Early Recognition Of Alzheimer’s Disease Based On Deep Neural Networks And SMRI

Alzheimer’s disease(AD)is a degenerative and irreversible neurological disease with a high mortality rate and a long course of disease,which has brought great psychological and economic burdens to families and society.According to estimates by Alzheimer’s Disease International,75%of dementia patients worldwide are not diagnosed through clinical diagnosis,which leads to the fact that when the disease is discovered,it is already in an irreversible stage of development,and patients at this stage will Significant memory loss and neurological decline appear.However,as far as the existing domestic medical conditions are concerned,the number of hospitals with a full set of early identification is limited,and it is difficult to meet the huge domestic demand for early AD screening.For patients,the early screening of AD using positron emission tomography PET imaging technology is expensive,which brings a heavy burden to patients.This prompted us to develop an adaptable and inexpensive computer-aided diagnosis technique for early screening to obtain early diagnostic markers.Aiming at the above difficulties and challenges,this paper proposes two methods of Alzheimer’s disease identification based on deep frequency domain network and Alzheimer’s disease identification based on metric learning.Specifically,the main work of this paper is as follows:1.Aiming at the problem that the feature dimension of 3D MRI image is too high,which leads to too long training time,and the low contrast of medical gray image itself is not conducive to the feature extraction of deep neural network,it is proposed to use discrete cosine transform to convert the image to the frequency domain The method,under the premise of not affecting the test accuracy,extracts effective information through the control of the communication signal bandwidth to shorten the training time.Compared with the traditional deep neural classification network,the model significantly improves the accuracy of Alzheimer’s disease based on MRI image diagnosis and shortens the training time.2.Aiming at the problem of large intra-class differences and relatively small inter-class differences caused by the lack of long-distance dependencies between blocks of 3D MRI images,and the problem that medical images themselves lack effective semantic information,a method of using metric learning is proposed.Reduce the intra-class distance while increasing the inter-class distance.At the same time,Transformer technology is introduced to capture the long-distance dependencies between blocks to improve the effective semantic information of medical images,making the model significantly improve the efficiency of Alzheimer’s disease MRI recognition.Compared with traditional convolutional neural networks,the model significantly improves the accuracy of MRI diagnosis of Alzheimer’s disease.3.According to the proposed two Alzheimer’s disease identification methods,a predictive evaluation system for this task is designed and implemented.The system is designed to provide better identification services for the auxiliary diagnosis of grassroots doctors and the early identification of the elderly population.The system design follows the principle of ease of operation,so that most users can quickly grasp and use it.The system could use 3D MRI brain images as input to the system,and finally output the user’s predictive assessment of Alzheimer’s disease.In order to ensure the usability and robustness of the system,tests have been carried out on some public data sets.