Detection and classification of diabetic retinopathy using convolutional neural networks and fundus images

Computer Engineering

There are countless number of people with diabetes around the world. Diabetic Retinopathy (DR), a major complication of diabetes, is a retinal disease that results in blindness. Early detection can prevent or delay the loss of vision due to DR. For early detection, it is essential for diabetic patients to undergo frequent retinal tests. Diagnosing DR using fundus images is a complex process and thus difficult and time consuming task requiring professional expertise to identify if the DR traits are present in eye. We propose a system that can provide immediate feedback thus effectively simplifying the process. Convolutional Neural Network (CNN) architecture based on fundus image database to accurately diagnose DR with minimum efforts on the part of patient and doctor is developed. The database contains multiple cases of retinal hemorrhage, micro-aneurism, cotton-woolspots, etc. The retinal image of the patient is fed to the system where image is processed to detect signs of DR. The network is also trained to classify various stages of DR depending on severity such as nonproliferative DR or proliferative DR. The output contains a report detailing if DR is present or not in the retinal image in question and if present, which stage it is.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.