Research Of Auxiliary Diagnosis Of Alzheimer’s Disease Based On Residual Neural Network

Alzheimer’s Disease International said in its global report on Alzheimer’s disease that there is a patient with Alzheimer’s disease(AD)every three seconds.The stages of AD can be divided into early,middle and late stages.In the early stage,the main manifestations are memory impairment and memory decline,in the middle stage,aphasia,agnosia and personality change,and in the late stage,bed rest and inability to take care of themselves.In particular,advanced AD patients bring huge troubles to the lives of their families and great pressure to the society.AD has become one of the major medical and health problems in the world.From the perspective of deep learning,this study aims to improve the ability to assist diagnosis of Alzheimer’s disease,and strive to alleviate the above problems.The research mainly includes two parts.One is from the perspective of algorithm model,two kinds of improved residual neural networks(1)and(2)are used respectively to carry out three classification tasks for AD;the other is from the perspective of engineering(3)to solve the problems existing in the clinical auxiliary diagnosis of Alzheimer’s disease.(1)An Alzheimer’s classification model based on the migration of the Med-3d network was proposed to improve the classification accuracy of Alzheimer’s disease.This deep learning model utilized the idea of transfer learning to transfer the information in the source domain of the Med-3d network to the classification task of Alzheimer’s disease.The ADNI dataset was used to retrain the Med-3d network,and the recognition accuracy,diagnosis accuracy and recall rate of Alzheimer’s disease tripartite task(AD,MCI,NC)reached 89%,96% and 89%.In order to improve the performance of the model,the channel attention was further added on the basis of the migration model to improve the performance of the model.The accuracy of the three classification tasks was 92%,the diagnostic accuracy was 93%,and the recall rate was 89%.(2)A multi-scale feature recombination Alzheimer’s classification model was proposed.OASIS data set was used for training.The recognition accuracy,diagnosis accuracy and recall rate of Alzheimer’s disease triad task(AD,MCI,NC)reached93%,95% and 87% respectively.The ablation results show that the multi-scale and Channel Shuffle modules have a significant effect on improving the classification ability,and the comparative experiments also show the superiority of this method.(3)A new auxiliary diagnosis system for Alzheimer’s disease based on hot update(Hot update is when users are unaware of software updates)SDK is proposed to solve some problems existing in the auxiliary diagnosis system in hospitals.At present,the machine classification algorithm model in most hospitals is old and cannot be updated in time,but the accuracy of classification model is constantly improving.Some hospital machine classification algorithms are still at the level of a few years ago,or even longer,and updating algorithms means replacing machines or upgrading software versions,which is costly.In order to solve this pain point,this paper developed a Web-based thermal update of Alzheimer’s disease auxiliary diagnosis SDK package,which can timely update the latest algorithm model and reduce the replacement cost of hospitals.