The development of a deep learning system for the detection of diabetic retinopathy

Abstract

Diabetes patients run the risk of developing diabetic retinopathy, an eye disorder that can impair vision and lead to blindness. Permanent and total blindness may be avoided with early detection. Consequently, a reliable screening mechanism is required. According to academic literature, deep neural networks (DNNs) have become the most preferred method for DR detection. In medical image classification, convolutional neural network (CNN) models are the DNN techniques that are most frequently applied. A new CNN architecture requires a lengthy and complicated effort to develop. Furthermore, it can be difficult to train a lot of factors. As a result, utilizing pre-trained models has been proposed in recent years as a transfer learning technique as opposed to building CNNs from scratch. To facilitate the screening process, we present a wide range of CNN architectures in this study along with multiple classifiers that can identify various stages of diabetic retinopathy.