Non-Compactness Attribute Filtering to Extract Retinal Blood Vessels in Fundus Images

Non-Compactness Attribute Filtering to Extract Retinal Blood Vessels in Fundus Images

I. K. E. Purnama, K. Y. E. Aryanto, M. H. F. Wilkinson
Copyright: © 2010 |Pages: 12
DOI: 10.4018/jehmc.2010070102
(Individual Articles)
No Current Special Offers


Retinal blood vessels can give information about abnormalities or disease by examining its pathological changes. One abnormality is diabetic retinopathy, characterized by a disorder of retinal blood vessels resulting from diabetes mellitus. Currently, diabetic retinopathy is one of the major causes of human vision abnormalities and blindness. Hence, early detection can lead to proper treatment, and segmentation of the abnormality provides a map of retinal vessels that can facilitate the assessment of the characteristics of these vessels. In this paper, the authors propose a new method, consisting of a sequence of procedures, to segment blood vessels in a retinal image. In the method, attribute filtering with a so-called Max-Tree is used to represent the image based on its gray value. The filtering process is done using the branches filtering approach in which the tree branches are selected based on the non-compactness of the nodes. The selection is started from the leaves. This experiment was performed on 40 retinal images, and utilized the manual segmentation created by an observer to validate the results. The proposed method can deliver an average accuracy of 94.21%.
Article Preview

Materials And Methods

Retinal Image Source

The proposed method are tested and evaluated on DRIVE database that is publicly available and consists of 40 colored retinal images. The DRIVE that stands for Digital Retinal Images for Vessel Extraction (DRIVE) was established by Staal et al. (2004). Each image of DRIVE has resolution of 565x584 pixels, and stored in GIF format.

The images of DRIVE database consists of 20 training images and 20 test images. One benefit of using DRIVE is the available reference images resulted from the manual segmentation procedure. The reference images are required to calculate the accuracy of the proposed method. Each training image has one reference image that was created by an observer, while each test image has two reference images created by two different observers. Our experiment is not related with the training procedure. Hence, we used 40 reference images created by the first observer. As it is mentioned in their web site, the observers were instructed and trained by an experienced ophthalmologist. They were asked to mark all pixels of the expected vessel. Figure 1a shows one of the retinal images of the test images, while Figure 1b, and Figure 1c, respectively, show the corresponding images created by first observer and the second observer.

Figure 1.

(a) An image of the test images of DRIVE database, (b) and (c) the corresponding images of the image in (a) created by the first observer and the second observer


Complete Article List

Search this Journal:
Volume 15: 1 Issue (2024): Forthcoming, Available for Pre-Order
Volume 14: 1 Issue (2023)
Volume 13: 5 Issues (2022): 4 Released, 1 Forthcoming
Volume 12: 6 Issues (2021)
Volume 11: 4 Issues (2020)
Volume 10: 4 Issues (2019)
Volume 9: 4 Issues (2018)
Volume 8: 4 Issues (2017)
Volume 7: 4 Issues (2016)
Volume 6: 4 Issues (2015)
Volume 5: 4 Issues (2014)
Volume 4: 4 Issues (2013)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing