Purchase this article with an account.
Kornelia Lenke Laurik, Boglarka Eniko Varga, Fanni Pálya, Erika Tatrai, Janos Nemeth, Joachim Hornegger, Attila Budai, Gabor Mark Somfai; The Assessment of Diabetic Retinopathy using Retinal Vessel Segmentation. Invest. Ophthalmol. Vis. Sci. 2014;55(13):230.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Diabetes mellitus is the leading cause of avoidable blindness amongst working-age adults in developed countries. The growing number of diabetic patients increases the urge for telemedical screening programmes, which may be further facilitated by automated tools for diabetic retinopathy (DR) classification based on retinal vascular changes. Our purpose was to determine the parameters that most accurately describe the differences of the diabetic and healthy retinal vasculature by using manual segmentation of fundus images.
7 macula centered fundus photographs of 7 eyes of 7 diabetic patients with mild nonproliferative diabetic retinopathy were randomly exported from the database of Semmelweis University’s Diabetic Retinopathy Reading Center. Additionally, 7 macula centered fundus photographs were also exported of healthy subjects with no diabetes. The vascular tree was manually segmented using binary vessel images, density and distance maps were generated and vessel thickness data were calculated. Statistical evaluation was implemented by Mann-Whitney U-test (Statistica 8.0, Statsoft Inc.,Tulsa, OK).
The vessel density and vessel thickness values were significantly lower in eyes with DR compared to controls (15,62±1,48 vs 14,18±1,32, 5,59±0,51 vs 4,77±0,49, Control vs. DR, mean ± SD, p=0.03 and p=0.01, respectively). However, vessel distance showed no diference between the two groups (114,57±3,86 vs 111,25±2,17, Control vs. DR).
We observed DR-related retinal vascular changes which could be used for the discrimination between diabetic and healthy eyes. Our results showed differences in the vessel density and thickness parameters, supporting their possible use in the development of automated DR classification methods.
This PDF is available to Subscribers Only