Enhanced model to detection of Diabetic Retinopathy using Deep Learning techniques

Abstract

Diagnosing and treating diabetic retinopathy (DR) early on can prevent vision loss. None, moderate, mild, proliferate, and severe are the top five DR phases. This work presents a deep learning (DL) model that identifies all five stages of DR more accurately than earlier approaches. the suggested method with image enhancement using a contrast limited adaptive histogram equalization (CLAHE) filtering algorithm in conjunction with an enhanced super-resolution generative adversarial network (ESRGAN), and circular mask with using random search for hyperparameter. The next step was using augmentation techniques to create a balanced dataset using the same parameters for both scenarios. The created model outperformed previous techniques for identifying the five stages of DR, with an accuracy of 90% , using Efficient net B3 applied to the Asia Pacific TeleOphthalmology Society (APTOS) datasets.