Retinal Lesions Detection for Screening of Diabetic Retinopathy

Abstract

Diabetic retinopathy is an eye disease, which causes blindness in diabetic patients. Early detection and periodic screening of diabetic retinopathy can reduce the progress of the disease and reduce vision loss. Due to diabetic retinopathy, there are lesions in the retina. Retinal lesions are mainly dark and bright lesions. Lesions have different properties such as colour, shape, and size. Microaneurysms (MAs) and hemorrhages (HEMs) are dark lesions and exudates(EXs) are bright lesions. This work proposes a retinal lesions detection system for screening diabetic retinopathy. Top hat and bottom hat transform are used for the enhancement of the image for the dark lesion detection. The optic disc is suppressed first to facilitate further processing of bright lesion detection. Morphological opening and closing are used for the detection of the bright lesion. Two SVMs are used in this work, one SVM is used to classify the dark lesion into microaneurysms and hemorrhages, and other one classify the bright lesion into hard exudates and non-hard exudates.