Abstract
Abstract: :
Purpose: We developed a digital technique, called the "dots" method, to reconstruct the background reflectance of a color fundus image for the purpose of segmenting drusen. This technique has been validated to a gold standard of manual digital tracings using stereo pairs for accuracy and reproducibility. In this study, we assessed the performance of this system at two independent institutions to demonstrate portability and ease of use. Methods:A digital system utilizing Adobe Photoshop (Adobe Systems Inc., San Jose, CA) and Matlab 6.5 (Mathworks Inc., Natick, MA) was used to reconstruct and level the macular background. Ten digital high–resolution images with either soft or hard drusen were selected by an expert grader in King’s College Hospital of London. Two graders, one from the Digital Fundus Photo Reading Center of Columbia University and one from King’s College Hospital, used this technique to quantify the percentage of drusen in the 3000 µm subfield. All correspondences and file exchanges were either performed electronically through a network or by phone. Results: The differences between the gradings from each institution and their 95% limits of agreement were calculated. The mean absolute difference of the ten gradings was 2.34 ± 1.52% and the 95% limits of agreement were –5.25% to 6.00%. Agreement could be improved by dividing the region into a smaller 1000 um central zone and a middle annulus.No significant difference was found between the two institutions on segmenting hard or soft drusen. Conclusions:Two institutions independently used the same sophisticated software to achieve comparable drusen measurements in the 3000 um subfield. Limitations included a tendency to include retinal pigment epithelial atrophy as drusen. Such software portability is a prerequisite to establishing digital image analysis standards.
Keywords: drusen • imaging/image analysis: clinical • age–related macular degeneration