Abstract
Abstract: :
Purpose:Current manual methods of drusen segmentation and quantification are laborious and subjective. 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, and tested its accuracy using patient data. Methods:Color fundus photographs of fourteen patients with significant soft drusen were selected for this study. A digital technique utilizing Adobe Photoshop (Adobe Systems Inc., San Jose, CA) and Matlab 6.5 (Mathworks Inc., Natwick, MA) was used to reconstruct the macular background. Manual selection of dots 25 µm square on normal background between the drusen were used to fit two quadratic polynomials that modeled the central subfield (1000 µm diameter around the central fovea) and the middle annulus (between 1000 µm and 3000 µm diameters). This created a two–zone geometric reflectance model, which was subtracted from the green channel of the image and accurately leveled the background reflectance. Global thresholds in the two zones were chosen to segment drusen. To assess the accuracy of drusen segmentation, measurements of drusen areas as percentage of the 3000 µm diameter field by the dots method were compared to our previously validated semi–automated method (R.T. Smith et al., BioMed. Eng. OnLine. 2:10. 2003.) and a gold standard of manual tracing from stereo pairs. Inter– and intra–observer reproducibility were also determined for the dots method. Results:Measurements obtained by the dots method were compared to the gold standard method and to our semi–automated method to yield 95% limits of agreement of –8.3% to 2.8% and –7.1% to 4.8%, respectively. Intra–observer reproducibility was demonstrated by standard deviations for two trials ranging from 0.1 to 4.1%. Inter–observer reproducibility evaluated by two investigators yielded 95% limits of agreement between –2.7% and 6.3%. The two manual steps of selecting dots and threshold each took about two minutes of user time. Conclusions:The method of reconstructing the macular background reflectance by the mathematical model yielded digital drusen measurements that were reproducible, and performed accurately with respect to both our semi–automated method and the gold standard. The total operator time was comparable to that of standard fundus photo grading. With the proper implementation of commercially available software, this technique for macular image analysis has potential for use in clinical research.
Keywords: age–related macular degeneration • imaging/image analysis: clinical • drusen