ABSTRACT
Diabetes is becoming a common disease almost in all the ages of people and is impacting a variety of human organs such as eyes, teeth, kidney, heart, skin, etc. This disease is causing various diseases in different organs of human body. Eyes are one of the important organs in human body through which one can see the beauty of the world. But, because of diabetes, many people are facing eye retinopathy problems. Due to these issues, many people are losing their eyesight and also going complete blind sometimes if it is not identified and treated at early stages. Damage to the retinal blood vessel obstructs the light passing through the optical nerves, rendering the diabetic patient blind. There are some manual procedures for detecting diabetic retinopathy (DR), one of which is screening, but it requires a professional ophthalmologist and a great deal of time. So, there is an immediate need of automation in detecting the diabetic retinopathy. By using convolutional neural network (CNN), we were able to identify multiple phases of DR severity. The main objective of this research work is to identify the diabetic retinopathy stages by analyzing the eye fundus images. Thus, we developed a hybrid deep neural network model to automate the diagnosis process of diabetic retinopathy using the transfer learning models with prior training of ResNet and DenseNet. The model is effective in detecting various stages of diabetic retinopathy.
