Classification of diabetic and normal fundus images using new deep learning method

摘要

Diabetic Retinopathy (DR) is a complication resulting from diabetes due to changes inthe retinal blood vessels. The person may lose eyesight because of damaged bloodvessels. Although diabetes cannot be cured, but diabetic retinopathy can be treatedsuccessfully with laser surgery. So far, there has been a lot of research for diabetesdiagnosis and classification of eye images by machine-learning algorithm and neuralnetwork, but these algorithms are good for a few images and features are extracted frompictures by manual or semi-automatic methods. With the emergence of deep learningleading to very good results in various areas of computer vision, the ConvolutionalNeural Network (CNN) was used by researchers as one of the most active algorithms inthis area. One of the areas where this type of network has achieved very good results isthe classification of the images. In this article, one of the available architectures of thedeep convolutional neural network called ResNet is reviewed and will be used forclassification of eye fundus Images into two groups of normal and diabetic images. Thisarchitecture has been used on the collection of eye fundus images available on theKaggel website and simulation result has achieved 85% accuracy and 86% sensitivity.