An efficient technique to detect and classify diabetic retinopathy in fundus images

Abstract

Diabetic Retinopathy (DR) is a severe eye disease that originates from diabetes newlinemellitus, and it mainly causes blindness in diabetic patients. DR can be suppressed or newlineits spreading slowed down at an initial stage by giving early treatment to patients. DR newlineis related to blindness, which takes place due to an incidence or an absence of newlineabnormal new vessel, such as Non-Proliferative Retinopathy and Proliferative newlineRetinopathy. DR is also caused due to the presence of hemorrhages. It is newlinecharacterized by the changes observed in the retina that includes lipid, and micro newlineaneurysms, and changes in the diameter portion of the blood vessels. newlineDR is a persistent evolution that develops certain threatening infections related to the newlineretina with expanded hyperglycaemia and further side effects that are associated with newlinediabetes mellitus, including hypertension. Ophthalmologists use Digital fundus images newlineto diagnose DR. To prevent vision loss, proper and early diabetic retinopathy detection newlinethrough regular scanning is very important. In this research work a Haar and an average filter is preferred to remove the noise newlinefrom fundus images and hybrid segmentation algorithm is offered to segment the newlinedifferent stages from its fundus image. A new feature extraction method is processed newlineseparately using Scale Invariant Feature Transform (SIFT) technique, and classified newlineusing neural classification network along with Neovascularization Elsewhere (NVE) newlineand Neovascularization of the Disc (NVD) parameter for PDR type classification. newlineFeatures are calculated from each binary vessel map to produce key point sets. The newlinesystem then combines these individual classification key points to produce a final newlinedecision. The proposed work, using different datasets of various images, achieves a newlinegood accuracy. newline