Research On Intelligent Detection And Classification Of Alzheimer's Disease Based On Deep Learning

Alzheimer's disease(AD),commonly known as Alzheimer's disease,is mainly an irreversible neurodegenerative disease.As the disease progresses,patients often experience cognitive decline,memory function decline,and severe loss of vitality.At present,the pathogenesis of the disease is still unclear,so it cannot be cured by current medical methods.Early detection and early intervention is the best way to deal with the disease.However,because the onset of the disease is hidden and the development of its condition is often slow,a large number of nerves have died when the patient's symptoms are clearly diagnosed,which has caused many difficulties in the prevention and treatment of the disease.In recent years,with the development of science and technology,people can make diagnosis by combining other factors such as laboratory and imaging examinations,which has improved the accuracy of Alzheimer's diagnosis to a certain extent.However,because this is a slow and progressive disease,the changes in medical images obtained through laboratory and imaging examinations in the early stage of the disease are very small,difficult to observe with the naked eye,and have its irreparable limitations.At the same time,the accuracy of diagnosis is often related to the doctor's experience,so the phenomenon of missed diagnosis and misdiagnosis often occurs,which makes it difficult to accurately diagnose Alzheimer's disease.Facing the increasing aging population worldwide,there is an urgent need for a method that can accurately diagnose Alzheimer's disease,assisting brain doctors to make faster and more accurate judgments on the patient's condition,so that timely and effective treatments can be taken.This subject is mainly based on 920 healthy T1 weighted brain MRI medical image data sets of 175 healthy people and 287 clinical Alzheimer's patients.Based on the deep learning method,a very accurate neural network model is trained for Alzheimer's disease.Detection and classification of Haimer disease.During the research process,the Alzheimer's disease medical image data set wascreated.The original data set contains a total of about 920 brain MRI medical images.At the same time,the image category is divided according to the severity of the patient's condition,which is divided into healthy and light.Four grades:degree,moderate and severe.According to the project objectives and the needs of neural network model training,effective data preprocessing and deep processing are performed on the data set.Preprocessing includes data screening,data naming,size normalization,etc.The purpose is to adapt the data image to the structure and training requirements of the neural network model.Deep processing includes data enhancement,feature enhancement,etc.,mainly through the optimization of the data set to improve the accuracy of neural network model training.According to the requirements of the subject,we divided the entire experimental process into two stages and trained two different neural network models.In the first stage,the original data set is divided into two types,health and disease,and labeled with 0 and 1.The target training is a two-class neural network model that can detect human brain MRI medical images for health and disease.The second stage is to improve the classification accuracy.The data set is divided into four types:health,mild,moderate,and severe,which are labeled with 0,1,2,and 3,respectively.The target training one can be used to perform health on human brain MRI medical images.A four-class neural network model for mild,moderate,and severe disease detection.In the design of the neural network model,a lightweight neural network model is selected for training based on the salient features of the small sample data set in this paper.SqueezeNet can achieve the same or even super-level training effect with less than one-fifth of the parameter amount of AlexNet,and has achieved remarkable results in many international image recognition competitions.However,due to the excessive compression parameters in the SqueezeNet network,the accuracy is not ideal,and the amplitude is too large.We introduced a series of improvements by introducing residual ideas in the above network to increase the layer jump structure,expand the network width,and optimize the activation function.Method A new ultra-lightweight neural network model NewNet is proposed.The improved lightweight neural network model improves the accuracy and efficiency of imageclassification under the premise of slightly increasing the parameters.The experiment proves that NewNet's classification accuracy of the test set has reached 93.47%in the binary classification experiment.In the four-category experiment,the recognition accuracy of the test set can reach 87.85%,which is not only higher than SqueezeNet,but also higher than the mainstream neural network models such as LeNet,Alexnet,GoogleNet,VGG16,ResNet,etc.The experimental goal is initially achieved.The basic experimental environment of this subject is Intel Core i5 4570(4 X 3.2 GHz processor),16 GB memory,GPU is GTX1080Ti graphics card and TensorFlow2.0 learning framework based on Ubuntu 16.04 operating system,and the integrated development environment is Spyder3.3.3,Use Python3.6 language as the programming language.At present,artificial intelligence has been widely used in all walks of life,and the application in medicine is more and more.It is believed that the intelligent detection and classification technology of Alzheimer's disease based on deep learning will definitely help Alzheimer's medicine Play an important role in diagnosis.