Semantic Segmentation of Ultra Widefield Images Displaying Regmatogenous Retinal Detachment and Comparative Analysis of Hybrid Deep Learning Techniques

Rhegmatogenous retinal detachment (RRD) is a vision-threatening condition that requires early and accurate detection for effective treatment. In this study, we utilized Ultra-Widefield Fundus Images to develop a deep learning-based semantic segmentation framework for identifying RRD regions. Three deep learning architectures such as U-Net, U-Net++, and ResU-Net were implemented. The U-Net model had 98.26% accuracy, 98.56% precision, 97.77% sensitivity, and 99.08% specificity. The U-Net++ model had 93.84% accuracy, 93.30% precision, 91.50% sensitivity and 95.78% specificity. The ResU-Net model had 98.16% accuracy, 98.50% precision, 97.55% sensitivity and 99.05% specificity. As U-Net proves to be the most effective model due to its outstanding performance and computational efficiency, these findings underscore the potential of deep learning in automated RRD detection. This advancement paves the way for enhanced computer-aided diagnosis (CAD) systems, aiding ophthalmologists in early disease detection and treatment planning.