Automated Retinal Image Analysis

Abstract

Eyes are not only the windows to our soul but also to our body’s overall health. From the past few years’ doctors and ophthalmologists are diagnosing many serious diseases from eyes and in many cases eyes help to early detect many diseases which cause serious problems if not detected early. Diabetes, high blood pressure, autoimmune diseases, sexually transmitted diseases and cancers are among the diseases that can be detected during an eye exam. This is because eyes represent unobstructed view of our blood vessels, nerves and connecting tissue. The eye has the same microscopic tissue as our other major organs. Hypertensive Retinopathy is the retinal abnormality caused by the high blood pressure. Timely treatment of Hypertensive Retinopathy is very important because it can cause permanent vision loss. Similarly diabetic retinopathy is a diabetes complication that affects eyes.. The proposed system consists of novel method for classification of vessels as arteries and veins using convolutional neural network. Our project scope encompasses the domain of Image Processing and will be focusing on segmentation and classification of vascular structure of eye through Retinal image analysis. We aim to make use of machine learning and image processing techniques to identify and differentiate the arteries and veins in the retinal structure of an eye in order to tackle the time-consuming problem of ophthalmologists i.e. manual analysis of retinal images to diagnose Hypertensive Retinopathy, diabetic retinopathy and many other retinal diseases. The deliverables constitute a system that will classify arteries and veins i.e. result of analysis of retinal images. The main objective behind this project is to develop a computer aided diagnosis (CAD) system that can be implemented on a large scale in the hospitals everywhere in the country. The system will be able to detect Hypertensive Retinopathy early, but our proposed system will capture a retinal image and analyse it to classify the image into arteries and veins using image processing and machine learning techniques. After that by applying geometry detection and measure technique of image processing, we can easily detect the hypertensive retinopathy. The strength of the project is using machine learning technologies for segmentation of retinal vascular structure in order to improve the early detection of many retinal diseases. For the creation of system, we also manually created the dataset by labelling of arteries and veins which is a novel approach used for their classification. Through convolution neural network, accuracy increases, which helps doctors, screen far more patients than currently possible.