Premium
Semi‐automated retinal vessel analysis in nonmydriatic fundus photography
Author(s) -
Schuster Alexander KarlGeorg,
Fischer Joachim Ernst,
Vossmerbaeumer Urs
Publication year - 2014
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/aos.12242
Subject(s) - fundus photography , retinal , reproducibility , software , fundus (uterus) , computer science , medicine , ophthalmology , optometry , artificial intelligence , biomedical engineering , computer vision , mathematics , statistics , fluorescein angiography , programming language
Purpose Funduscopic assessment of the retinal vessels may be used to assess the health status of microcirculation and as a component in the evaluation of cardiovascular risk factors. Typically, the evaluation is restricted to morphological appreciation without strict quantification. Our purpose was to develop and validate a software tool for semi‐automated quantitative analysis of retinal vasculature in nonmydriatic fundus photography. Methods matlab software was used to develop a semi‐automated image recognition and analysis tool for the determination of the arterial–venous ( A / V ) ratio in the central vessel equivalent on 45° digital fundus photographs. Validity and reproducibility of the results were ascertained using nonmydriatic photographs of 50 eyes from 25 subjects recorded from a 3 DOCT device ( T opcon C orp.). Two hundred and thirty‐three eyes of 121 healthy subjects were evaluated to define normative values. Results A software tool was developed using image thresholds for vessel recognition and vessel width calculation in a semi‐automated three‐step procedure: vessel recognition on the photograph and artery/vein designation, width measurement and calculation of central retinal vessel equivalents. Mean vessel recognition rate was 78%, vessel class designation rate 75% and reproducibility between 0.78 and 0.91. Mean A / V ratio was 0.84. Application on a healthy norm cohort showed high congruence with prior published manual methods. Processing time per image was one minute. Conclusions Quantitative geometrical assessment of the retinal vasculature may be performed in a semi‐automated manner using dedicated software tools. Yielding reproducible numerical data within a short time leap, this may contribute additional value to mere morphological estimates in the clinical evaluation of fundus photographs.