Screening For Brain Network Diagnostic Biomarkers Of High-functioning Autism Spectrum Disorders By Data-driven Methods

?Objective?Human brain is an extremely complex network where segregation and integration are basic rules governing brain functions e.g.,cognition,learning,emotion and memory.That is the regularity aspect of brain among populations.Meanwhile,variability across individuals is as prominent as is regularity among populations.A natural yet fundamental question is: how to quantitatively and effectively measure the regularity of human brain architectures,with full consideration of the enormous individual variability? This question is of great significance in terms of building correspondences across subjects,brain network construction,and comparing and reporting brain imaging data across populations.In response to this critical question,this thesis performed systematic research on the individualized representation of common brain architectures via a ROI(region of interest)optimization strategy using multimodal brain imaging data including Diffusion Tensor Imaging(DTI)and functional MRI(fMRI).Specifically,this thesis:?Methods?Study1: The whole-brain white-matter structural network study of Highfunctioning autism spectrum disorder in children and adolescents based on graph theoryThis study used diffusion tensor image tractography to construct the human brain white-matter networks of 36 high-functioning ASD children and adolescents ranging from 6 to 16 years old and 39 age and gender matched TD controls,followed by a graph theoretical analysis to establish topological structures of the cerebral anatomical network,and further to explore the relationship between topological properties and clinical symptoms of the high-functioning ASD group.Study2: Brain Function Network Study of HF-ASD in children and adolescents Based on DICCCOL ModelIn order to establish a precise relationship between the structure and function of different individuals' brains,we combined DTI and rsfMRI data to study the brain functional network of 37 HF-ASD patients and 33 controls based on white matter structure to finding more consistent biomarkers.In the present paper,we use a connectomes-scale assessment of structural and functional connectivity in ASD via multimodal DTI/fMRI dataset.We first used DTI-derived structural profiles to explore and tailor the most common and consistent landmarks DICCCOLs,then applied them in a whole-brain functional connectivity analysis.The next step fused the results from two independent datasets together and resulted in a set of functional connectomes with the most differentiation power,hence named as “connectome signatures”.Study3: Data-driven Exploration of Functional Connectivity Based Biomarkers in Children and Adolescents with High-functioning Autism Spectrum Disorder using dictionary and sparse coding technologyThis study examined the empathy of 80 individuals with HF-ASD and 50 matched controls by using the Griffith empathy measure parent ratings(GEM-PR)and resting state functional magnetic resonance imaging(rsfMRI)dataset.HF-ASD patients and control group were studied to explore stable biomarkers for HF-ASD patients.A set of functional networks were extracted via an effective data-driven approach for functional brain network identification-dictionary learning and sparse coding(DLSC)in a groupwise manner.Then,the localized common functional brain networks of both ASD and matched control groups were automatically decomposed into a set of regions of interests(ROIs)for further functional connectivity analysis.With the derived functional connectivity matrix,we investigated three measures including correlation,partial correlation and tangent embedding for differentiating ASD from controls.?Results?Study1: The whole-brain white-matter structural network study of Highfunctioning autism spectrum disorder in children and adolescents based on graph theoryCompared with TD group,the global topological properties of the brain whitematter networks in high-functioning ASD children and adolescents are abnormal,as indicated by decreased local efficiency(Eloc)(P = 0.02)and the coefficient cluster(?)(P = 0.00).There were sixteen hub nodes in the high-functioning ASD group and all of them were the same as the TD controls.The right caudate nucleus is one of the hub nodes in TD group,but it is not the hub node in high-functioning ASD group.And the classification accuracy,sensitivity and specificity of the two groups are 60%,59.6%and 60.7% respectively based on the global topological properties of the brain network.Study2: Brain Function Network Study of HF-ASD in children and adolescents Based on DICCCOL ModelOur results indicate that these “connectome signatures” have significantly high HF-ASD vs-controls classification accuracy,reached 91.43%,sensitivity was 97.29%,specificity was 84.85%.The HF-ASD group has 30 high-recognition functional connections Increase and 10 functional connections decreased.Through functional meta-analysis,we found that high-recognition functional connections not only come from the same network,but also from different networks.Cognitive-cognitive,cognitive-emotional,and emotional-emotional networks show higher participation.Study3: Data-driven Exploration of Functional Connectivity Based Biomarkers in Children and Adolescents with High-functioning Autism Spectrum Disorder using sparse coding technologyWe achieved classification accuracies of 95%,95% and 100%,respectively,which indicates that the proposed DLSC method can extract representative and characteristic brain ROI atlases for both ASD and controls.Further analysis of functional connectivity results showed that high functioning ASD(HF-ASD)have multiple abnormal connections,especially the ones connecting left inferior temporal and left inferior parietal which belong to temporoparietal junction(TPJ),and the connections related to right insula and anterior cingulum which belong to the salience network(SN).Our results suggested that ASD individuals have lower empathy capability than controls and this abnormality may be linked to the damage of the salience and social brain networks.?Conclusions?1.First,we find that the high-functioning ASD children and adolescents exhibited“small-world” character of the whole-brain white-matter structural network.Compared with the TD group,HF-ASD group showed potential decreased efficiency of local information processing in the whole brain.Using the AAL template to classify the HFASD and the control group based on the graph-based topological attribute values of the whole brain network,the accuracy is not high.2.Then,based on the study of the whole brain white matter structure,using white matter structure to locate cerebral cortex nodes,and the FLIRT mapping function in FSL software was used to locate the functional network nodes.The functional "connectome signatures" was used to classify HF-ASD and control group,and the accuracy could reach 91.43%.The two-stage feature extraction process can effectively classify HF-ASD and locate the brain network area damaged in HF-ASD.Our work provides support for using functional "connectome signatures" as neuroimaging biomarkers of ASD.3.Finally,we increase the sample size and locate the nodes of functional network research based on data-driven sparse coding and dictionary learning methods,the functional connections based on this method can be used for HF-ASD classification,and the accuracy rate can reach more than 95%.The potential biomarkers screened mainly fall in brain network of SN,JPG,SCN,DMN and so on.The empathy ability of HF-ASD patients is lower than that of the control group,and there may be a potential correlation between this abnormal empathy ability and the damage of salience and social brain networks.