Study On The Early Diagnosis Of Alzheimer's Disease Based On CNN And GWAS

Alzheimer's Disease(AD)is one of chronic neuro degenerative disease.At present,the number of worldwide AD patients is increasing and growing rapidly.The clinical manifestations of AD patients are amnesia,aphasia,cognitive dysfunction,etc..AD patients suffer severe decline in quality of life and need care,that bring great pain and burden to patients and their families.Currently,the cause of AD is unknown,and the course of the disease is irreversible.There is no way to truly cure AD yet.Early diagnosis of AD and intervention treatment may delay or prevent the disease from continuing to deteriorate,so early diagnosis of AD is of great significance.Mild Cognitive Impairment(MCI)is a state between AD and Healthy Controls(HC).Studies have shown that patients at MCI state are more likely to convert to AD than those at HC state.Therefore,accurate screening of MCI patients has become one of the research hotspots of early diagnosis of AD.Magnetic Resonance Imaging(MRI)is an important tool for AD research.Studies have shown that the brain structure of AD patients has obvious pathological signs and biomarkers,and clear brain structure imaging can be obtained by MRI for early diagnosis of AD.With the deepening of AD research,some studies have shown that AD is greatly affected by genetic factors,so the study of AD genetic biomarkers will help to diagnose patients with susceptible AD earlier.At present,the use of machine learning(ML)or Deep Learning(DL)methods to analyze MRI images has become a popular way for AD research,but these research methods have some shortcomings.Usually,using machine learning methods to analyze MRI images requires select regions of interest(ROI)manually and then compute features for training the classification model,which increases the labor cost of the experiment and human factors increase the uncertainty of the experiment.Besides,although deep learning methods can automatically extracte features from MRI by the adaptive training,the feature interpretability is poor,and it is not convenient to find the neuroimaging biomarkers of AD.The chapter 3 of this paper combines the current research hotspots of AD and uses deep learning methods to analyzes MRI images.We analyzed the advantages and disadvantages of traditional Machine Learning(ML)methods,Convolutional Neural Networks(CNN)based methods,and 3D convolutional neural networks based methods.Finally,we proposed our AD diagnostic model which combines convolutional neural networks and ensemble learning.The model showed higher accuracy and stability than the comparative paper in the classification of AD,MCI and HC.In addition,the model's special model structure makes it easy to find imaging markers for AD.In chapter 4,We used brain atlas to describe the AD constructive neuroimaging biomarkers identified in this paper.The neuroimaging biomarkers we have identified include well known AD-related brain regions such as the hippocampus,amygdala,and temporal lobe,as well as new brain regions.This indicates that the constructive neuroimaging biomarkers we have found are in line with current AD research experience and have high reference value.In addition,we further performed behavioral domain analysis on the AD neuroimaging markers we found.The experimental results show that the behavioral domain distribution of these biomarkers is consistent with the clinical features of AD patients,which further prove the correctness of the structural neuroimaging biomarkers of AD we have found.In chapter 5,based on our AD early diagnosis model and AD neuroimaging biomarkers presented in this paper,we combined genetic data to explore the genetic biomarkers of AD.We try to apply Genome-Wide Association Study(GWAS)method to genetic data of the subjects and the volume data of the brain regions and aim to find out the genetic variation that related to the volume of AD structural neuroimaging biomarkers.These genetic variations can be used as genetic biomarkers of AD and are important for the early diagnosis of AD.