Instance-Based Learning for Blood Vessel Segmentation in Retinal Images

Abstract

© 2020, Springer Nature Switzerland AG. Diabetic retinopathy is a fatal disease that affects the majority of those who have diabetes for a period of time. It could lead to a blindness in the end. Therefore, it is important to detect the diabetic retinopathy in the early stage, in order to prevent the blindness. One of the key indicators of the disease is the abnormality of blood vessels in the retina. This research paper is thus to propose the technique to automatically segment blood vessels in retinal images, which could be used further for the disease analysis. It begins with using the color transfer approach to normalize the color statistics of all input images based on the reference image in the lab color space. Then, the magenta channel is extracted and used for the best distinct of blood vessel structure from the background. The morphological operators and binarization process are applied here for segmenting the blood vessels with the noise reduction using CLAHE or contrast limited adaptive histogram equalization. The proposed method is validated using the published dataset, namely STARE. The proposed method achieves the promising sensitivity and specificity.