Automated Diabetic Retinopathy Detection and classification using ImageNet Convolution Neural Network using Fundus Images

Diabetic retinopathy (DR) is one of the major complications of diabetes and it occurs when the blood vessel rupture and leakage of blood get collected into the retina. DR is one of the major causes of vision loss. Retinal bleeding is the foremost symptom of diabetic retinopathy and it is asymptotic in nature. So the early detection of DR can prevent patients from visual loss. This paper proposes an automated system to detect and classify diabetic retinopathy by using ImageNet model to achieve higher accuracy. Publicly available dataset `Kaggle' of retinal images has been used to compare and analyze the performance of the algorithm. The ImageNet model achieved an impressive performance in DR detection and classification. The model achieved 98.6% of training accuracy. The model has achieved an 86.6% accuracy of No DR, mild with 62.5%, moderate with 66.6%, severe with 57.1% and PDR with 42.8%. The accuracy can be improved by training the network with more number of images in each classification.