Purchase this article with an account.
Jong-Mo Seo, Hyun Sung Nam, Hyun Kim, Jeong Min Hwang, Jung Kim, Young Min Lee, Hae Young Lee, Hyo Soo Kim; Automated Measurement Of Retinal Vessel Diameters On Digital Fundus Photographs By Focus Assessment And Fast Marching Method. Invest. Ophthalmol. Vis. Sci. 2011;52(14):2892.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
It is a challenging issue to measure the retinal vessel diameter on the fundus photograph for analyzing its correlation with systemic disease, because semi-automated or manual measurement shows low repeatability and reproducibility. We developed a novel, automated method to measure the retinal vessel diameter and tested its performance.
Fundus photograps were taken with Canon CR-1 nonmydriatic fundus camera attached with Canon EOS 50D digital camera, in 4752 x 3168 pixel JPEG format. All the photographs were evaluated by focus assesment with the normalized sum-modified-Laplacian (SML) value, and images with the normalized SML value more than the threshold were used for analysis. Optic disc was segmented from the red channel of the photograph, and the annulus between the concentric circles with the two and three times of the optic disc radius was used as an evaluation area. The direction and diameter of the vessels are obtained by using the Hessian matrix on the green channel image. The path of the vessels can be determined by the geodesic path using the fast marching method. All the analysis were done on a computer with Intel Core2Duo-P8600 CPU and 4GB RAM.
Experiments were performed on nine healthy volunteers. Six normal fundus photograps were taken for left/right eyes of four volunteers, and if the normalized SML value of the image was unsatisfactory, that image was discarded and another photo was taken. Vessels were well segmented and the diameter of the vessels could be automatically acquired by the proposed method. Calculation time was about 23 sec. per image. The average standard deviation value of all the vessels was 3.68%. For arteries, it was 4.07% and for veins, 4.46%.
Proposed method automatically detected and analyzed retinal vessels without intervention of user. Focus assesment improved the measurement of the retinal vessel diameters.
This PDF is available to Subscribers Only