Abstract
Alzheimer's disease is one of the most common types of neurodegenerativedisease, characterized by the accumulation of amyloid-beta plaque and tautangles. Recently, deep learning approaches have shown promise in Alzheimer'sdisease diagnosis. In this study, we propose a reproducible model that utilizesa 3D convolutional neural network with a dual attention module for Alzheimer'sdisease classification. We trained the model in the ADNI database and verifiedthe generalizability of our method in two independent datasets (AIBL andOASIS1). Our method achieved state-of-the-art classification performance, withan accuracy of 91.94% for MCI progression classification and 96.30% forAlzheimer's disease classification on the ADNI dataset. Furthermore, the modeldemonstrated good generalizability, achieving an accuracy of 86.37% on the AIBLdataset and 83.42% on the OASIS1 dataset. These results indicate that ourproposed approach has competitive performance and generalizability whencompared to recent studies in the field.