Diabetic Retinopathy Severity Detection an Automated Tool

Diabetics start to occur when there is not enough insulin secretion in the body. Diabetes is the primary cause of diabetic retinopathy in patients, which leads to vision impairment. The main reason for vision impairment is that lesions get accumulated in the retina. Early detection of diabetic retinopathy using the fundus images surely helps in faster recovery of the patient. Manual detection of the severity of diabetic retinopathy requires time, which may cause a delay in cure. To automate the process of severity detection, a convolutional neural network (CNN) is developed and presented, which is capable to predict the severity of the diabetic retinopathy in the fundus images. A model developed can learn the features of the retinal images and predict either of the five stages of diabetic retinopathy that is non-proliferative diabetic retinopathy, mild non-proliferative diabetic retinopathy, moderate non-proliferative diabetic retinopathy, severe non-proliferative diabetic retinopathy, and proliferative diabetic retinopathy. Images were enhanced using various methods. After conducting tests on the model using random test data, the model proved to be reliable resulting in 95.5% specificity and 93.5% sensitivity. With this specificity and sensitivity, our model is reliable when deployed to the real world where the doctors in the medical field can rely on in detecting the presence of diabetic retinopathy.