Research On Feature Identification Of Dynamic Brain Network System

The human brain functional network which constructed based on resting-state functional magnetic resonance imaging is dynamic network system.In recent years,researchers have attempted to analyze resting-state functional magnetic resonance imaging data from the perspective of complex networks and dynamic characteristics,and the goal is to discover the evolution pattern of brain function network.Many studies using machine learning algorithms have found that dynamic human brain network built with brain network construction techniques presents several different states and features during the whole process of resting-state magnetic resonance imaging data acquisition.For the research of dynamic brain network feature identification,all the factors such as state observation window window setting,threshold selection,brain network vectorization or network similarity measurement may affect the identification result.However,comprehensive researches are still insufficient for these key technologies and impacts.Considerating above situation,several key technologies for feature identification of dynamic brain network systems are studied in this paper based on sliding window technology,this paper focuses on the brain region correlation analysis method in the state observation window,the threshold selection method in the dynamic brain function network construction process,and the brain networks of different time points vectorization method.The integrity and time complexity of the expression of functional connections between regions of whole brain are studied with different correlation analysis methods,then evidence for the selection of correlation analysis methods in the state observation window is provided.A small-world and integrity dynamic brain network threshold determination method is proposed as the theoretical basis for the threshold selection in dynamic brain network construction process.A dynamic brain network feature embedding and weighted topology overlap coefficient similarity measurement method is given,which provides an effectiveexpression method for the identification of dynamic brain network features.Based on these technologies and methods,the resting-state functional magnetic resonance imaging data from the human brain open database were used to carry out experiments,and the identification of dynamic brain network system features and the clustering of states at different time points were completed by DBSCAN.Compared with some machine learning algorithms this method can express the characteristics of brain network at a specific time point clearly,and reflect the similarity of the brain networks at different time points reasonably,More importantly,the temporal properties of the brain network during the identification process are preserved.This work provides a basis for further research on the properties and evolution process of dynamic brain networks.