Enhancing Pneumonia Detection using Vision Transformer with Dynamic Mapping Re-Attention Mechanism

Pneumonia is a prevalent respiratory infection with potentially life-threatening consequences. In this research, we propose a novel deep learning approach to enhance pneumonia detection using the Vision Transformer (ViT) architecture with a dynamic mapping re-attention mechanism. The re-attention mechanism in the ViT architecture facilitates the flow of information between patches of the input image, enabling contextual understanding through self-attention. This mechanism captures long-range dependencies and enhances the ViT's ability to effectively recognize and interpret images. The dynamic mapping attention mechanism allows the model to dynamically assign attention weights to different regions of the input data based on their relevance to pneumonia detection. By adaptively adjusting the attention weights, the model can capture both local details and global dependencies, enabling more accurate diagnosis. The ViT utilizes the dynamic mapping re-attention mechanism with attention dropout. Here the performance of this ViT compared to the conventional ViT and the ViT incorporating the re-attention mechanism. This comparative analysis offers valuable insights into the effectiveness of the proposed dynamic mapping re-attention mechanism. Additionally, it highlights the added benefits of attention dropout in enhancing the overall performance of the model.