Comparative Analysis of Alzheimer's Disease Detection via MRI Scans Using Convolutional Neural Network and Vision Transformer

Progressive damage to brain neurons is caused by a neurodegenerative disease (ND) that the body cannot heal or restore. Dementia like Alzheimer's Disease (AD), which affects millions of lives each year, is a well-known instance of such illnesses. Despite extensive research, the aforementioned disorders currently have no viable therapies. However, a timely diagnosis is essential for the management of diseases. For neurologists, diagnosing NDs is difficult and needs years of education and experience. The paper focuses on the detection of Alzheimer's Disease via Brain MRI Scans using Convolutional Neural Network (CNN) which minimizes an image's high dimensionality without sacrificing its information and using Vision Transformers that are for image classification and apply an architecture akin to a Transformer over selected areas of the image. In the proposed work three different Vision Transformer namely: Vanilla Vision Transformer (Vanilla ViT), Deep Vision Transformer (DeepViT), Class Attention in Image Transformer (CaiT). The Vision Transformers turned out to be better than CNN as when training on fewer datasets, ViT exhibits inductive bias, which increases dependency on model regularisation or data augmentation (AgReg). In terms of accuracy and computing efficiency, ViT models exceed the present CNN by almost a factor of four.