Diabetic Retinopathy Grade and Macular Edema Risk Classification Using Convolutional Neural Networks

Diabetic retinopathy (DR) is one of the major causes of blindness in the western world. Effective treatment of DR is available, when detected early enough, which makes this a vital process. Computers are able to obtain much quicker classifications once trained, giving the ability to aid clinicians in real-time classification. This work employed a deep convolutional neural network (CNN) based method for diabetic retinopathy classification. Three independent CNNs were employed for the classification of DR grade, macular edema risk and multi-label, which included the combination of both grade and risk classes. A fusion method was used to combine all features extracted by the CNNs and make the final classification result. The classification accuracy of the grade and risk were 0.65 and 0.72, respectively. The classification results showed the proposed network fusion method can improve the performances on both task - DR grading and macular edema risk.