Diabetic Retinopathy has become one of the most common causes of blindness and vision loss worldwide. It is difficult to find this disease in the early stages but finding it early can make you take the step to save your vision ability. Usually, the diabetic retinopathy diagnosis requires retinal fundus images and optical coherence tomography (OCT) using images. However, (OCT) is a time-consuming technique with expensive equipment. Therefore, patients are mostly diagnosed by taking retinal fundus images. In this paper, we have shown our technique based on CNN deep learning. We used a deep learning model to evaluate diabetic retinopathy detection. The problem that we faced to do image classification was to predict which class the given image belongs to, and the classes are 0 for Normal, 1 for Mild, 2 for Moderate, 3 for Severe, and 4 for Proliferative. We used the resnet34 model, which was pre-trained on the image net dataset then we replaced the final layer of pre-trained resnet34 with new layers. To optimize the model, we called the function of train and validate for (no of epochs) times. It returns a tuple of lists containing losses for all the epochs. No of images in Training = 3302 and no of images in Validation set = 360 after 60 epochs the results are Training accuracy = 92.1865, Validation accuracy = 79.444. So, we can improve our accuracy by increasing the dataset size, increasing the model complexity, using ensemble models, and increasing the number of epochs.
Index Terms- Cornea Classification, Convolutional Neural Network, Deep Learning, Diabetic, Retinopathy, Fundus images, Retina.