Detection of Retinal Neovascularization Using Optimized Deep Convolutional Neural Networks

Abstract

The most common disease that is found among people across the world is Diabetes and it is predicted to increase more in the upcoming years by The World Health Organization (WHO). People who are diabetic for a longer period are more likely to have Diabetic Retinopathy (DR), an eye disease which can lead to blindness and this cannot be reversed. One of the severe stage problems of DR is Retinal Neovascularization (RN), i.e., outburst of retinal blood vessels. Residual Network (ResNet) has an effective technique called Skip or Residual Connections which solves the problem of vanishing gradient during backpropagation. ResNet50 has 50 layers which is a deep network that omits signal representations and learns from residual representations leading to predict RN with 88.97% accuracy.