Diabetic Retinopathy Classification Using Swin Transformer: A Vision Transformer-Based Approach

Early and precise detection is the most effective way to manage diabetic retinopathy (DR), the leading cause of blindness worldwide. Vision Transformers (ViTs) has also recently developed with increased feature representation capabilities, and has contributed to the great success of deep learning-based methods in automating DR detection. To introduce a Swin Transformer-based classifier for automatic detection and classification of DR severity using the APTOS 2019 Blindness Detection dataset. The hierarchical vision transformer SWIN transformer (shifted window attention) captures both local and global retinal characteristics efficiently. To train our proposed model, and to use transfer learning and deploy pre-trained Swine Transformer weights on retinal fundus Images, which helps to gain rich hierarchical information. These characteristics are assessed against five stages of the severity of DR: No DR, Mild, Moderate, Severe DR and Proliferative DR. This article applies various data preparation methods like image augmentation, noise removal, and contrast adjustment, to help improve the robustness of the model. Use Running City Game Optimiser (RCGO) Try to find out which model parameters are most important next step: Improve classification accuracy To get a better result from the model, to implement both adaptive learning rate scheduling, along with methods for dealing with class imbalance, such as weighted sampling and focal loss. Notably, the Swin Transformer-based classifier also score-wise outperforms classic CNN-based architectures such as ResNet, VGG, and Inception. Microaneurysms, hemorrhages, and exudates also play an important role in the development of DR, which is well-captured by the implemented model. Due to the low computing burden and strong generalizability of the proposed method, it holds a promise for automated DR screening as well as telemedicine applications in practice. Studies in deep-learning-based medical-image analysis can draw upon this central effort, demonstrating the potential of Vision Transformer architectures for ocular disease detection.