Comparison of clinical and textural approach for diabetic retinopathy grading

Abstract

Diabetic Retinopathy (DR) is an eye disease due to complication of diabetes. Early screening and treatment of DR disease is crucial to prevent blindness. This paper presents a comparison of different approaches for automated DR grading. In the current development of automated system for DR grading, generally there are two approaches; clinical and textural. Clinical approach is based on the clinical study that refer to the abnormalities exist in retina such as microaneurysm, haemorhage, exudates and cotton wool spot. While textural approach is based on the visual or texture features of retina image. In this work we compare both approaches for performance analysis. Both approaches were applied to a standard dataset and the results show that both approaches capable to classify the severity of DR disease. This is based on the average classification accuracy obtained using support vector machine (SVM) with 82.74% and 83.87% for clinical and textural approach respectively. The pros and cons for each approach based on the experimental study also discussed in this paper.