Research On Complex Network Module Detection Algorithm And Modularization Analysis Of Brain Network

Module(also known as community)structure is an important feature of complex networks.Research on the modular structure helps us understand the organizational principles of complex networks,and has important theoretical significance and practical value in function analysis and behavior prediction of networks.The brain is also a complex system.We can construct the brain network model through brain imaging data,and then studying the brain functions by analyzing network topology.The brain network also presents modular structure,which can help us understand operating mechanisms of brain.In this paper,we proposed a new module detection algorithm of complex networks,and analyzed modular structures of brain networks according to different specific characteristics of brain.Firstly,the Label Propagation Algorithm(LPA)is a very simple module detection algorithm,but the results obtained is not unique,and it may produces meaningless results that all nodes belong to the same module when the module structure is blurry.Based on the basic idea of label propagation,we proposed an improved algorithm based on similarity called Stepping LPA-S,and defined a new evaluation function.Experiments were conducted with both synthetic networks and real-world networks.The results showed that our algorithm obtained more accurate results than LPA and avoided the meaningless when the modular structures were relative fuzzy,indicating that it is a promising method to detect modular structure of complex networks.Second,some research shows that the modular structure of the brain network changes dynamically with time,but existing brain network modularization methods only focus the static partitions of static network.In order to solve this problem,we proposed a consensus modularization method to extract the dynamic information in blood oxygenation level dependent(BOLD)signal through the sliding window,and then integrate the dynamic module structure into an optimized modularization based on the consensus algorithm.Experiments show that,compared with the static method,results of this method is more consistent with the operation mechanism of the brain,and can fully explore the common module structures in networks with different connectivity modes.In the comparison of healthy subjects and Alzheimer's disease(AD)subjects,independent test samples can be effectively classified with consensus results,and group-level results are also consistent with the pathology and clinical symptoms of AD.Finally,researchers has demonstrated that brain network has a multi-scale modular structure.Organizational units of the brain transmit information at different levels to complete various subtle functions.In order to analyze the multi-scale modular structure of brain,we utilized a hierarchical clustering algorithm called LEGClust and set adjacency matrix as the input to detect the hierarchical modular structure of brain network.We verified the validity of LEGClust algorithm through experiments on synthetic brain network.Then this algorithm was also applied in real brain network division,and experimental results showed that there is hierarchical modular structure in brain network and it is in accordance with the robustness of brain and cooperative relationship between brain regions.