A novel approach for blood vessel segmentation with exudate detection in diabetic retinopathy

Abstract

Non-Proliferative Diabetic Retinopathy (NPDR) sabotage the blood vessels of the human eye that causes visual impairment. This paper proposes a blood vessel segmentation method on retinal images. It is proceeding by tracing the exudates present in the retinal image. Taking care of exudates in the retinal image, it segments the blood vessels in the image. The segmentation is done, by using the Frangi filter with the enhancement of contrast, using morphological bottom hat transform in an integrated platform that uses the fuzzy C-Means clustering, and to handle the pathological cases it uses the Laplacian of Gaussian (LOG) filter also. The morphological hat transforms extracts the details of tiny elements from the image, hence it is used to intensify the line, edge, and the vasculature structure. The Frangi filter is one of the most advanced methods for vessel enhancement, which is then applied to extract the edge shaped vasculature structure from the retinal image. Then it is followed by fuzzy C-Means clustering to delineate the vessel structure from the retinal images. The overall performance is evaluated on the DRIVE, STARE dataset that is publicly available on the internet for research purposes in the field of DR.