Research On The Early Diagnosis Of Alzheimer's Disease Based On Deep Learning

Alzheimer's disease(AD)is a primary neurological disease with memory and cognitive impairment as the main symptoms.It is the most common type of senile dementia and seriously threatens the quality of life and safely of the elderly.China is facing an increasingly severe population aging problem,and the increasing prevalence of AD will bring heavy economic and living burden to patients,family and society.Due to the pathological characters and technical limitations of AD itself as well as the limitation of medical technology,it is difficult to achieve the early diagnosis of AD.Besides,there is no very effective treatment method in the global medical community at the present stage,so we can only rely on drugs to control the deterioration of AD as far as possible.Therefore,the study of early diagnosis of AD has been one of the hot and difficult issues at home and abroad.This paper aims to build and optimize the applicable early classification diagnosis model of AD by studying applicable machine learning methods,and help clinicians realize the early diagnosis of AD,which has extremely important theoretical and practical application value.Firstly,aiming at the problem of limited data sample and high dimension of AD,this paper establishes a novel switching delayed particle swarm optimization support vector machine(SDPSO-SVM)model to realize the early diagnosis of AD.Then,in order to further improve the accuracy of early diagnosis,considering the characteristics and difficulties of complex relationship and representational capacity of AD data samples,the applicable deep learning model is researched and improved to realize the early diagnosis of AD.Finally,the comparison of results proves that the improved deep learning method proposed in this paper can provide a reliable diagnostic method for early diagnosis and timely treatment of AD.The main contents of this paper are summarized as follows:(1)In view of the complexity and specificity of magnetic resonance imaging(MRM)of AD,this paper proposes an image preprocessing method which is suitable for the early diagnosis of AD.In this paper,five steps including skull stripping,non-uniformity correction,registration,tissue segmentation and spatial standardization,smoothing are used for image preprocessing,and then 90 regions of interest(ROIs)related to AD lesions are preliminarily extracted based on the template.In order to eliminate irrelevant or redundant features in ROIs,the principal component analysis(PC A)is firstly utilized to proj ect the original 90-dimensional data features into a new data feature space,where the new 38-dimensional data features containing the original information are obtained.And then,the multi-tasking feature selection(MTFS)method is adopted to select 12-dimensional feature subset from the 38-dimensional data features which are relevant to all tasks with strong representational force,so as to further cut dow^n the dimension of data feature.(2)In view of the limitation that non-linear of data sample and high dimensional feature in AD data,this paper proposes an optimized S VM to realize the early diagnosis of AD.Considering that the overall performance of SVM is largely dependent on the value of penalty coefficients and kernel parameters,thus,SDPSO algorithm is applied to optimize parameters of SVM in this paper.Among them,SDPSO algorithm considers the individual and population delayed information of particles in the searching process,adaptively adjusts the algorithm parameters according to the information of evolution factor and the non-homogeneous markov chain,promotes the mutual communication among particles,and thus effectively avoids particles trapping into local optimal value Introducing SDPSO algorithm into SVM is conducive to search global optimal value of model parameters,which provides suitable and reliable parameters for building and optimization of the early diagnosis model of AD based on SVM,and the experimental results display that this method can obtain satisfactory classification effect(3)In order to further improve the accuracy of the early diagnosis model of AD,considering the characteristics and difficulties of complex relationship and representational capacity of AD data samples,this paper proposes an improved deep learning method to realize the early diagnosis of AD.Among them,considering that deep learning network is difficult to detect the internal correlation information among multiple tasks in terms of learning features,this paper introduces the multi-task learning method to learn the similarities and differences among multiple tasks,capture potential information,and strengthen the representational ability of features.Meanwhile,problems such as parameters optimization and data overfitting exist in the training process of deep learning network,dropout technology and zero-masking technology are introduced in this paper to improve the structural mechanism and learning performance of network.Finally,the improved deep learning method is adopted to the early diagnosis of AD,and compared with other deep learning methods and the SDPSO-SVM method,which can be seen that the proposed method can acquire higher accuracy,and has a good theoretical research and application value.