Diabetic Retinopathy Detection from Retinal Fundus Images Using Pretrained Convolutional Neural Network, VGG19

Diabetic retinopathy (DR) is a progressive syndrome that affects the retina and is a result of diabetes mellitus, in which high blood glucose levels cause impairment on the retina. For diabetic patients, particularly those of working age in developing countries, it is considered to be the main cause of blindness. Retinal fundus image examination for DR diagnosis can be used to decrease the possibility of vision loss of diabetic patients. Early identification may lead physicians to treat patients effectively, while in the opposite scenario, if retinal disease is detected too late, it may result in blindness. In order to prevent vision loss, these lesions should be properly identified and treated as away. To identify the early symptoms of DR, real-time automated systems can apply deep learning algorithms. As a result, it is easy to lower the possibility of human error and the effort required for the ophthalmologist. In this research, pretrained convolutional neural network (CNN), VGG19 is used to detect DR in retinal fundus images. The Asia Pacific Tele-Ophthalmology Society 2019 Blindness Detection (APTOS 2019 BD) dataset was utilized in this study which was collected by India’s Aravind Hospital. To calculate the effectiveness of the proposed model, the APTOS-2019 BD dataset is split into two sections, 80% of which are used for training the proposed model and 20% for testing it, in order to assess the model's efficacy. The efficacy of the suggested approach is calculated based on a number of factors, including accuracy, precision, sensitivity, and specificity.