Early Detection of Alzheimer's Disease using Vision Transformer based Convolutional Neural Network

In recent scenarios, Alzheimer’s Disease (AD) has posed early recognition challenges due to complex and overlapping symptoms by making difficulties in timely care and intervention. The existing models such as Convolutional Neural Networks (CNN) struggled with limited interpretability. In this research, Vision Transformer based CNN (ViT-CNN) is proposed for AD diagnosis. The proposed ViT-CNN utilizes query-attention mechanism which mainly focuses on the specific region of input that improves the interpretability issues. This procedure begins with the collection of input images from AD Neuroimaging Initiative (ADNI) dataset. Then, the pixel intensities are normalized and data augmentations such as flipping is performed to increase diversity of input data. After that, ViT is incorporated to extract features by capturing the spatial as well as temporal relationships within the image. Finally, the extracted features are further processed into the CNN model for early diagnosis of AD detection. From the results, proposed ViT-CNN outperformed the existing CNN in terms of accuracy (0.83), specificity (0.79) sensitivity (0.85), and F1-score (0.85) respectively.