Purpose
To evaluate the accuracy of macular atrophy quantification by a semiautomated region finder technique in fundus autofluorescence (FAF) images of eyes with macular atrophy due to Stargardt disease.
Methods
Macular atrophy quantification was performed by masked, certified reading center graders using both a commercial semiautomated software (Heidelberg Spectralis Region Finder tool) and manual segmentation in 20 eyes of 12 subjects with Stargardt disease and available FAF imaging in this retrospective study. The Spectralis Region Finder tool facilitates semiautomated detection and quantification of dark areas in FAF images, termed definite decreased autofluorescence (DDAF). Seed points within candidate areas of atrophy are selected by the graders and the graders then adjust the threshold of the region growing algorithm until the entire area of atrophy is delineated. For manual delineation of atrophy, the borders of the atrophy were segmented using a previously described planimetric grading tool (GRADOR). The correlation between semiautomated and manual measurements was assessed by intraclass correlation coefficients (ICC).
Results
The mean (standard deviation) atrophy area was 6.15 (5.16) mm2 with manual grading and 5.87 (4.81) mm2 with the semiautomated region finder (their correlation is shown in Figure 1). The mean absolute difference was 0.77 (0.65) mm2, with greater apparent differences in eyes with larger atrophic lesions. The two measurements, however, were highly correlated with an intraclass correlation coefficient of 0.99 (95% CI: 0.98-1.00).
Conclusions
FAF-derived atrophy measurements in Stargardt disease obtained by semiautomated region finder techniques or manual delineation are highly correlated. The two measurements do not agree perfectly, however, particularly at larger areas of atrophy. The reproducibility of the two techniques requires further study.
Keywords: 550 imaging/image analysis: clinical •
494 degenerations/dystrophies