Detection of Diabetic Retinopathy

Abstract

Biomedical engineering faces a significant challenge in assessing the physiological changes occurring in the human body non-invasively, particularly when detecting abnormalities in the eye due to the complexities involved. Traditional disease identification techniques in retinal images rely on manual intervention, which can be error-prone and have a low success rate. Long-term diabetics are susceptible to the multistage progressive condition known as diabetic retinopathy, which has aberrant retinal imaging characteristics such microaneurysms, haemorrhages, and exudates. It uses the dataset which consists of total 13970 images which includes the images of all stages collected from kaggle APTOS competition. Deep learning, a key method in health informatics, is used in this work to identify diabetic retinopathy and its stages.. Residual Network and Densely connected convolutional network deep learning approaches are employed, and transfer learning is used to achieve good accuracy in identifying the stages of diabetic retinopathy. The accuracy obtained by using the ResNet101 is 92.37% and with the DenseNet121 is 98.37%.