Feature extraction from Fundus images to Analyse Diabetic Retinopathy

Abstract

Diabetic Retinopathy is an eye disease and a common complication of diabetes that can cause vision loss if left undiagnosed at an initial stage. It is the prime cause of blindness in the working age population of the world. Colour fundus images are used by ophthalmologists to study eye diseases like diabetic retinopathy. Early detection of diabetic retinopathy through regular screening is particularly important to prevent vision loss. However, with a large number of patients undergoing regular screenings, more amount of time is needed for ophthalmologists to analyse and diagnose the fundus images. In India, there are not enough resources, in terms of time and available expert ophthalmologists, for carrying on an extensive screening. Thus, a reliable automatic tool for diagnosis of diabetic retinopathy is strongly needed. Any automatic tool for diagnosis of diabetic retinopathy must go through some well defined steps. First, it has to detect the major anatomical structures of the retina viz blood vessels, optic disc and fovea. Second, it has to identify abnormalities in the retina like hard exudates, cottonwool spots, hemorrhages and microaneurysms that cause diabetic retinopathy. This thesis mainly focuses on developing a Fundus Image Analysis system that extracts the anatomical and abnormal features of the retina in order to diagnose diabetic retinopathy. The research is carried out in six phases: In the first phase, Histogram Matched Local newlineRelative Entropy (HMLRE) method is developed to segment the vasculature of fundus images. This method uses the intensity information of red and green channels of the same fundus image to correct the non-uniform illumination in colour fundus images. Matched filtering is employed to improve the contrast of retinal blood vessels against the background. The enhanced retinal blood vessels are then segmented by using Local Relative Entropy based thresholding that can efficiently maintain the spatial structure of the vascular tree segments.