Study On Classification Of Alzheimer’s Disease Based On Structural Magnetic Resonance Imaging

Alzheimer’s disease is a common neurodegenerative disease,which occurs frequently in the elderly over 65 years old.With the rapid development of social economy and medical level,the global population is showing an aging trend,and the threat of Alzheimer’s disease to human health is becoming more and more obvious.In order to find an effective diagnosis and treatment method as soon as possible,many related researchers have worked hard in different fields.Among them,making a diagnosis at the early stage of Alzheimer’s disease and then intervening treatment is one of the important research areas of current research.Structural magnetic resonance imaging is an important basis for clinicians to diagnose patients with Alzheimer’s disease.With the continuous development and wide application of machine learning and deep learning in image segmentation,image recognition and image classification,many researchers at home and abroad have been committed to using deep learning algorithm to classify structural magnetic resonance images of Alzheimer’s disease.Combining classification algorithms can help clinicians diagnose Alzheimer’s disease faster and more accurately,so as to intervene and treat patients with mild cognitive impairment or Alzheimer’s disease as soon as possible.This study hopes to improve the classification accuracy of Alzheimer’s disease based on structural magnetic resonance imaging on the basis of existing research,and help clinicians identify and diagnose Alzheimer’s disease as soon as possible.The main research content is based on the characteristics of one-dimensional brain network attributes,two-dimensional gray matter voxel images,and three-dimensional gray matter images,combined with machine learning and deep learning algorithms,three groups,Alzheimer’s disease,mild cognitive impairment,and normal cognition were subjected to two-by-two classification and three-category classification.The data in this study come from the ADNI database,including 276 cognitively normal subjects,240 mild cognitive impairment and 276 Alzheimer’s patients.The methods used mainly include building an individualized brain network based on the DTW algorithm,using four algorithms of KNN,SVM,DT,and RF classify Alzheimer’s disease into two categories and three categories based on brain network properties;convert three-dimensional structural magnetic resonance images into two-dimensional gray matter voxel images and use 2D CNN and transfer learning(VGG19,Inception V3,ResNet50)algorithm is based on gray matter voxel images to classify Alzheimer’s disease in two and three categories;3D CNN and 3D Res Net-18 are used to classify Alzheimer’s disease in three categories based on gray matter images.Among the two classification results,the accuracy of 2D CNN classification based on two-dimensional gray matter voxel images is the highest.The average classification accuracy of the AD vs.CN,MCI vs.CN,and AD vs.MCI through ten-fold cross-validation is 99.1%,96.7%,93.8%.Among the 3-way classification results,the random forest algorithm based on one-dimensional brain network attributes has the highest accuracy rate,and the average accuracy rate of the validation set of ten-fold cross-validation is 94.5%.To improve the classification accuracy,on the one hand,the DTW algorithm is applied to the construction of individualized brain network,and then the GRETNA software is used to analyze the brain network attributes for machine learning classification,which achieves 94.5% of the three classification accuracy.On the other hand,extracting the whole brain gray matter voxel value and converting the sMRI to 234*234 gray matter voxel image not only preserves the whole brain feature but also reduces the size of the dimension and single sample,greatly reduces the running time and calculation cost,and achieves 93.7% three classification accuracy.Both methods have achieved good results and can provide new ideas for early classification of Alzheimer’s disease.