Detection of Diabetic Retinopathy in Retinal Fundus Images using DenseNet based Deep Learning Model
P. Saranya, S. Kiruthika Devi, B. Bharanidharan
Abstract
Diabetic Mellitus is one of the world’s most common diseases. Diabetes is quite prevalent, and it produces a variety of health issues such as Diabetic Retinopathy, nephropathy, diabetic foot, and so on. Diabetic Retinopathy is the most prominent problem (DR). DR starts with no symptoms or minor vision problems and escalates to the point when vision loss is a possibility. Since diagnosis takes time and ophthalmologists are scarce, patients endure vision loss even before they are diagnosed. So, an early detection of DR may help to mitigate the problem. To diagnose DR, however, numerous physical tests are required and in the early phases of the disease, it is difficult to diagnose by exams. As a result, a new diagnostic technique must be devised to detect the disease before it manifests itself in a test, allowing it to be treated sooner. The objective of the proposed model is to provide an automated diagnosis model for DR detection utilizing DenseNet-based deep learning models. As an input, the classification model was given a pre-processed retinal image that had not been enhanced with features. Only a few preprocessing steps are done on the noisy images to increase DR detection accuracy and achieved the maximum accuracy and precision of 0.83 and 0.99 respectively.