Research And Application Of Brain Network-based Tau Prediction Model For Alzheimer's Disease

Alzheimer's disease(AD)is now common and highly prevalent in the middle-aged and elderly population and has severely affected the normal lives of patients and their families.However,current therapeutic drugs can only slow down AD progression as much as possible but cannot cure AD patients at the root.Therefore,it is especially important to study the disease progression of AD by using computer technology to help physicians achieve early intervention in AD patients with potential AD lesions.Tau protein is one of the major causative agents of AD.Identifying Tau protein content in the brain can help distinguish between cognitively normal elderly,mild cognitive impairment,and AD.It can also be useful in terms of disease mechanisms and progression patterns.However,current methods of obtaining Tau protein levels are mainly through quantitative PET neuroimaging or extraction from cerebrospinal fluid,which is invasive and costly.In contrast,Rs-f MRI neuroimaging technology is cost-effective,non-invasive,high-resolution,and has received widespread attention.Given the above,combined with the current status of clinical research on functional brain network and Tau protein deposition,this paper proposes a research method to construct a functional brain network to predict Tau protein content based on Rs-f MRI neuroimaging,which includes:(1)To overcome the problem that it is more difficult to obtain the cerebrospinal fluid Tau protein content in AD,this paper proposes a Tau protein content prediction method based on the MLP-ATT model of brain functional connectivity.Firstly,the Rs-f MRI neuroimages were preprocessed,then the functional connectivity matrix was constructed by Pearson correlation coefficient,and the relationship between brain regions and Tau protein deposition was investigated by the Lasso algorithm.The results showed that the functional connectivity between the right lingual gyrus(Lingual?R)and the right superior temporal gyrus(Temporal?Sup?R)brain regions had a greater influence on Tau protein deposition.Finally,the regression prediction of cerebrospinal fluid Tau protein content by the MLP-ATT model had a root mean square error of 104.2 and a mean absolute error of 77.2,which was better than the traditional machine learning algorithm.(2)To resolve the problems of the high cost and invasive nature of Tau-PET for the quantitative analysis of Tau protein content,this paper proposes a method to predict Tau protein content using complex network topological properties based on the study of functional brain connectivity and regression prediction of Tan protein content using Sparse Light GBM(SparseLGBM),whose root mean square error is 59.2.The mean absolute error was 49.6,and the experimental results proved the validity of the model.(3)In response to the problems of Rs-f MRI data preprocessing,the more cumbersome operation of constructing functional brain networks and the application of prediction models,this paper designs a Tau protein content prediction aid diagnostic system,which has medical image conversion,functional brain connection construction,and Tau protein content prediction modules to facilitate doctors to use Rs-f MRI neuroimaging to study the Tau protein content of the subject's brain.