A Survey on Diabetic Retinopathy Classification Based on Deep Learning Approaches

Diabetic Retinopathy (DR) is a severe retinal disease that impact on blindnessto the patients who ha diabetes for long period of time. Thus, early detection and accurate classification of DR in retinal fundus images are important for reducing the risk of vision loss. However, machine learning models has limitations to learn subtle and complex feature effectively which is crucial for DR classification that lead to misclassification of DR severity levels. Therefore, Deep Learning (DL) models are mostly used for DR classification and also for extracting important subtle features from retina image to detect and classify the DR effectively. The DR classification based on DL methods is primarily divided into with segmentation and without segmentation. The DL methods used for DR classification are Convolution Neural Network (CNN), Long Short-Term Memory (LSTM), Bi-directional LSTM, Gated Recurrent Unit (GRU), Deep Neural Networks (DNN), Transformers and so on. This survey analyzes the strengths, relationships, and challenges of the DL-based classification approaches for DR.