Purpose:
Autofluorescence (AF) images with hypofluorescenceindicate geographic atrophy (GA) of the retinal pigment epithelium(RPE) in age–related macular degeneration (AMD). We adaptedthe image analysis technique of level set segmentation to provideaccurate segmentation of GA in AF images.
Methods:
Seven patients with GA had serial AF imaging over aperiod of 2 to 3 years. Two serial images from one eye of eachpatient were selected for analysis (14 images). The clinicallyapparent areas of hypofluorescence were drawn in Photoshop byan expert grader (RTS, IB, or MB). Level sets, a deformablemodels technique established in cranial CT and MRI, were appliedby one expert (NL). Images were pre–processed using adaptivehistogram equalization and edge preserving smoothing to improvesegmentation results. Adaptive contrast enhancement was chosento reduce amplification of noise in homogenous areas and tocompensate for nonhomogenous background illumination. Afterseed points in each apparent atrophic region were selected,the algorithm calculated segmentations automatically. Automaticmorphological post–processing removed over segmentationresulting from retinal vessels and noise.
Results:
The level set segmentations were compared to the expertgradings on a pixel–by–pixel basis. The mean sensitivityand specificity were 0.89 and 0.98.
Conclusions:
The relatively large and homogeneous areas of hypofluorescencein AF images of GA in AMD are quite amenable to accurate levelset image analysis. Extension to segmentation of atrophic areasin AF images of other retinal disorders seems likely.
Keywords: age-related macular degeneration