Purchase this article with an account.
Carolina Kunnen, Percy Lazon De La Jara, Brien A Holden, Eric B Papas; Automated assessment of lid margin lissamine green staining. Invest. Ophthalmol. Vis. Sci. 2014;55(13):1976. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop and validate an objective, automated method for the assessment of lissamine green (LG) staining of the lid wiper region and establish objective quantifiable parameters which correspond with the judgement by human observers.
Upper eyelid margins of 27 participants were photographed after staining with LG. Images were processed using custom designed software developed in MATLAB. After manual delineation of the lid margin, images were transformed into a HSV format (hue, saturation, value). Spurious reflections were automatically eliminated prior to contrast enhancement, thresholding and binarisation. An example of an original image together with the automatically binarised outcome is shown in Figure 1A and B. The algorithm automatically extracts the following variables: proportion of LG staining relative to area of the eyelid (PA), median intensity of staining (IRGB), median intensity of staining in red channel (IR), green channel (IG) and blue channel (IR), relative greenness staining (RGS), green-red difference (GRD), green-blue difference (GBD) and median intensity of hue value (IH). Repeatability was assessed by capturing and measuring two images of the same 27 eyelids. Validity was tested by comparing the automated results to assessment of the 27 images made by ten observers, using a 0-3 grading scale.
For PA, the mean difference between replicates was -0.06% with 95% limits of agreement being 1.07 and -1.13%. Correlation analysis showed significant associations (p < 0.05) between the mean grades by human observers and all parameters except for IRGB, GBD, IG and IB. PA showed the highest linear correlation with subjective grading (R2 = 0.63). The associations with the colorimetric parameters, GRD (R2 = 0.55), were less strong, but also showed a positive correlation with human assessment.
A semi-automatic system has been developed that permits objective, repeatable measures of lissamine green staining. Judgments of LG staining by human observers are not fully replicated by individual morphometric or colorimetric parameters.
This PDF is available to Subscribers Only