Improvement of the Application of Diabetic Retinopathy Detection Model

Abstract At present, CNN models for diabetic retinopathy detection have become a research focus and have achieved good results in experiments. However, still, these models have not been wildly used in real practices. The main reasons are the lack of interpretability and accessibility. Therefore, this paper proposed a new method which improved the interpretability through visualization based on results, and which increased the accessibility through web services and a message queue framework. Besides, in this paper, a report system was designed for doctors and a mobile APP was designed for patients. By means of that, the method was verified in real practices.