AD Classification Methods Based On Deep Learning And MRI Image

With the advent of an aging society,research on dementia is becoming more and more important.Alzheimer’s disease(AD)is the most common and incurable dementia,so it is especially important to distinguish patients with AD from normal or other dementia patients.In recent years,the rapid development of deep learning has pointed out a new research direction for AD classification.This paper mainly uses magnetic resonance imaging(MRI)data from ADNI to study AD classification methods based on deep learning from the following three aspects:1.AD classification based on deep metric learning and MRI images.metric learning is to learn a mapping rule that makes samples belonging to the same category closer and samples of different classes farther apart.This forces the mapped features to be more discriminating.For the classification model,the more discriminative features can speed up the convergence of the model and improve the classification ability of the model.In this paper the 3D convolutional neural network is used,which has a large amount of parameters results in a slow convergence.In this paper,the metric learning regular term is added to the deep learning model’s loss function,which makes the model’s feature space more discriminative,so as to accelerate it’s convergence speed and improve it’s classification performance.2.AD classification based on attention mechanisms and MRI images.The attention mechanism is inspired by the way the brain processes visual signals.For different sample data,it screens and focus on different regions of interest for the target task.The region of AD lesions varies from patient to patient.This paper considers that the attention mechanism can obtain the unique attention weights of input’s feature map.Therefore,the attention module is embedded in the deep learning model,and the model is used for AD classification.3.AD classification based on multi-task learning and MRI images.Multitask learning is a transfer learning method that reduces model overfitting by sharing features among multiple related subtasks.The deep learning model used in this paper has a large amount of parameters and the MRI image dataset of Alzheimer’s disease classification is small,so the model is easy to overfit.In this paper,multi-task learning is used to classify AD to weaken model overfitting.Specifically,this paper includes three sub-tasks: AD classification,CDR regression,and MMSE score regression.During the model training process,the three subtasks promote each other,and finally the realization of each subtask goal is improved;in particular,the accuracy of AD classification is improved;and the overfitting phenomenon of the model is also improved.