A Dual Deep Learning Approach with ResNet-50 and CycleGAN for Enhanced Retinal Lesion Detection

Diabetic Retinopathy (DR) is a vision-threatening complication in diabetic patients caused by high blood sugar levels, which affects the blood vessels in the retina. This research presents a dual deep-learning strategy involving ResNet-50 and CycleGAN architectures to detect and grade DR severity levels. CycleGAN is used for pre-processing the images by resizing, normalizing, denoising, and augmenting the dataset. While ResNet-50 is used for extracting Features from the retinal fundus images, capturing subtle retinal lesions associated with DR, and based on the lesions captured, the DR is classified according to its severity level. These two deep-learning strategies improve classification accuracy. Retinal fundus images are classified into 4 severity levels: Mild, Moderate, Severe, and Proliferative DR (PDR). This hybrid model exhibits encouraging results, illustrating the potential of ResNet-50 and CycleGAN in improving diagnostic accuracy and tackling global healthcare issues related to diabetic retinopathy.