Quantitative clinical marker extraction from colour fundus images for non-proliferative Diabetic Retinopathy grading

摘要

This paper proposes robust methods for segmentation of clinically significant features of fundus images, from the point of view of detection and gradation of Diabetic Retinopathy as well as Maculopathy. After pre-processing to remove intra- and inter- image illumination variances, the optic disk and fovea are detected and exudates, haemorrhages and microaneurysms are segmented out using modified morphological techniques. The retina is then divided into circular zones concentric around fovea and the pattern and extent of abnormalities in these zones is used to classify the abnormal images into different grades of non-proliferative diabetic retinopathy (NPDR) and maculopathy.