Purchase this article with an account.
Elizabeth Marlow, Margaret McGlynn, Nathan Radcliffe; A novel alignment and subtraction technique for the detection of glaucoma progression. Invest. Ophthalmol. Vis. Sci. 2013;54(15):4829.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Identification of features of glaucoma progression informs clinical management. While flicker chronoscopy is useful, a dynamic view is not always possible. Auto-alignment and subtraction of serial optic nerve photos (ASSOP) was used to generate static representations of differences between baseline and follow-up images that may be valuable for identifying glaucomatous change.
Subtraction maps were generated from auto-aligned (EyeIC, Narbeth, PA) baseline and follow up images using Adobe Photoshop software. They demonstrate progressive retinal nerve fiber layer (RNFL) defects, optic disc hemorrhage (DH), neuroretinal rim loss (RL), and peripapillary atrophy (PPA). A masked glaucoma specialist identified features of progression on subtraction map first, then assessed feature strength by comparison with original images using flicker. Control images with no progression and parallax-only images were included.
Seventy-six eyes of 71 patients were used to generate subtraction maps that detected glaucoma progression in 87% of DH (n=31, Sensitivity (Se) 84%, Specificity (Sp) 98%) and 84% of PPA (n=32, Se 81%, Sp 98%) cases. The lowest rates of detection were seen with RL (67%, n=31, Se 65%, Sp 98%). The subtraction technique was most sensitive for detecting parallax (41 of 40, Se 98%, Sp 94%). All features of glaucoma progression appeared equally strong in flicker and subtraction images. However, parallax was stronger in 13% of subtraction maps overall and in 50% of images containing RNFL defects in particular. Among control images selected for absence of features of glaucomatous change (n=11) in original flicker images, no features were detected on subtraction maps.
ASSOP reliably detects features of glaucoma progression with a single static image. Parallax identification may also be facilitated. Future investigations should determine an algorithm that can be uniformly applied to any set of temporally-spaced and aligned images to create a subtraction map yielding the highest quality visualization of change.
This PDF is available to Subscribers Only