Research On Alzheimer's Disease Classification Algorithm Based On Deep Neural Network

Alzheimer Disease(AD)is the most common type of Alzheimer’s disease,with a high incidence and prevalence in the elderly.Mild Cognitive Impairment(MCI)is in the transitional phase of normal controls(NC)and dementia.Studies have found that drug intervention in patients with MCI or early AD can improve the symptoms of the disease and delay the progression of the disease.Therefore,early diagnosis or prediction of MCI and AD patients is very practical.In the traditional AD classification study,relying mainly on prior knowledge to design representative features for classification tasks.This method of manually selecting features is time consuming and laborious and has a strong subjectivity.Therefore,its selects deep neural networks that can be automatically learned to extract deep features in Structural Magnetic Resonance Image(sMRI)and is used for classification of AD,MCI and NC.The specific work of it is as follows:Firstly,an Alzheimer’s disease classification algorithm based on migration learning and deep residual network is proposed.Firstly,according to the parameter sharing of CNN,the idea of transfer learning is introduced.The pre-trained deep residual network model in ImageNet is transferes to the sMRI data set for fine-tuning,instead of training a brand-new model from scratch.In addition,the entropy of all sMRI slices is calculated,and they are arranged in descending order according to the entropy size.Only the sMRI slices with larger information entropy are used to train the CNN model,thus enhancing the overall robustness of the model.Finally,AD classification experiments were performed based on the selected sMRI slices.The results show that the proposed method solves the problem of the small number of available medical image samples and shortens the training time of the CNN model.In the binary classifications such as AD/NC,AD/MCI and NC/MCI,the accuracy of 97.65%,90.36% and92.37% was achieved respectively,and the accuracy of 87.52% was also achieved in the AD/MCI/MCI classification.Secondly,a classification algorithm for Alzheimer’s disease based on depth features and nonlinear dimensionality reduction.First,the final layer of VGG16 is fine-tuned to extract the deep features of the sMRI slice.Then use LargeVis to reduce the dimensionality of nonlinear sMRI deep features.Finally,the deep features after dimensionality reduction are input into the Softmax layer for classification,and 93.52% high accuracy is obtained in the ternary classification of AD/MCI/NC.In addition,in order to prove the excellent performance ofLargeVis,several classic dimensionality reduction methods are compared.The results show that the classification method based on LargeVis dimensionality reduction reduces the loss caused by the dimension reduction process and reduces the computational cost of the algorithm.And the method is superior to PCA,MDS and t-SNE in terms of accuracy and specificity.Finally,the classification performance of the CNN middle layer features after LargeVis dimensionality reduction is compared.The results show that the classification performance of the last layer features is better than other CNN layers.In summary,this studies the application of deep neural network in Alzheimer’s classification,and proposes two classification algorithms based on convolutional neural network for Alzheimer’s disease.These two algorithms solve the problems caused by the small number of available medical images and high dimensional nonlinearity,and improve the binary classification and ternary classification accuracy of AD,MCI and NC.