Open AccessAutomated Approach for Extraction of Retinal Blood Vessels
Author(s) -
Dheyaa M. Abdulsahib,
Hussain F. Jaafar
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/978/1/012037
Subject(s) - artificial intelligence , computer science , thresholding , computer vision , fundus (uterus) , contrast (vision) , retinal , segmentation , adaptive histogram equalization , human eye , histogram , image processing , image segmentation , pixel , ophthalmology , image (mathematics) , histogram equalization , medicine
In the field of ophthalmology, retinal image analysis is crucial to extract many details that help doctors identifying retinal diseases at early stages, such details are the blood vessels, optic disk, and fovea. The exponential increase in number of diabetic retinopathy patients necessitates the use of computers to help doctors to diagnose and treat retinal diseases of the human eyes. Computer vision effectively helps doctors by analyzing and treating human retinas. In the field of ophthalmology, a color image (RGB image) of the eye fundus is captured by ophthalmoscopy. In this work, an automatic technique for extraction of human retinal blood vessels is proposed. The proposed method is based on three main stages, namely image pre-processing, initial segmentation of blood vessels and image post-processing. In the first stage, the contrast-limited adaptive histogram equalization is applied to the green component image to improve its contrast, and then it is remerged again with the red and blue components. In the second stage, the blood vessels are extracted using mean-C thresholding. Finally, in the third stage, many morphological operations are used to refine the segmented blood vessels image. The proposed method is validated using the expert ground truths with the DRIVE dataset in terms of pixels, and experimental results show sensitivity, specificity, accuracy and positive predictive value of 0.770816, 0.977575, 0.959993 and 0.7611835, respectively. The performance measures are compared with many recent related works and found to outperform most of them. The superior performance of this method proves that it is promising for mass screening of human fundus images.