MobileViT-based classification of Alzheimer's disease

Alzheimer's disease (AD) is a progressive neurodegenerative disease, and the detection and discovery of early AD is still difficult due to the insidious nature of its pathogenesis. With the rapid development of deep learning nowadays, Transformer has been commonly used in the field of computer vision and the results are very good. Therefore, in this paper, we propose a MobileViT style model BrMobile for AD diagnosis using 1.5T T1-weighted baseline magnetic resonance images (MRI). The model combines the features of the Transformer and Self-Attention mechanisms and was tested on the ADNI dataset, achieving excellent accuracy. The accuracy of the method was 96.7%, 91.8% and 90.9% for the classification of AD with NC, MCI with NC and AD with MCI, respectively. In addition to these dichotomous tasks we also tested multiclassification AD vs. NC vs. MCI with a classification accuracy of 88.8%. Compared to other state-of-the-art studies, this method provides an effective and accurate diagnosis in AD classification due to the light weight and efficiency of the model.