Early Prediction And Evolution Model Of Alzheimer’s Disease Based On Methylation And Imaging

Alzheimer’s disease is a degenerative disease of the central nervous system.In the early stage,patients tend to miss the golden treatment time due to symptoms similar to aging,leading to aggravation.Therefore,it is a major issue that how to use medical data to assist doctors in understanding the condition and needs to be resolved.The latest research found that methylation has the advantages of early disease,easy detection,and minimally invasive sampling.Therefore,this project intends to study the problem of early diagnosis and prediction of disease evolution in the intermediate period through multi-omics data containing methylation.The main research contents of this article include:1.Aiming at the high-dimensional problem of methylation,a multiregular feature selection algorithm combining density clustering is proposed.Specifically,based on the medical priori that the abnormal regions are clustered,it is proposed to cluster features in advance under the premise of controlling the clustering scale.On this basis,based on the sparse regression model,a variety of regular terms are combined on the clustering and feature scales to ensure the clustering and discrimination of the selected features.Experiments prove that the selected features have excellent classification capabilities.2.Multi-model experiments were carried out for the fusion of methylation data.Specifically,multi-angle analysis is performed based on multiple machine learning algorithms and feature fusion methods.Experiments show that fusion of methylation features can improve the performance of each model.3.Aiming at the problem of integrating multi-omics data such as magnetic resonance imaging,clinical genetic and so on,a complex relationship mining model based on graph convolutional neural network is proposed.Specifically,according to the medical priori of patients with similar conditions that have similar characteristics,a method for constructing an adjacency matrix based on sample similarity is proposed.On this basis,in order to further explore the complex relationship between samples,Chebyshev convolution is introduced.Finally,aiming at the problem of fusion priors,an adaptive adjacency matrix update method that strengthens the prior background is proposed.Experiments have proved that the model can be adjusted automatically and flexibly,and the prediction results are more accurate than a variety of machine learning algorithms.In summary,the main research content of this topic is to design a methylation feature mining algorithm based on spatial information and a model based on multi-omics to achieve early diagnosis and prediction of disease evolution in the mild cognitive impairment period.First,through the fusion of density clustering multi-regular feature selection algorithm,the methylation features are effectively screened from a medical perspective.Then,the validity of the selected features is verified by a variety of classic algorithms.Finally,the model is proposed for in-depth mining of disease characteristics and evolution rules.Experiments prove that the model has application potential and can assist doctors in timely intervention and reduce the burden on patients’ families.