Automatic detection of diabetic retinopathy

Abstract

Diabetes mellitus is a major health problem in modern India. There are estimated 73 million cases of diabetes mellitus in India, of which around 22% have diabetic retinopathy. So early detection and treatment is imperative for the prevention of visual morbidity. Diabetic retinopathy is diagnosed manually by fundus examination using ophthalmoscope which requires an ophthalmologist and is time consuming, expensive and can have subjective variations. In an automated method, using artificial intelligence these difficulties can be circumvented and early detection of diabetic retinopathy is facilitated. This study is based on blood vessel segmentation, detection of hard exudates, micro aneurysms and statistical feature extraction. Here 100 retinal images having varying stages of diabetic retinopathy were taken from MESSIDOR database and image processing done using MATLAB 2018a software. This is an effort to classify diabetic retinopathy into mild, moderate and severe cases and help in detecting diabetic retinopathy in early stage itself so that visual morbidity can be prevented