Quantitative evaluation of drusen on photographs.

Visualizing and tracking drusen is important in managing age-related macular degeneration. Color fundus photography (CFP) is common in clinical practice, but quantitative evaluation of drusen in CFPs is difficult because of the varying background of the macula, retinal pigment epithelium (RPE), and choroid. Spectral-domain optical coherence tomography (OCT) provides high-resolution visualization of drusen, and 2-dimensional projections resembling CFPs can be produced from spectral-domain OCT using summed-voxel projection (SVP). Although SVP is useful for visualizing the fundus, most drusen are not visible this way. Anything that produces high reflections along axial lines produces bright areas in the SVP image and obscure drusen (Fig 1A, B, yellow circles; available at http://aaojournal.org). Thus, drusen that are visible on B-scans may not be visible on the SVP image (Fig 1B, C, red circles). Furthermore, drusen produce distortions in the RPE layer, but do not substantially change the intensity sums in the image columns traversing the drusen and remain invisible on SVP. The “slab SVP” in a commercial spectral-domain OCT system (Carl Zeiss Meditec, Inc., Dublin, CA; unpublished data, June 2011) and similar methods have improved the SVP; however, user interaction is required to localize the RPE, and the slab SVP provides no information about drusen thickness, which could be useful for characterizing drusen. We have developed a novel, fully automated approach, the restricted SVP (RSVP), to generate improved visualizations of the retinal surface that incorporates image processing to enhance drusen visualization. The RSVP has 2 steps: (1) restrict the projection to the subvolume of OCT data in close proximity to the RPE, and (2) supplement the projection with image processing methods that enhance the brightness of drusen. In the first step, the method detects and interpolates 2 contours in the RPE: (1) The abnormal contour of the upper boundary of the RPE which is distorted by drusen and (2) the expected normal (undistorted) lower contour that would be expected if drusen were absent. The posterior boundary of the normal RPE layer is taken as the baseline of the projection region used for RSVP generation, whereas the top boundary of the projection region is determined by displacing the normal RPE layer anteriorly the same distance as the largest drusen peak in the cube. All other structures in the retina that could interfere with visualization of drusen are excluded (Fig 2, available at http://aaojournal.org). In the second step, the method replaces the values of the pixels underneath distorted RPE with the maximum intensity value in each image column of the B-scan. (Fig 3, available at http://aaojournal.org). Because the projection is based on the sum of the pixel values in each image column, the greater the height of the drusen, the brighter the drusen in the projection; thus, in addition to enhancing drusen visualization, the method provides a visual indication of the thickness of drusen. Using 46 OCT datasets from 8 patients, we found that the RSVP method was more effective than conventional SVP in displaying drusen and retinal vessels (Fig 4, available at http://aaojournal.org), and was more accurate in detecting drusen in qualitative and quantitative evaluations. Qualitative evaluation was conducted by visual inspection of SVP and RSVP images, and comparing them with CFPs. Quantitative evaluation consisted of a comparison of drusen detected by manual segmentation of the SVP and RSVP images in 4 eyes from 3 patients. Gold standards were obtained by manual segmentation of each of the b-scans in the cubes. They were performed twice by 2 separate readers. The results showed that although, on average, only 2.1% of drusen could be detected from SVP images, 89.3% of drusen were detected in RSVP images. Our method differs from prior approaches in that it is fully automated and it includes a highly restricted volume of the OCT in proximity to the RPE where drusen reside; thus, drusen visualization with a RSVP are degraded less than in the current, slab-like SVP methods. Furthermore, our method incorporates image processing to enhance drusen. To date, the RSVP method is substantially more effective than current SVP methods for visualizing drusen, and it may be useful to ophthalmologists in rapidly and quantitatively assessing drusen in age-related macular degeneration patients.