Inspecting Transfer Learning Models for Diabetic Retinopathy Detection

Astonishingly, 285 million individuals of the global diabetic population, exhibit sign of diabetic retinopathy. It has become a prominent global health challenge arising as a complication of diabetes. Diabetic Retinopathy damages the retina's delicate blood vessels leading to blurred vision, dark spots in vision, retinal detachment and ultimately vision loss. To address this concern, we implemented a study to explore the possibilities for automated DR detection using advanced deep learning techniques. We assessed four models namely, EfficientNet-B5, ResNet152V2, EfficientNet-B6 and MobileNetV2, on the Asia Pacific Tele-Ophthalmology Society (APTOS) dataset of retinal images. MobileNetV2 emerged as a top-performing model, closely followed by EfficientNet-B6, with the potential to revolutionize diabetic retinopathy diagnosis and reduce the risk of vision loss. These findings contribute to the field of medical image analysis, offering promise for enhanced early detection of diabetic retinopathy.