatric disorders.