Early Diagnosis Model Of Multimodal AD Based On MRI And SNP

Alzheimer’s Disease（AD）^[1]is a brain disease,a chronic neurodegenerative disease of the central nervous system.After several years of brain changes,obvious symptoms such as memory will appear.Fading and language problems,patients will not be able to take care of themselves in life,need other people’s care around the clock,causing a heavy burden on patients and their families,mild cognitive impairment（MCI）^[2],is a kind of between AD and healthy elderly group（Healthy Controls,HC）during this period,the individual has subtle cognitive changes,but does not interfere with daily activities.Therefore,treatments that prevent mild cognitive impairment from deteriorating into Alzheimer’s disease will have a significant impact on the quality of life,the burden of nursing staff,and the cost of use and care.It is current research to identify which MCI patients are more likely to eventually develop into AD patients.The main goal,with the development of deep learning and machine learning,fusion of information from multiple modal data will enhance the recognition rate of early AD diagnosis.This article conducts research from the perspective of multimodal data,by studying the two modal data of magnetic resonance imaging and single nucleotide polymorphism,and studying how to combine to improve the early diagnosis of AD.The main research contents are as follows:（1）Convolutional neural networks can improve the automatic learning and representation capabilities of the network by deepening the network structure.However,as the network depth increases,a network degradation problem will occur.In response to this problem,this paper proposes a residual neural network 2DResnet,the AD classification model for network and ensemble learning,uses multiple data expansion methods to increase the number of slices,and then trains multiple 2DResnet models for each slice and finally votes for integration.The experiment has achieved good results;（2）SNP data has high dimensionality and direct classification.The model is easy to overfit.Nowadays,SNP data of disease-related genes are generally selected manually when SNP data is used for disease diagnosis.However,the manually selected SNP data may be in the selection process.There are omissions.In addition,many disease-related SNPs have not been recorded.This article uses genome-wide association analysis to preprocess the SNP locus data,encode it into a binary SNP data set,and then use machine learning algorithms for classification Research and experiments have found out the data of AD-related SNP loci,which will provide help for the research of AD in genetics;（3）The existing AD classification models mostly use single-modal data for research.The single-modality considers the problem from a single angle,which affects the recognition result,and the classification accuracy of the obtained model is not high.For this reason,this article uses multi-modal data From the perspective of research,a multi-modal AD diagnosis model based on the weighted probabilistic classifier is proposed.First,expand the previous MRI model from single-axis slicing to three-axis slicing,use the validation set data to select the TOP3 base classifier in each axis direction for voting integration,generate three MRI classifiers,and then The ensemble learning model based on the decision tree is used to construct the SNP classifier,and the ensemble learning method based on the weighted probabilistic classifier proposed in this paper is used to combine the MRI and SNP classification models.The experimental results are improved compared with the results of any single mode,which shows the effectiveness of the multi-modal method.