Human Brain Functional Network Construction Based On Interpretable Machine Learning Methods

The human brain is the human nerve center and the most complex organ of the human body.Researchers have found that many neurological diseases are related to the abnormal topology of brain structure and brain function networks.Constructing the human brain functional network can not only be used to study nervous system diseases and provide a new perspective for understanding the pathological mechanism of nervous system diseases,but also be used to explore the relationship between human brain aging and cognitive function.With the continuous development of neuroimaging,a variety of technical methods for obtaining human brain structure data and human brain nerve activity signals have emerged,which has greatly promoted the development and progress of brain science.The method of constructing the human brain function network can be based on the network construction between human brain regions on a macro scale,or can be based on the network construction on the voxel level.Aiming at the problem of constructing the human brain functional network,this paper mainly uses magnetic resonance imaging(MRI)and functional magnetic resonance imaging(f MRI)to obtain information related to the human brain,and focuses on the construction method of functional network based on voxel level.Use explainable machine learning and methods to construct a human brain functional network,and on this basis,use a complex network method to analyze the constructed functional network.The main work of this paper includes the following two aspects:1.Based on SHAP(SHapley Additive ex Planations),a method for constructing human brain functional networks is proposed,and the interpretability of SHAP is used to visually explain the individual and global aspects.For image-stimulated f MRI data,first implement the classification task based on the XGBoost classifier;then combine the SHAP and the training model to obtain the interactive information between features,and use these interactive information to achieve the construction of a human brain functional network at the voxel level;and finally Use the interpretability of SHAP to analyze and interpret the results.The results show that for different classification tasks,the response area of the human brain is different,and the functional network connections of the human brain under different tasks are also different.2.A functional network construction method based on second-order polynomial kernel functions is proposed for the MRI data of 403 subjects.First,in the framework of the generalized additive model,the interaction between features is considered,and the polynomial kernel function support vector regression(SVR)is combined to achieve age prediction;then the functional network is constructed by using the interaction between model weights and features.The experimental results show that with the increase of age,the connections between voxels have a certain corresponding relationship with age,and the connectivity between different functional network regions is enhanced.In addition,through the analysis of the topological structure of the two functional networks constructed,it is found that the functional network corresponding to visual stimuli satisfies the scale-free characteristics of complex networks,while the age-related functional network had the small-world properties,and as the age increases,the average degree of the network will also increase,the scale-free characteristics of the network gradually disappears,and the modularity of the network is declining.