An Ensemble Deep Learning Framework for Prediction of Diabetic Retinopathy

The disease Diabetic Retinopathy represents the main factor behind blindness because medical teams need to discover it early. Recent research in deep learning demonstrated breakthrough DR detection results through the development of Efficient Net, Vision Transformers (ViT) and attention-based combined models. All clinical applications benefit from Efficient- Net due to its high accuracy level supported by low parameter requirements. Through its design ViT maintains extended structure dependencies and performs faster generalization than conventional CNNs do. Combining Efficient Net with ViT through attention mechanisms enables both improved diagnosis performance and human-readable analyses that focus on significant retinal areas. The models achieve high diagnostic accuracy and excellent generalization when applied to different stages of the APTOS 2019 dataset which helps clinicians make better decisions and monitor patient screening efficiency. The architectures demonstrate potential to improve diagnostic precision of DR and shorten diagnostic times which provides valuable benefits to clinical healthcare.