Higher-order Properties Of Human Brain Network And Its Application In Autism

With advances in computer science,cognitive science,neuroscience,physics,statistics,optical imaging,and more,scientists have unprecedented access to a wide range of tools to study the brain.Considering the complexity,flexibility,and plasticity of the brain,interdisciplinary collaboration is necessary to gain a full understanding of how the brain works and how cognition and consciousness arise.Computer science is one of the foremost representatives of interdisciplinarity thanks to its algorithmic capabilities.This enables researchers to address different research questions,including climate change,sustainability,neural networks,image recognition,etc.This dissertation explores the brain as a complex system,where nodes are brain regions and edges represent the statistical association between brain regions.Unlike previous studies of large-scale brain networks focusing on the lower-order(nodes and connected edges)level,this dissertation focuses on the topological properties of myelin microstructural networks and is dedicated to developing a robust higher-order analytical framework to measure the higher-order relationships between complex systems of different sizes and scales,applying higher-order algorithms to systematically study the complex structure-function interactions of the brain,revealing the fundamental principles of brain structure-function relationships,and applying this higher-order approach to the study of autism to explore its potential theoretical and applied value.First,this dissertation uses kernel density estimation and Kullback-Leibler divergence to construct the morphological brain network based on the myelin map.The results show that individual myelin networks have small-world properties and modular organization and that global topological indices are significantly correlated with individual IQ,indicating that individual myelin microstructural networks can be a useful complement to functional network studies and have biological significance.Second,to solve the insufficiency of analyzing complex networks at the node or edge level,this dissertation develops an effective higher-order network analysis method based on higher-order subgraphs.Previous studies have been focused on the analysis at the node and edge levels,which has yielded valuable research results and provided critical references for our understanding of the mechanism of the brain at the large-scale level and potential abnormal areas in brain diseases but neglected the representation patterns of higher-order structures(motifs,subgraphs,hypergraphs,etc.)in the brain.Thus,this dissertation proposes a framework that combines higher-order subgraphs,mutual information,and statistical correlation to measure higher-order relationships between complex systems of different sizes and scales.The results show that the network of networks based on higher-order features is effective in capturing hidden information in networks.It supplies a methodological preparation for investigating the higher-order characteristics of the brain network.Third,to comprehensively describe the structure-function relationship of the brain,based on the proposed high-order method,this dissertation systematically studies the highorder relationship between brain structural and functional networks and finds new organizational principles of the brain structure-function relationship.Our brain has a relatively static structure,yet it can perform a wide range of cognitive functions(learning,socializing,thinking,sleeping,etc.).This one-to-many relationship is very intriguing.Various research studies have been conducted on structure-function relationships based on DTI and f MRI,but recent studies have shown a gradient variation in the correspondence between myelin covariance networks and functional networks,with the correspondence higher in the visual cortex and the correspondence lower in the association cortex.Since the myelin covariance network is constructed at the population level,it cannot be used to predict individual cognitive behavior.This dissertation examined the higher-order relationships between individual myelin structural and functional networks.The results revealed that higher-order relationships between myelin structure and function decreased with age,and this trend was most prominent in male participants,as well as that simulated attacks on brain hubs could significantly disrupt higher-order structure-function relationships.Furthermore,this structure-function relationship was kept from the whole brain to local neural circuits,supplying key information on understanding brain diseases and the development of brain-like intelligence.Additionally,to evaluate the effectiveness of the high-order analysis method for clinical diseases,this dissertation applied higher-order features to the study of morphological brain networks in autism.Abnormalities of higher-order features were detected in some subcortical regions(caudate nucleus,thalamus,and pallidum),which had also been detected by previous DTI and f MRI studies,demonstrating the value of morphological brain networks.Further,these higher-order features can be fed to machine learning models to achieve good accuracy(86.3%)in the classification of autism subjects and typical controls,proving their utility in the auxiliary diagnosis of diseases.Overall,this dissertation investigated higher-order patterns of brain structure-function interactions using an individual-level myelin microstructural network and applied the higherorder computational framework to the classification of autism participants,which provided new evidence for the higher-order perspective over a lower-order perspective and was valuable both in theory and in application.