A Comparative Study on Performance of Popular Deep Learning Models to Detect Diabetic Retinopathy

Currently diabetic retinopathy is best diagnosed manually by a trained professional who is not only expensive but also takes a lot of time. This paper focuses on the use of deep learning to provide an alternative path which is cheaper and faster. The concept of transfer learning in CNN was used to predict the scale of the disease in individuals. By using different architectures of neural networks, their ability to detect DR in patients was tested and compared. It was concluded that Inception Resnet V2 Architecture gave the best results when tested for accuracy on Cohen's kappa metric.