Photographic patterns in macular images: representation by a mathematical model

Abstract

Normal macular photographic patterns are geometrically described and mathematically modeled. Forty normal color fundus photographs were digitized. The green channel gray-level data were filtered and contrast enhanced, then analyzed for concentricity, convexity, and radial resolution. The foveal data for five images were fit with elliptic quadratic polynomials in two zones: a central ellipse and a surrounding annulus. The ability of the model to reconstruct the entire foveal data from selected pixel values was tested. The gray-level patterns were nested sets of concentric ellipses. Gray levels increased radially, with retinal vessels changing the patterns to star shaped in the peripheral fovea. The elliptic polynomial model could fit a high-resolution green channel foveal image with mean absolute errors of 6.1% of the gray-level range. Foveal images were reconstructed from small numbers of selected pixel values with mean errors of 7.2%. Digital analysis of normal fundus photographs shows finely resolved concentric elliptical foveal and star-shaped parafoveal patterns, which are consistent with anatomical structures. A two-zone elliptic quadratic polynomial model can approximate foveal data, and can also reconstruct it from small subsets, allowing improved macular image analysis.