Application Of Deep Learning Technology In Classification And Prediction Of Alzheimer's Disease

Alzheimer's disease(AD)is a progressive and neurodegenerative disease happened in elderly people.AD brings suffers and economic burdens to the patients and their family as well as the society.A series of scales are used to diagnose for AD in clinically.Not only the doctors' workload increase greatly,but the diagnosises are subjective.When patients show cognitive impairment,it already develops into the late stage of the disease.Alzheimer's disease can't be cured but only delay the process by drugs.Therefore,early disgnosis and early intervention is very important for AD patients.Deep learning is a promising technology for intelligent healthcare.The aim of the study is to achive classification and prediction of AD based on deep learning technology.The study uses three networks,such as simple convolutional neural network(CNN),Dense Net,and multi instance learning network(MILNet)in AD classification and prediction research.The structural magnetic resonance imaging data in ADNI1 dataset is used and separated as training set,validation set and testing set.For the classification results of the model,the accuracy,recall,sensitivity,specificity,F1 score,and the area under the ROC curve(AUC)were evaluated.In order to use as much data as possible,a new data augmentation method is proposed as adding follow-up data into training set as independent subject.As a result,the accuracy of the testing set can be improved by 2 percent.And the clinical data from Affiliated Drum Tower Hospital of Nanjing University Medical School is used as an extra testing set.In conclusion,the Dense Net model achieves the best classification performance with accuracy of 0.91 for AD/NC classification,0.76 for MCI/NC,0.73 for AD/MCI,0.58 for AD/MCI/NC,0.66 for p MCI/s MCI,and 0.45 for AD/p MCI/s MCI/NC.Based on the MILNet,the study found that: the left lingual gyrus performs well in AD/NC classification;the right upper temporal gyrus and bilateral hippocampus performs well in MCI/NC classification;the right frontal lobe performs well in AD/MCI classification;the right temporal lobe performs well in p MCI/s MCI classification.These results might suggest that: in the early stage of Alzheimer's disease,brain atrophy mainly occurs in the right temporal lobe and bilateral hippocampus;in the late stage of the disease,the left lingual gyrus and the right frontal lobe atrophy seriously.