Diagnosis of Alzheimer's disease based on siamese cross-attention network
In order to improve the accuracy of longitudinal data in the classification of Alzheimer's Disease(AD),a siamese cross-attention network is proposed to diagnose AD through longitudinal data at two time points.Firstly,two structural magnetic resonance images(sMRI)at different times are input into the network,and the image features are extracted by using the 3D-Densenet network.Based on the clinical significance,the feature map of the image(M24)24 months from the first acquisition is enhanced by the self-attention mechanism,and then the cross-attention network is established for information fusion at different stages.The mutual response tensor generated by the dual time feature is classified by the full connection part.The experimental results show that the siamese cross-attention network improves the accuracy of AD classification,which shows that the network can effectively combine the disease information at different time stages and enhance the expression ability of disease characteristics.
国家科技期刊平台
NSTL
万方数据
