Research On Alzheimer’s Disease Diagnosis Algorithm Based On Deep Learning

In recent years,with the rapid development of artificial intelligence,deep learning has been widely used in various fields.In the medical field,especially in medical image processing,deep learning is applied to medical image segmentation,image registration,image reconstruction and other directions.In lung disease,heart disease,skin disease,breast disease,eye disease,etc.,deep learning has achieved outstanding research results.However,for the diagnosis of Alzheimer’s disease,we still need to use deep learning to explore further.Alzheimer’s Disease(AD)is a typical senile dementia,which is progressive and irreversible.Due to the complex pathogenesis of Alzheimer’s disease,the accuracy of clinical detection of Alzheimer’s disease is very low.At the same time,because the brain changes caused by Alzheimer’s disease are not obvious,it is difficult to detect Alzheimer’s disease in the early stage of the disease.When patients are diagnosed,they are mostly in the middle and late stages of the disease,causing patients to miss the best time for treatment.Therefore,in order to solve the problem of low diagnostic accuracy and difficulty in early diagnosis of Alzheimer’s disease,this paper uses deep learning methods to diagnose Alzheimer’s disease,allowing patients to detect the condition as early as possible and provide timely and correct treatment.At the same time,the deep learning-assisted diagnosis method can also provide doctors with reliable and consistent reference.Magnetic resonance imaging(MRI),as a commonly used means of brain clinical examination,plays an important role in the diagnosis and treatment of Alzheimer’s disease.It is very important to study the MRI diagnostic algorithm for Alzheimer’s disease.practical significance.Therefore,this paper carried out the following studies,the main contents of which include:1.A neural network framework based on the attention mechanism is proposed for the brain changes of patients with Alzheimer’s disease,which mainly focus on brain changes such as the expansion of the ventricles and the atrophy of the hippocampus.By introducing the attention mechanism into the Convolutional Neural Network(CNN),the local changes in brain MRI are captured.In addition,the introduction of probabilistic fusion method and instance-batch normalization method also improves the diagnostic ability.The entire neural network framework has achieved a more accurate diagnosis of Alzheimer’s disease,and achieved an accuracy rate of 84.17% on the ADNI dataset.2.Aiming at the fact that the brain MRI in adjacent stages of the disease is very similar,it is difficult to extract the features in 3D MRI,and the previous methods are more complicated.A feature extraction method based on 3D MRI is proposed.By using 3D asymmetric convolution to directly extract 3D features,it helps 3D CNN to extract more discriminative features in 3D space while avoiding the loss of feature information.And,aiming at the problem that the overall structure of MRI at each stage is similar,but there are local differences in MRI,a feature fusion method based on multi-scale channel attention is proposed.This method can simultaneously focus on changes in the entire MRI and local regions of the MRI.Therefore,the convergence of these two methods in the same neural network greatly improves feature extraction and feature fusion,improves the early diagnosis rate of Alzheimer’s disease,and achieves an overall accuracy rate of 88.33%.3.Aiming at the lack of transparency in the process of neural network diagnosis of Alzheimer’s disease,which cannot directly provide reasoning and explanation like experts,a convolution visualization method based on Class Activation Mapping(CAM)is applied.By opening the "black box" of deep learning and providing visual explanations,the outside world can better understand the working mechanism and diagnostic basis of 3D CNN.