Purchase this article with an account.
D.G. Hafford, J.L. Reimers, R.P. Danis, Jr., M.D. Davis, T. Harding, B.A. Esser, L.–Y. Lee, B. Zhang, R. Susman; Diabetic Retinopathy Severity Assessment From Digital Images Compared to Film . Invest. Ophthalmol. Vis. Sci. 2006;47(13):986.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To compare gradings of diabetic retinopathy severity according to the ETDRS Diabetic Retinopathy Severity Scale in scanned film images displayed on an LCD monitor to standard grading of film images.
Modified 7–standard field, stereoscopic, color film photographs of 39 eyes were scanned as TIFF images at a resolution of 1319 x 1252 and 100 pixels/mm (approximately 1.65 megapixels). These images were then displayed at 72 ppi from compressed JPEG images on a 20.1" LCD monitor. Each of 8 graders thoroughly familiar with the definitions of the retinopathy severity levels performed a single evaluation of each eye using a hand–held stereo viewer and limited digital tools to determine ETDRS diabetic retinopathy severity level directly, after examining all 7 fields. The results were then compared to the original film grading (duplicate, field–by–field gradings with the level determined by a software algorithm and adjudication of major disagreements). Percent agreement and kappa statistics (unweighted and weighted as follows: 1.0 for agreement, 0.75 for 1 step disagreement and zero for all other disagreements.) were calculated.
Percent agreement within 1 step was ≥ 90% and weighted kappa was ≥ 0.78 for only 2 graders. Average values were 83% and 0.72, respectively, compared with 91% and 0.78 for film vs. film grading in a group of 91 different, but representative eyes. Of 110 disagreements, in 103 the digital grade was lower. The primary areas of disagreement indicate undergrading of microaneurysms, IRMAs, and subtle neovascular complexes in the digital grading.
Microaneurisms, IRMA, and neovascular complexes were missed upon the single grading of digital images compared to multistep evaluation of film. When grading scanned images of this resolution, there is a concerning loss of sensitivity. Further investigation of resolution, compression, and display parameters is necessary to optimize grading sensitivity. These results cannot be extrapolated to grading of de novo captured digital images of comparable resolution where standardization of color spacing, brightness, and contrast between camera systems introduces additional variables to consider.
This PDF is available to Subscribers Only