UNet-CrossViT: A Deep Learning Model for Classifying Alzheimer's Disease

In recent times, deep-learning (DL) methods are implemented to classify neuroimaging data associated with Alzheimer's disease (AZD), and notable advancements have been achieved. Traditional Machine Learning (ML) has been shown to be helpful in aiding the diagnosis of AZD, however only a limited number of research have explored the capabilities of DL and deep transfer learning (DTL) in this intricate process. In this research, a novel DTL model called Cross attention multi scale Vision Transformer (CrossViT) is proposed to classify the AZD using MRI images. For this classification, a set of MRI images of AZD is collected from the Kaggle website. The collected dataset has 6400 images of MRI related to AZD and its various stages. The MRI images are preprocessed and divided into training and testing sets as 80% and 20%. After preprocessing, the segmentation is performed using the UNet architecture. The segmented images are fed to the proposed CrossViT model for AZD classification. The classification is performed as a binary classification, as it classified the results as either the AZD is present or not. The performances of the research model are assessed based on accuracy, specificity, sensitivity, f1-score, and precision. The research model has attained 99.63% accuracy, 99.51% sensitivity, 99.69% precision, 99.35% specificity, and 99.50% f1-score in training. The model attained 99.21% accuracy, 98.94% sensitivity, 99.25% precision, 99.10% specificity, and 99.04% f1-score in testing. The test results were compared with the current models for validation. The research model has outperformed all the compared model in comparison.