Deep Learning Models for Diabetic Retinopathy Detection: A Review of CNN and Transformer-Based Approaches

: This article reviews the progress of many deep learning neural network models in the detection of diabetes retinopathy (DR), and discusses the significance of these advances in clinical practice. It examined improved Convolutional Neural Networks (CNN) and Transformer based DR detection models. Models based on CNN, such as Romero Oraa's framework, two-layer neural networks, and weighted fusion deep learning networks, have shown promising results in addressing challenges such as lighting and image quality. Transformer based models, including dual transformer encoder models and self supervised image transformers, utilize their unique architecture to improve performance. These models improve the accuracy and efficiency of diagnosis, promoting the effectiveness of early intervention for DR treatment. In addition, the integration of these advanced technologies not only simplifies the diagnostic process, but also has the potential to alleviate the burden on the healthcare system by providing scalable solutions for extensive screening, ultimately helping to improve patient outcomes.