Application Of Graph Network Model In Prediction Of Human Brain Structural-Functional Connectivity Relationship

In recent years,deep learning has achieved success in many fields,including image processing,natural language processing and other fields.However,how to extend deep learning to graphs has still not been completely solved.Graphs,also known as networks,have rich definitions and strong expressive abilities,and are of great significance to the development of deep learning in non-traditional fields.We call the model defined on the graph as the graph network model,and call this field graph neural network,or graph network for short,to distinguish it from convolutional neural networks.For now,graph neural networks have been transplanted and expanded on graphs using many concepts of convolutional neural networks,and have achieved excellent results on many tasks.However,most of the current researches are mainly focused on the study of the properties of nodes and edges.There are relatively few studies on the overall properties of graphs,and the research on the generation of weighted graphs is even rarer.Based on the above situation,this paper uses the published Human Connectome Project data set,combined with the graph convolution operation in the graph network model,to improve the relevant models in the medical field.On the basis of maintaining its good theoretical interpretation,through graph convolution operations and other adjustments,the model's ability to predict the target has been effectively improved.The research work of this paper is mainly divided into the following three parts.First,we introduce the research background in the field of graph neural network,and clarify the current mainstream model and its principle mechanism in the field of graph neural network.On this basis,we conducted in-depth research on the graph convolution network,analyzed the theoretical basis and model evolution using relevant literature,and finally obtained the conceptual principles and theoretical advantages of graph convolution operations.Subsequently,we introduced the relevant knowledge and basic concepts of human brain network research.Based on the relevant subject background,we introduce and analyze the relationship between the structural connectivity and functional connectivity of the human brain network,and on the basis of the existing model,combined with graph convolution operation to improve it,to obtain the improved model in this paper.After this,we introduce the current mainstream graph generation network model,and select the appropriate model for subsequent comparison according to the data requirements.Finally,we check the results of our model through comparative experiments.The comparison experiment is mainly divided into two parts.The first part focuses on the prediction performance of the model and compares the prediction performance of different models on the data set.The second part mainly focuses on the stability of the model and compares whether the model can maintain good performance under different interference conditions.Through comparison experiments,our model maintains good results in both parts.