Identifying the Level of Diabetic Retinopathy Using Deep Convolution Neural Network

Diabetic Retinopathy is the leading cause of blindness in the last 100 years. The traditional screening process for DR and its stages takes a lot of time, and it is not practical. Using machine learning techniques and image processing, we can automate detecting diabetic retinal disease and disease stage with acceptable performance. In this work, we have used multiple deep convolution neural networks (CNN) with the same architecture of InceptionV3. Each of the pre-trained Inception V3 architecture is retrained with 2200 preprocessed and leveled images. The dataset is preprocessed using multiple high performing and effective image processing techniques. Then the newly trained models are used for identifying the level of DR. In the final stage, we use a voting scheme for classifying the level of DR from the output of each model. We have achieved 90.5% accuracy in binary classification (Normal/DR) and 81.1% accuracy in 5-class classification.