Detection and Classification of Diabetic Retinopathy Using Artificial Intelligence Algorithms

Diabetic Retinopathy (DR) is considered as a sight-threatening complication of diabetes mellitus, the primary cause of blindness among working-age individuals. Ophthalmologists use fundus images to diagnose diabetic retinopathy and measure its severity by observing retinal lesions with high accuracy. However, diagnosing DR manually from fundus images requires a high level of expertise and effort from professional ophthalmologists. Early diagnosis of diabetes helps in saving the patient’s eye and preventing possible risky complications. In this context, the current paper proposed a model aims to detect DR using image processing and deep learning methods. A fully automatic diagnosis system that exceeds manual techniques to avoid misdiagnosis, reducing time, effort and cost were presented through this paper. A publicly available dataset of fundus images was used to apply the current paper’s proposed neural network model and transfer learning models, to classify each image into one of the five diabetic retinopathy stages. The simple proposed model achieved an accuracy of 66.68% in predicting the right label of the image. On the other hand, the second approach of fine-tuning the pre-trained models achieved higher testing accuracy (ranging from 93.13% to 100%,), which exceeds the current state-of-the-art results.