AUTOMATIC LESIONS DETECTION AND CLASSIFICATION OF DIABETIC RETINOPATHY USING FUZZY LOGIC

Abstract

Diabetic Retinopathy (DR) is one of the visual abnormality of diabetes which affects the retina. The disease gradually damages the retina and leads to blindness in future. Detection and classification of DR at an early stage can reduce the severity of this vision loss significantly. In this paper, we propose a robust system which detects retinal lesions (blood vessels, microaneurysms and exudates) from retinal images and classifies the stages of DR automatically. At first, blood vessels, microaneurysms and exudates are detected by using image processing techniques. Next, blood vessels area, microaneurysms count, exudates area, contrast and homogeneity are measured from the processed images as the retinal features. These features are finally fed to knowledge based fuzzy classifier for classifying normal, mild NPDR, moderate NPDR, severe NPDR and PDR stages. A total of 400 retinal fundus images are collected from STARE, DIARETDB0 and DIARETDB1 databases and the images are successfully classified by the fuzzy classifier with accuracy up to 95.63%.

Existing System

Deep convolutional neural networks (DCNN) approach

Proposed System

The novelty of our proposed system lies in developing Fuzzy classifier for classifying the stages of DR. First, the retinal fundus image was preprocessed to remove the background noises and get a smooth normalized image. Then blood vessels, MAs, and EXs are detected and calculated the area of each candidate lesion, contrast, and homogeneity as features. Finally, the features are classified by the fuzzy classifier to identify normal, mild NPDR, moderate NPDR, severe NPDR and PDR stages.

Architecture

BLOCK DIAGRAM