Computer aided diagnosis systems using deep learning for retinal diseases: A survey

Abstract

Classification and segmentation of blood vessels, lesions, exudates in the retina play a great role in the detection and classification of different ocular diseases like diabetic retinopathy, glaucoma and Familial Exudative Vitreoretinopathy (FEVR). To monitor and analyze microvascular and systematic diseases, classification and segmentation of blood vessels and optic disk are primary tasks. Many research works are being carried out from past one decade concentrating on various classification and segmentation algorithm which helps the ophthalmologists in the quicker diagnosis of retinal diseases. Image processing-based segmentation algorithms were used in the early days. After the evolution of Machine Learning (ML) and Deep Learning (DL) algorithms, many researchers started focusing on applying it in various fields. Here the review focuses on medical image analysis of the retina. In this review, a detailed survey is done on deep learning methods applied for retinal diseases in fundus images with high-quality papers published from 2012 to 2021. It typically focuses on appropriate preprocessing methods, classification, and segmentation models designed with deep learning algorithms. A glimpse of available and mostly preferred datasets for various retinal diseases, costs, and benefits of each model are given and highlighted in terms of the results produced. An attempt is made to assess the deep learning architectures with the help of accuracy, sensitivity, specificity produced by each model. The impact of this study helps in future research directions in medical image analysis and the segmentation of retinal diseases.