TOWARDS A HYBRID STATISTICAL FEATURE EXTRACTION AND HIERARCHICAL CLASSIFICATION MODEL FOR DIABETIC RETINOPATHY DIAGNOSIS

Abstract

Diabetic retinopathy (DR) is one of the leading causes of blindness worldwide. It is a disease that is caused by diabetes which affects the retina. Early detection of the disease can prevent blindness but it is affected by few or lack of visible symptoms in its early stage. The application of digital image processing, machine learning and pattern recognition techniques has provided fast, cost effective, accurately and automated screening of the disease using fundus images which solves the problems of manual screening. However, automated screening of diabetic retinopathy using fundus images are generally affected by poor fundus image quality and high correlation of the in-between DR grade fundus image statistical features which affects the performances of classifiers. We propose an improved hybrid statistical feature extraction approach using first order and second order gray level co-occurrence matrix (GLCM) and hierarchical classification model using artificial neural network (ANN) for diabetic retinopathy screening. The implementation success will minimize correlation effect and improve classifier performance, enable fast, effective, accurate, automated and convenient means of diagnosing diabetic retinopathy.