Drusen ultrastructure can be imaged with SDOCT and reliably characterized by readers viewing unprocessed high-resolution scans. The high resolution and lack of motion artifact in SDOCT scans allow the assessment of multiple morphologic parameters and thus makes possible more precise characterization of drusen than previous imaging modalities. This grading system will be useful for longitudinal clinical trials to correlate these intradrusen characteristics with genotype, severity of disease, and risk of progression in AMD.
With this grading system, we were able to identify unique drusen tomographic findings and validate previous ones. Our observation that most of the drusen graded as calcified photographically were of low predominant internal reflectivity tomographically is consistent with clinicopathologic OCT studies that observed calcific sites in atherosclerosis17
and with clinical studies of calcific ocular lesions imaged with conventional OCT.18
As did Pieroni and colleagues,6
who used a time-domain system, we observed a “saw-toothed” pattern of RPE elevation in a majority of eyes (). Likewise, we found that some drusen have hyper-reflective foci overlying them (, available at http://aaojournal.org
). It has been suggested that these hyper-reflective foci represent RPE cell migration,6
although in this study only 4 of 13 sites (31%) showed any hyperpigmentation on the color photograph. Histologic correlation would aid in sorting this out. In addition, the patterns of “RPE excrescences overlying moderately reflective material” (, available at http://aaojournal.org
) and “disruptions of the RPE” (, available at http://aaojournal.org
), reported by Pieroni et al,6
were clearly resolved with this imaging system. However, with the SDOCT system, we were able to specify the relative degree of internal reflectivity and assess homogeneity to add greater precision to drusen assessment. We elucidated several patterns of drusen characteristics that have not been described in earlier studies. For example, we found convex nonhomogeneous drusen that had predominantly low internal reflectivity (, druse number 5). Also, we were able to demonstrate convex-shaped drusen with predominantly high internal reflectivity, nonhomogeneous, and without overlying foci of hyper-reflectivity (, available at http://aaojournal.org
). Finally, with high-resolution SDOCT imaging we were able to identify 5 eyes with drusen that had hyper-reflective “cores” within them ( [available at http://aaojournal.org
]). This finding has not been reported with OCT drusen imaging and may relate to drusen structures associated with complement activity as seen in ex vivum studies.2,3
The non-core, nonhomogeneous reflectivity patterns within drusen may reflect distribution of other components such as amyloid β
OCT imaging is based on narrow angle reflectance from tissue, and this modality has been used to differentiate fibrous, fibrocalcific, and lipid-rich sites within atherosclerotic plaques and to identify cellular processes and alignment in engineered tissue.17,19
Until we have clinicopathologic correlation of SDOCT imaged drusen, we will not know which specific components are responsible for these SDOCT image characteristics.
Although several drusen with a similar photographic appearance had a similar tomographic appearance (), there was also great tomographic variability in many photographically similar drusen (, , and ). Although most of the soft indistinct drusen were convex with medium internal reflectivity, homogeneous, and without overlying foci when imaged with SDOCT, making these the most common photographic and tomographic drusen types seen in our study, it is important to remember that 50% of the soft-indistinct drusen have a different tomographic appearance. These sites of different tomographic appearance may be precursors of change; whether of drusen growth, clearing, choroidal neovascularization, or atrophy. Sites of unique drusen reflectivity may predict focal change at that locus, or the presence of these drusen may predict risk status for the eye or the patient overall. These questions will be examined in the upcoming 5-year Age-related Eye Disease Study 2 Ancillary SDOCT Study.
