Purchase this article with an account.
Jaya Sadda, Nizar Saleh Abdelfattah, Srinivas R. Sadda, Zhihong Hu; Inter-Grader Repeatability of Geographic Atrophy Measurements from Infrared Reflectance Images. Invest. Ophthalmol. Vis. Sci. 2018;59(9):3245.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Geographic atrophy (GA) is commonly measured on fundus autofluorescence (FAF) images as an outcome measure in clinical trials. Companion infrared reflectance (IR) images are often used to assess for foveal involvement by GA, and are more comfortable than FAF images. The reliability of IR images alone for GA measurement has not been well-studied and is the subject of this analysis.
In this retrospective analysis, 88 confocal IR images (Heidelberg HRA+Spectralis) from 88 patients/eyes with GA were collected from existing anonymized reading center datasets. Two masked reading center graders (JS and NS) independently manually segmented the GA lesions on each IR image using GNU Image Manipulation Program (GIMP, ver 2.8.22) software. GA on IR images was defined in accordance to recently published CAM criteria as sharply-demarcated hyper-reflective regions at least 250 microns in diameter. The agreement in GA area measurements between graders were assessed by mean difference, intra-class correlation coefficients (ICC), and Bland-Altman plots. In addition, to better assess the point-to-point correspondence between gradings, overlap ratio (OR) and Dice similarity coefficient (DICE) were also computed.
Figure 1 shows a typical case with the grading result for the two graders. Among the 88 eyes with GA, mean area was 7.80 ± 6.01 mm2 for Grader 1 and 7.47 ± 6.25 mm2 for Grader 2, with a mean difference of 0.33 ± 1.65 mm2 and an ICC = 0.98 with a 95% confidence interval from 0.97 to 0.99 (F(87,87) = 54.15, p<0.001). A Bland-Altman plot illustrating the level of agreement is shown in Figure 2. The OR between graders was 0.83 ± 0.22 and the DICE coefficient was 0.88 ± 0.20.
GA lesions can be quantified reliably using IR images. As IR imaging is comfortable for patients and is commonly obtained along with OCT images, IR-based GA assessment may be a useful tool in clinical research studies and trials.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
Figure 2. Bland-Altman Plot of the difference between graders versus the mean of the two graders measurements of the area of GA on IR images.
This PDF is available to Subscribers Only