Abstract
Purpose :
A wide field-of-view (FOV) is desirable in retinal imaging for managing peripheral retinal disease and speeding up clinical workflow; however, any wide-angle view presented on a flat screen is a projection and thus exhibits distortion. Widefield imaging systems typically produce “fisheye” distortion, where the apparent shape of retinal features varies depending on their position within the imaged field; this may make interpretation of the image more difficult. With knowledge of the optics, however, it is possible to numerically remap the images to any standard geometric projection. In this retrospective study, we present different methods of re-projecting wide FOV retinal images. We then quantify their respective impact on distortion by examining shape properties of the optic nerve head (ONH).
Methods :
For each of 10 eyes, a pair of widefield color fundus images (CLARUSTM 500, ZEISS, Dublin, CA) were assembled. The patient’s fixation was steered between captures such that the ONH presented peripherally in the first image, and centrally in the second. The ONH was manually segmented in each case. The inherent fisheye distortion of the optical system was computationally remapped to each of four standard geometric projections: equidistant, equal-area, stereographic, and rectilinear. The eccentricity of the segmented ONH was calculated for each projection type.
Results :
Figure 1 shows a comparison of each map projection type. Figure 2 shows the correlation coefficient, ρ, between the central and peripheral ONH eccentricity across subjects for each projection type.
Conclusions :
Among the tested map projections, stereographic projection rendered the most consistent ONH shape between captures with different eye orientations (ρ=0.71). It may be advisable for widefield retinal images to be remapped to a stereographic (or other conformal) projection, as the property of rendering features with a shape independent of eye orientation may aid in interpretation and analysis across repeat examinations.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.