Research On Prediction Of Brain Networks In Alzheimer's Disease Based On Multi-Modal Imaging

In the course of research on a range of brain disorders,it was found that the concept of "brain network" can be obtained by dividing the brain into several brain regions and defining them as network nodes,and defining certain connections between brain regions as edges.With the in-depth study of brain images,many different brain networks have emerged.Then,we use the knowledge of graph theory to interpret them,and further explore the structure or functional connection of the brain and the relationship between them.Diffusion tensor imaging(DTI)can reflect the direction of white matter fibers in the brain by detecting the diffusion of water molecules in white matter fibers.Functional magnetic resonance imaging(f MRI)can capture the functional activity signals of the brain by measuring the blood oxygen level dependence of the brain.In this paper,DTI and f MRI are taken as the research objects.By constructing brain structure connection network and brain function connection network respectively for these two images,we study the prediction of brain function correlation by brain structure correlation.The main research content of this paper is as follows:1)In response to issues such as how to dynamically collect richer node interaction information and how to more effectively perform small sample learning in brain network fusion research,We propose a Random Walk-Grassmann model to integrate the structural connections and functional interactions of the brain.First,we derived a matrix of structural connectivity and temporal features of the brain from multi-modal data for each subject.Then,both matrices were fused using a random walk algorithm and the Grassmann pooling method.Finally,we performed feature selection by recursive feature elimination method and put the selected features into a support vector machine to obtain the final classification result.This method performed four dichotomous classification experiments on the ADNI dataset,and the classification accuracy all outperformed traditional brain network classification methods.2)Although multi-modal data can help to improve the diagnostic effect of AD,it will bring greater economic burden to patients.Therefore,in view of the demand for economical early screening of AD,we studied the prediction of the topological characteristics of functional connections based on the brain structural connections of AD patients.Firstly,the brain network was constructed by DTI and f MRI images of subjects,and then the network topology attributes were extracted.Xgboost regression,support vector regression,ridge regression and random forest regression are used as prediction models to select the optimal model through experiments.It has been experimentally demonstrated that the global clustering coefficient,the shortest path length,the global efficiency and the node efficiency of the brain structure network can be used to predict the global clustering coefficient and the shortest path length of the functional network respectively.3)An assisted diagnosis system for Alzheimer's disease based on multi-modal brain image data was designed and implemented.It contains five functional modules: login validation,data preprocessing,intelligent diagnosis,Intelligent prediction and record query.When using intelligent diagnosis,users need to upload three complete DICOM format data files of DTI,T1-MRI,and f MRI taken simultaneously to make a preliminary understanding of their condition through this system.When using intelligent prediction,users only need to upload DTI and T1-MRI files to predict the topological characteristics of functional brain network through the system.At the same time,the system retains the user's diagnostic or predicting results many times and displays them in the personal Center for easy viewing.