A Survey on Diabetic Retinopathy Datasets

Abstract

Healthcare is a field generally operates with numerous images, where obtaining labels from the image data is a difficult task in many domains. Deep learning has helped with automating tasks like early and automated detection of many diseases, with great accuracy. So acquiring good datasets with properly labeled images is a prime condition in deep learning because quality cannot be compromised when it comes to dealing with cases in healthcare. Diabetic Retinopathy is becoming prevalent in many countries and the disease becomes nearly incurable after it reaches a severe condition and results in serious vision impairment and partial blindness in some cases. The disease can be cured properly if it is diagnosed early, but the analysis of the disease becomes costlier and complex if it gets severe. Computer-aided tools can help us in achieving early detection of Diabetic Retinopathy that helps in cutting down costs as well as curing the patients in a better way. Therefore, using deep learning can be useful in the early detection of Diabetic Retinopathy. So this paper includes a discussion on some good datasets available that could be used for understanding the lesions in the retina responsible for Diabetic Retinopathy and building deep learning models required for accurate analysis and early detection of the disease.