Age Prediction Model Based On Decomposition Of Brain Network Line Diagram

Age prediction is a method of using biomarkers or neuroimaging data to estimate an individual’s physical or mental age.Age prediction can be used to assess aspects of an in-dividual’s health,cognitive abilities,aging speed,etc.Because the various regions of the brain are not independent of each other,but are connected with each other and influence each other to form a network to play their role.Therefore,in recent years,based on various neuroimaging information collected,such as diffusion tensor imaging(DTI),functional magnetic resonance imaging(f MRI),etc.,to construct structure or function brain network on.Specifically,each brain region of the brain is first considered as a node.Then,if there is a structural or functional connection between the two brain regions,connect them with an edge,and the weight value of this edge represents the connection strength of the two brain regions.Then the topological properties of their brain networks were extracted,such as characteristics such as small-worldness,modular organization,highly connected or concentrated hubs of the network,to explore different age-related network topologies.This method often only focuses on the spatial topology of brain networks,but never con-siders whether brain networks also have topological frequency information.That is,under the constraints of the topological structure(i.e.not belonging to European data)for each connection between the brain regions,how the connection values between the brain re-gions change according to their topological structure.In order to further verify our conjecture,this paper proposes a brain network-based line graph decomposition model to explore the application of topological frequency infor-mation in brain connection networks in age prediction tasks.The line graph decomposition model proposed in this paper is a method based on the theory of graph signal processing,which can construct sub-brain networks of different topological frequency bands.This pa-per takes the brain connection network constructed by DTI and f MRI data as an example,selects the DTI and f MRI data of 387 subjects matched by the public dataset Cambridge Center for Aging Neuroscience(Cam CAN),the following work was done:(1)This paper first preprocesses the DTI and f MRI image data and constructs the structural connection network and functional connection network respectively.(2)Then convert the two types of connection networks into line graphs,and perform graph Fourier transform on the basis of the line graphs to obtain topological frequency information,thereby reconstructing sub-brain networks in different topological frequency bands.And based on the community detection algorithm,it is found that these sub-brain networks form different connection patterns.For example,in the sub-brain network of DTI low topological frequency band,the left putamen,inferior prefrontal gyrus,insula and amygdala,and the left and right temporal poles Subcortical structures such as the up-per gyrus and parietotemporal gyrus and temporal lobe structures play an important role in the entire network?while in the sub-brain network of f MRI low topological frequency bands,the left and right putamen,hippocampus,amygdala,caudate nucleus and Subcor-tical structures such as the parahippocampal gyrus and limbic system structures such as the parahippocampal sulcus are more closely related to other brain structures and play important roles in the overall network.(3)Finally,based on the reconstructed sub-brain networks in different topological frequency bands,this paper extracts common network topological features such as node strength,global clustering coefficient,local clustering coefficient,and local assortment,and extracts them based on mutual information.The characteristics of interactive infor-mation between sub-brain networks of different modalities and different topological fre-quency bands.Finally,after feature selection,the features are input to various regressors for age prediction,and the optimal mean absolute error(MAE)is only 6.35,the coefficient of determination R~2=0.76,and the root mean square error(RMSE)is 8.15 and the indi-cators(MAE,R~2,RMSE)based on the brain network line diagram decomposition model are better than the traditional method of age prediction using brain network on multiple learners.Based on the brain network line diagram decomposition model,this paper found that sub-brain networks with different modes and topological frequency bands have differ-ent functional characteristics and organizational modes,which can more effectively un-derstand the physiological mechanism of brain network cross-frequency and cross-modal interaction,and provide a basis for brain aging.Research offers a new approach.