An efficient method of detecting exudates in diabetic retinopathy: Using texture edge features

Abstract

Ophthalmologists analyze fundus images of eye extensively as a non invasive diagnosis tool for various internal eye defects. Diabetic retinopathy is an eye complication specially seen in diabetic patients, causing damage to retina which may lead to blindness. The major symptoms of this disorder is the presence of exudates, a pus like fluid oozed from damaged blood vessels due to high blood sugar. This hardens on the retina of patient, leading to blindness. In this paper, we propose a methodology for automatic detection of exudates. We remove the non exudates like optic disc, blood vessels, and blood clots in two phases using Gradient Vector Flow Snake algorithm and region growing segmentation algorithm. This improves efficiency of detection by masking false exudates. Then, we detect exudates using Gabor filter texture edge detection based segmentation algorithm. To reduce computational complexity, only Gabor filters tuned to two higher frequencies and four orientations are used. We have implemented the proposed methodology on 850 test images. We have obtained a high efficiency of 87% true exudates.