Purchase this article with an account.
H. K. Li, L. D. Hubbard, R. P. Danis, T. Harding, J. F. Florez-Arango, E. A. Krupinski; Digital vs. Film Stereo Color Retinal Images for Grading Diabetic Retinopathy Severity. Invest. Ophthalmol. Vis. Sci. 2008;49(13):2748.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To compare diabetic retinopathy (DR) severity levels graded from enhanced stereo color digital retinal images to those graded from slide film photographs.
Film (F) and digital (D) color images of 154 eyes of persons with diabetes (pre-selected for full range of DR) were taken with a Topcon 35o camera as 7 stereo fields [7SF] per DRS/ETDRS protocol. Three graders independently determined Early Treatment Diabetic Retinopathy Study(ETDRS) DR level from F (slide transparencies on light boxes with Donaldson viewers) and D (2392 x 2048 pixel, uncompressed images viewed on calibrated 20" LCD monitors with hand-held stereo viewers in Topcon IMAGEnet). Digital images were standardized for brightness, contrast and color balance according to the AREDS2 image model, yielding film-like consistency. In D, the green channel (similar to red-free) was also examined. DR levels were defined by central tendency among graders. Custom software controlled order/timing of grading to minimize bias and recall.
Graders classified DR from F images as follows: no DR = 26 eyes, non-proliferative DR (NPDR) - 97 eyes (microaneurysms only, mild/moderate/severe = 8/35/25/29), proliferative DR (PDR) = 30 eyes (mild/moderate/severe = 10/10/10), and ungradable DR = 1 eye. Comparison of D vs. F classifications for specific ETDRS level (9-step scale) yielded 66.9% exact agreement and 94.8% +1 step (unweighted Κ = 0.62, SEΚ = 0.04; linear weighted Κ = 0.83, SEΚ = 0.03). For historical comparison, agreement in ETDRS from single replicate gradings of F using the same scale (ETDRS report #12, Ophthalmology 1991) was lower: 53% exact and 88% +1 step (unweighted Κ = 0.42). In our study, variability observed between media was similar to that observed between graders within each medium.
Comparing D 7SF with F 7SF after standardized enhancement of D, our DR evaluations from D were similar to those from F. Unlike some previous reports using unenhanced digital images which showed lower sensitivity of D for levels defined by subtle abnormalities (microaneurysms, IRMA, new vessels), we found no evidence of systematic disadvantage for D compared to F over a broad range of DR severity.
This PDF is available to Subscribers Only