Detection and Grading of Diabetic Retinopathy from Fundus Images by Applying Transfer Learning

Diabetic Retinopathy (DR) is a common complication of diabetes that can cause damage to the veins in the retina, resulting in significant vision loss. Timely treatment can help prevent this vision loss, but due to delays in diagnosis and a shortage of ophthalmologists, patients often experience vision problems before receiving proper care. Therefore, it is crucial to detect DR at an early stage. Automating the analysis of diabetic retinopathy is essential for addressing these challenges. While deep learning has achieved high accuracy in binary classification, its performance in multi-stage classification, particularly in the early stages of the disease, is less impressive. To address the abovementioned problem, Densely Connected Convolutional Networks are proposed for detecting Diabetic Retinopathy in retinal images. The proposed method has two cases (case 1- Binary classification, i.e., DR and No DR, and case 2 Multilevel classification, i.e., classification with five stages of DR). Also, the proposed method is tested with a publicly available retinal image dataset, i.e., APTOS. The accuracy of binary and multiclass classifications is 99% and 95% for the APTOS. The accuracy of the DR screening system is improved due to preprocessing.