A Systematic Review of Deep Learning Methods for Fundus Image-Based Diabetic Retinopathy Classification

Diabetic Retinopathy (DR) is a microvascular complication of diabetes mellitus. DR is the leading cause of irreversible low vision and blindness. The early diagnosis of DR, i.e., DR classification and accurate DR classification, is required for vision conservation and early DR treatment. Computer-based techniques with deep learning made huge promises in DR grading from fundus images with various advantages. DR is the most frequent and perilous complication of diabetes. DR detection in a computer is gaining widespread use as deep learning methods expand across the field of medical image analysis to make results faster and more efficient than the traditional human screen methods. Transfer learning has also been applied as a low-cost alternative to overcome limited data challenges with diabetic retinopathy (DR) screening based on this review. ResNet and VGGNet tens and even hundreds of layers are the most common platforms used for DRgrading. APTOS2019 and EyePACS are the most common datasets used for DR classification. Other than that, some restricted DL models like SqueezeNet and MobileNet are recommended to be used for application of DR grading in restricted source of data and computational resource situations.