Aggregating Dual Attention Residual Network and Convolutional Sparse Autoencoder to Enhance the Diagnosis of Alzheimer’s Disease

For the diagnosis of Alzheimer’s disease, we proffer a pioneering computer-assisted methodology, the salient innovations of our approach are threefold: 1) A dual attention residual deep neural network has been proposed to capture localized features inherent in MR images; 2) A convolutional sparse autoencoder, fortified with dual hidden strata, was devised to extract both global and spatial intelligence from PET images, thereby bolstering diagnostic efficacy; 3) An innovative multi modal framework has been introduced, which integrates MRI and PET images into a unified network architecture for end-to-end learning. Pursuant to rigorous experimental evaluations on the publicly accessible dataset from the ADNI, empirical results underscore that our propounded multi-feature fusion model attains an impressive classification accuracy, registering at 95.12%.