Classification of Alzheimer's Disease Based on Residual Network and Attention Mechanism
This study addresses the problem of relying mostly on two-dimensional slices of MRI data,which is difficult to fully capture the detailed information of the brain,and the low accuracy when using three-dimensional image data,which is prevalent in the current classification models for Alzheimer's disease(AD),Mild Cognitive Impairment(MCI),and Normal Controls(NC).In this paper,we propose an innovative ap-proach to construct a new classification model using 3D brain MRI data,combining the channel attention mech-anism and deep residual networks.By introducing a channel attention module in the residual network,the mod-el is able to focus more on key features while ignoring irrelevant background information.In addition,the last two residual modules use deep separable convolution instead of traditional convolution,which effectively re-duces the model parameters and lowers the risk of overfitting.Experimentally verified,the model achieves 87.12％accuracy on AD/NC classification and 61.60％accuracy on AD/MCI/NC triple classification problem,which are significantly better than other existing models.This not only improves the accuracy of AD classifica-tion,but also simplifies the data preprocessing process,which is of great significance for future AD diagnosis and research.
NETL
NSTL
万方数据
