Early Diagnosis Of Ad Based On 3D Convolutional Neural Network And Brain Network Analysis

Alzheimer’s Disease(AD),commonly known as Dementia,is an irreversible degenerative disease of nervous system.In recent years,the number of AD patients worldwide has risen sharply and the cost of treatment is huge,which seriously affects the normal life of patients and families.It has become an urgent social issue.At present,the most effective way to curb AD is to carry out early intervention,which can slow down and control the development of the course of AD.Mild Cognitive Impairment(MCI)is generally considered to be a transition state from Normal Control(NC)to AD,especially late MCI is likely to develop into AD.Therefore,how to diagnose AD / MCI quickly and accurately clinically at an early stage is still a challenging problem.Magnetic Resonance Imaging(MRI)has gradually become an important medical tool for studying brain diseases because of its non-invasiveness to the brain and its sensitivity to changes in brain structure.The use of MRI technology can help capture regional brain atrophy and understand anatomical changes associated with AD,which can reflect changes in disease stages to a certain extent.In recent years,with the rapid development of Machine Learning(ML),in particular,deep Convolutional Neural Network(CNN)are widely used in various medical image segmentation,classification and registration tasks.Therefore,more and more researchers are trying to analyze MRI using ML methods to achieve early diagnosis of AD.However,the early diagnosis method of AD based on traditional ML requires manual extraction of features,which is easily interfered by human subjective factors.At the same time,the AD diagnosis model based on 2D Convolutional Neural Network have insufficient use of MRI spatial information,resulting in a low diagnosis rate and lack Issues such as the effective determination of AD biomarkers.In view of the above problems,this paper analyzed MRI based on 3D Convolutional Neural Network and Brain Network Analysis methods,and then realized the diagnosis of AD and the determination of AD biomarkers.The main contents of this paper are as follows:1)In view of the problem that the insufficient use of image spatial information leads to the low diagnosis rate of AD,this paper proposed an early diagnosis model of AD based on a combination of Residual and Attention Mechanism(AM)3D Convolutional Neural Network.Experimental comparison results show that the model has higher accuracy and stability in AD,MCI,and NC classification.2)Based on the prior knowledge of atrophy of local related brain regions such as AD,in order to improve the diagnosis rate of AD,a method of early diagnosis of AD based on Regions of Interest(ROI)3D Convolutional Neural Network was proposed.Compared with other methods,this method can further improve the accuracy of AD,MCI,and NC classification,and confirms the effectiveness of ROI selection.3)In order to further improve the diagnosis rate of AD,and also to search for AD biomarkers,an AD Ensemble Learning(EL)model based on the combination of ROI-based CNN and Genetic Algorithm(GA)was proposed.It used GA to find the brain regions of AD biomarkers with significant classification effect,and used the brain map to analyze the behavioral domains of these brain regions.Experimental results show that the model not only further improves the classification effect,but also finds the AD biomarkers basically in line with current AD research results and experience,and has certain reference value and significance.4)In addition,we studied the Brain Structure Network of different people based on the gray matter volume of the brain,and combined with the Brain Network theory to analyze the Network Properties of the related people,and then realized the determination of AD biomarkers.The experimental results show that the structural networks of different populations have the properties of small worlds,and the connections between the Hubs of their brain regions are also very different,and their corresponding brain functions also match the performance of the patient population to a certain extent,which may be AD pathological research provides structural help.