Topological Analysis Of Multiple Templates Brain Networks Based On Relationship Induced Sparse Model

The human brain is one of the complex and powerful systems in the real world.At present,the structural and functional connection patterns in the human brain have become the subject of increasing attention in the field of neuroimaging research.In traditional neuroimaging research and analysis,a brain map is usually divided by a single brain map template to construct a brain network.However,brain network topological features extracted using a single brain map template may not be sufficient to reveal potential differences between patient groups and control groups.Specifically,different brain map templates have a great influence on the structure of the constructed network and its topological properties.In addition,the number of different split nodes of the network has a significant impact on the small world attributes,local attributes,functional connection strength,and network connectivity of the network.At the same time,the impact of different brain map templates on the network is also reflected in the classification of network topology attributes.However,previous studies based on multiple templates brain networks neglected the existence of potential topological association information in the brain network constructedby each brain template when selecting templates.Based on this,this paper proposes a Relationship Induced Sparse multiple templates feature selection model based on parameter self-optimization framework.By establishing different templates and the relationship between different subjects,the corresponding relationship between multiple templates brain regions is mined.The correspondence between the same subjects under different templates characterizes the significant influence of multiple brain map templates on network topology attributes.The main research work of this paper is as follows:Firstly,a Relationship Induced Sparse multiple templates feature selection model based on parameter self-optimization framework is proposed.The Relationship Induced Sparse model was used to extract the correlation between different subjects in the same brain map template and the correlation of the same subjects under different brain map templates.Aiming at the optimization problem of the three parameters involved in the model,this paper constructs a parameter self-optimization framework by combining grid search and random search.To evaluate the efficacy of the feature selection method under this multiple templates,this paper was validated in two groups of experiments based on functional magnetic resonance imaging-based depression classification studies and diffusion tensor imaging based Eysenck personality classification studies.Second,validation was performed in a functional magnetic resonanceimaging-based data set for depression.In this experiment,the Relationship Induced Sparse model of the parameter self-optimization framework was applied to the feature selection of depression subjects.The results show that the dataset collected by functional magnetic resonance imaging has achieved more than the traditional multiple templates method by using the method proposed in this paper.High classification accuracy.The results demonstrate that the Relationship Induced Sparse model based on the parameter self-optimization framework is feasible and effective in the functional magnetic resonance imaging dataset.It also proves that there are potential correlations among multiple brain map templates,and the multiple templates are characterized.The significant influence of the topological attributes of the brain network,in view of the shortcomings of using only a single brain map template and the traditional multiple templates method in the existing brain network field,the breakthrough in the classification research of resting state functional magnetic resonance imaging is realized for depression.The classification of data provides a new reference scheme.Third,it was validated in the Eysenck Personality Data Set based on diffusion tensor imaging.In this experiment,the Relationship Induced Sparse model based on the parameter self-optimization framework is applied to the feature selection of different personality subjects.The experimental results show that the data set collected by diffusion tensor imaging obtains a more traditional template method by using the proposed method.Higher classification accuracy.Based on the above results,this paper affirms the feasibility and effectiveness of the Relationship Induced Sparse model based on the parameter self-optimization framework in the diffusion tensor imaging dataset,and lays a foundation for multiple templates research based on the pathological diagnosis of diffusion tensor imaging.At the same time,it provides strong support for the future breakthrough personality classification research.The research was awarded by the National Natural Science Foundation of China(No.61672374,61873178,61741212,61876124),Shanxi Provincial Natural Science Foundation(201601D021073),Shanxi Provincial Science and Technology Department Applied Basic Research Project(201801D121135),Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi(2016139).Research funding.The research was also supported by the Key Research and Development(R&D)Project of Shanxi Province(201803D31043)and the Ministry of Education's Xaar Network Next Generation Internet Technology Innovation Project(NGII20170712).This paper focuses on the application of the Relationship Induced Sparse feature selection model under multiple templates,and explores the setting of the model parameters.The parameter self-optimization framework is proposed to obtain the optimal combination of parameters.The effectiveness of the method was verified by applying this method to functional magnetic resonance imaging-based depression and Eysenck personality data based on diffusion tensor imaging.This paper provides a new method for the diagnosis of braindiseases,and proposes a research method for the related research of behavioral and neuroimaging.