Abstract
Abstract: :
Purpose:To validate digital techniques for grading color fundus photographs in a large multicenter clinical study of age related macular degeneration (AMD). Methods: Twenty stereo pairs from nine patients enrolled in the Columbia Macular Genetics Study were selected for validation of digital analysis. The study grading is a modification of the International Grading System. Nine representative grading categories (drusen morphology, size and number, pigmentary changes, etc) were selected for comparison. Each pair was first evaluated in stereo by two independent readers. In cases of disagreement between readers, gradings were adjudicated by a retinal specialist. The digital analysis was performed by the first author, selecting one slide from each stereo pair. The digital methods included parafoveal background leveling, luteal pigment correction, and graphic tablet tracing. Results:The raw concordance of the two initial clinical gradings was 83.4%. After resolution of disagreements where possible, the concordance of the final clinical grading with the digital method was 80.3% There was no agreement among the three clinical readers on 4.1% of the gradings. Of the 32 total disagreements between digital and stereo grading, there were 20 in drusen categories and 12 other, roughly proportional to the number of categories graded. Conclusion:The concordance of digital grading of contrast enhanced fundus photos with stereo grading in AMD is comparable to the concordance of independent stereo gradings by two readers. Advantages include ease of identification of drusen and pigment variations, which our humans preferred over the fatigue of light box stereo grading, but there was a tendency to overcall drusen or confuse them with pigment atrophy without human oversight.
Keywords: 308 age-related macular degeneration • 430 imaging/image analysis: clinical • 391 drusen