Detection and classification of Diabetic Retinopathy Lesions using deep learning

Abstract

Diabetic retinopathy (DR) is a frequent consequence of diabetes mellitus that induces retinal lesions, which affect vision. DR can lead to poor vision and blindness if not treated quickly. Unfortunately, DR is not reversible, and therapy just prolongs vision. As a result, tools are needed that initially identify and prevent poor vision in diabetics at an early stage. Early identification and treatment of DR can decrease the risk of vision loss considerably. Unlike computer-aided diagnosis systems, the manual diagnosis of DR retina fundus images by ophthalmologists is time-consuming and is prone to misdiagnosis. Recent technological advances have brought optical imaging systems to the market in relation to smartphones, which allows for low power, DR viewing in a variety of settings. On the other hand, deep learning (DL) has recently emerged as one of the most widely used approaches for improving performance in a variety of fields, including medical image analysis and classification. The purpose of this chapter is to use DL models to create an automated DR detection for the modern eye model. Moreover, DL models are implemented with the color fundus retina images. Transfer learning models such as InceptionResNet, VGG, and DenseNet architectures are also utilized for the color fundus retina image analysis. F1 scores, accuracy, area under the receiver operating characteristic curve (AUC - Area under the ROC Curve), and Kappa score are utilized to measure the performance of DL models for DR detection. It contributes significantly to improve DR identification by using different artificial intelligence (AI) methods with a variety of the color fundus retina public datasets. © 2023 Elsevier Inc. All rights reserved.