Abstract
Purpose:
To evaluate the validity and reproducibility of a novel computer-based semi-automated software method for quantification of retinal arterial central reflex, a hypertensive retinopathy sign, from digital color fundus images using images taken from the Singapore Malay Eye Study (SiMES).
Methods:
Color fundus images were selected from a large population based study SiMES. A novel central reflex quantification method was developed which only requires a singular user input to select the vessel edge points. The software then automatically determined the width of caliber and central reflex of the vessel, and computed their ratio. Specialized software was developed that allows the user to select the vessel edge start points and edit the width of the vessel and central reflex to be measured. The validity was tested using Spearman rank correlation of the quantified central reflex (CR) assessed by the proposed method against an expert grader who graded CR qualitatively (mild, moderate and severe level). The reproducibility was evaluated in two ways: 1) inter-observer variation between two graders who graded the same images; 2) intra-observer variations, via grading of two images of the same eye taken at different times by the same grader. Careful attention was given to measure at exactly the same vessel locations by the graders in the images. Reproducibility of these measurements was determined by Pearson’s correlation.
Results:
Two hundred and twelve retinal images of normal subjects were randomly selected by retinal image graders and CR grading on the largest arteries undertaken. The validity assessment showed that Spearman rank correlation (ρ) between the software measured values and expert graders’ ranking on CR severity was 0.83 (p<0.01). High inter-observer reproducibility was indicated: Pearson’scorrelation 0.81 for CR to arterial caliber ratio. Similarly, high intra-observer reproducibility was achieved. The Pearson’s correlation on this is 0.83 for CR caliber and 0.87 for the ratio of CR to arterial caliber.
Conclusions:
A novel semi-automatic measurement system for retinal vessel central reflex caliber quantification was demonstrated to be valid when against the grading by an experienced grader, and have high inter- and intra-observer reproducibility.
Keywords: 551 imaging/image analysis: non-clinical •
688 retina •
549 image processing