Optimized Deep Learning Framework for Automated Diabetic Retinopathy Detection

Diabetic Retinopathy (DR) is one of the leading causes of blindness worldwide, emphasizing the need for early detection and timely intervention. Traditional manual screening methods are time-consuming, subjective, and prone to variability, necessitating the development of automated and highly accurate diagnostic tools. This study introduces an advanced deep learning framework that integrates multiple state-of-the-art architectures to enhance DR detection performance. Vision Transformers (ViT) are utilized for superior feature extraction, capturing both global and local retinal abnormalities effectively. An attention-based U-Net is incorporated for precise lesion segmentation, improving localization and interpretability. Additionally, Efficient-Net, combined with Swarm Intelligence optimization, is employed for classification, ensuring high accuracy while maintaining computational efficiency. The proposed model is evaluated on the IDRiD and Messidor-2 datasets, demonstrating its robustness across diverse retinal images. Experimental results indicate that our framework achieves an outstanding accuracy of 99.12%, outperforming existing deep learning-based methods in DR detection. The integration of attention mechanisms and optimization techniques significantly enhances the model's reliability and generalization capability. Furthermore, the framework is designed for real-time applications, making it suitable for deployment in clinical settings. The high accuracy and efficiency of our model highlight its potential as a valuable tool for ophthalmologists, aiding in early diagnosis and treatment planning. By automating DR detection, this approach can help reduce the burden on healthcare professionals and improve patient outcomes. The findings underscore the effectiveness of deep learning-driven solutions in advancing retinal disease diagnosis and enhancing ophthalmic healthcare.