Diabetic Retinopathy Detection using a Lightweight Edge Intelligence-based Technique

Bangladesh currently ranks 8th globally in the number of diabetic patients, with the International Diabetes Federation projecting a rise from 460 million to 690 million patients by 2045. One of the serious consequences of diabetes is diabetic retinopathy, an eye disease affecting 80 percent of patients who have had diabetes for more than 10 years, damaging retinal blood vessels and potentially leading to blindness. To address this issue, we have developed an automated system, utilizing deep learning Convolutional Neural Network (CNN) frameworks for detecting diabetic retinopathy in its 5 stages. The trained model is then integrated into a mobile Android application. This innovative solution allows users to detect the condition remotely, providing instant results. While similar solutions exist globally, our project marks the first introduction of such a system to Bangladesh, with a focus on potential use in rural areas. By utilizing readily available Fundus Camera attachments, the system can produce high-definition retinal imaging and perform diabetic retinopathy screening on an edge device achieving an accuracy of 81.6%.