There was considerable interobserver and intraobserver agreement for each of the morphologic parameters, ranging from agreement on drusen reflectivity and shape, which was generally more than 90%, to drusen homogeneity, which had the lowest (albeit still >75%) agreement. Overall, the level of interobserver agreement was high. When compared with other systems that graded drusen type in AMD using color photography, our results are favorable.4,5
In the Wisconsin age-related maculopathy grading system, interob-server agreement for drusen type was 70.6% and intraob-server agreement was 62.5%.4
The AREDS Research Group found the contemporaneous reproducibility of drusen type to be 77.5% (exact agreement) and the temporal reproducibility of drusen type to be 59.0% (exact agreement).5
We would have predicted better interobserver agreement in summed (processed) scans and less intraobserver agreement when comparing non-summed (unprocessed) to summed scans because of the decrease in noise (hyper-reflective “speckle”) in summed scans (). This is important because acquiring multiple scans at 1 site to sum the data and remove speckle noise decreases the overall number of different retinal sites that can be sampled during the same time interval. However, the interobserver agreement appeared to be higher with the non-summed scans than the summed scan even for the overlying foci of hyper-reflectivity, a morphologic characteristic that potentially could be hidden in speckle noise. Although the reason for this outcome is unclear, the result is reassuring at least. The majority of commercial devices that currently perform spectral domain OCT do not implement techniques to reduce noise during volumetric scan acquisition. However, the clinical utility of this grading system will be maximized when acquiring volumetric scans of patients’ maculae (a possibility with SDOCT), not with single linear scans through the fovea (the only possibility with conventional time-domain OCT). Thus, our grading system should be reliable and reproducible in clinical practice regardless of the SDOCT device being used and regardless of whether post-acquisition noise-reducing image processing occurs.
One limitation of our study is small sample size. Although we successfully identified 17 different drusen types (using combinations of the 4 above parameters) in our 21 eyes and 120 total drusen, based on our grading methodology, a total of 38 different drusen patterns could mathematically exist (concave or convex  × degree of internal reflectivity  × homogeneity  × overlying hyper-reflective foci presence  + saw-toothed with overlying hyper-reflective foci  + saw-toothed without overlying hyper-reflective foci = 36 + 1 + 1 + 38). We have no reason to believe that there are in fact actually 38 different drusen patterns that exist, but there may be more than the 17 we found. This large number of theoretic possibilities suggests that searching for phenotypic and genotypic links will be difficult. However, we think that it is important to first describe all the possible tomographic patterns before grouping them into larger, more functional categories, if warranted.
The goals of this study were to categorize the various drusen substructures with SDOCT and to correlate these tomographic appearances with photographic appearance, the current gold standard. Future studies should aim to correlate these structures with function, disease status, and genotype. This would be aided not only by correlation with appearance on fluorescein angiography or autofluorescence but also by histologic analysis. These correlation studies are under way. In addition, this study did not address total drusen area or volume or drusen size, because this is the focus of a separate semiautomated computer segmentation algorithm study that is reported separately.20
Future studies include refinement of computer algorithms to identify and characterize drusen with this grading system automatically in an effort to speed the process of drusen analysis, quantify these characteristics, and eliminate potential observer disagreement by standardizing the morphologic parameters. These further enhancements in drusen analysis would be used for longitudinal studies that could potentially produce prohibitively large amounts of data for manual analysis.
Likely future advances in pharmacologic therapy for non-neovascular AMD and the need for improved phenotyping for genetic studies will demand a level of precision in both clinical trials and the clinical setting that is not possible with color photography alone. The existence of diverse tomographic patterns of drusen is consistent with the considerable histologic variability of drusen but may be problematic in future attempts to automate drusen recognition based on internal characteristics. These in vivo morphologic parameters may relate closely to substructural elements imaged with light and electron microscopy of cadaveric eyes. Our imaging findings may very well have pathologic implications. For example, the appearance of hyper-reflective foci over drusen may represent retinal pigment migration6
or liberation that may signal a progression of disease. Also, the cores of hyper-reflectivity may relate to the presence of known drusen substructures, such as the above-mentioned glycoprotein cores or processes of choroidal dendritic cells. They could also represent the degree of complement-related activity in patients’ eyes or the beginning of a choroidal neovascular membrane and may potentially be an imaging biomarker for level of disease or risk of progression. The potential for tomographic diversity in drusen features even among similar-appearing, photographically soft-indistinct drusen, for example, cannot be overstated. The additional information regarding drusen structure revealed by SDOCT could provide an evolutionary step in our current paradigm of risk determination in AMD ().