Semiautomated Segmentation of the Choroid in Spectral-Domain Optical Coherence Tomography Volume Scans

Abstract

Purpose.: Changesin the choroid, in particular its thickness, are believed to be of importance in the pathophysiology of a number of retinal diseases. The purpose of this study was to adapt the graph search algorithm to semiautomatically identify the choroidal layer in spectral-domain optical coherence tomography (SD-OCT) volume scans and compare its performance to manual delineation.

Methods.: A graph-based multistage segmentation approach was used to identify the choroid, defined as the layer between the outer border of the RPE band and the choroid-sclera junction. Thirty randomly chosen macular SD-OCT (1024 × 37 × 496 voxels, Heidelberg Spectralis) volumes were obtained from 20 healthy subjects and 10 subjects with non-neovascular AMD. The positions of the choroidal borders and resultant thickness were compared with consensus manual delineation performed by two graders. For consistency of the statistical analysis, the left eyes were horizontally flipped in the x-direction.

Results.: The algorithm-defined position of the outer RPE border and choroid-sclera junction was consistent with the manual delineation, resulting in highly correlated choroidal thickness values with r = 0.91 to 0.93 for the healthy subjects and 0.94 for patients with non-neovascular AMD. Across all cases, the mean and absolute differences between the algorithm and manual segmentation for the outer RPE boundary was −0.74 ± 3.27 μm and 3.15 ± 3.07 μm; and for the choroid-sclera junction was −3.90 ± 15.93 μm and 21.39 ± 10.71 μm.

Conclusions.: Excellent agreement was observed between the algorithm and manual choroidal segmentation in both normal eyes and those with non-neovascular AMD. The choroid was thinner in AMD eyes. Semiautomated choroidal thickness calculation may be useful for large-scale quantitative studies of the choroid.

Introduction

Spectral-domain optical coherence tomography (SD-OCT) is an interference-based, noninvasive, in vivo imaging technique. It provides a three-dimensional, cross-sectional, microscale depiction of the optical reflectance properties of biological tissues. ¹ SD-OCT can directly access the spectrum, and thus provides a rapid acquisition of the three-dimensional images of interest. ² Improvements to existing SD-OCT, such as frame averaging, despeckling, and enhanced image contrast, have provided even better definition of deeper intraocular structures, such as the choroidal stroma and choroidal vasculature structures. ^3,4  

The choroid is a major vascular layer of the eye and provides oxygen and nourishment to the outer layers of the retina. ⁵ Recently, many commercial SD-OCT instruments have received software updates that enable enhanced depth imaging to better visualize the choroid. Changes in the choroid, in particular its thickness, have been hypothesized to be of critical importance in the pathophysiology of a number of retinal diseases. ^6–13 Diseases for which changes in OCT-determined choroidal thickness are of relevance including glaucoma, ⁷ high myopia, ⁸ neovascular and non-neovascular AMD, ^9–11 central serous chorioretinopathy, ¹² and Vogt-Koyanagi-Harada disease. ¹³  

The choroidal thickness determinations in these studies, including a recent investigation of healthy subjects by our group, ¹⁴ were performed by manual delineation of the choroidal layer. This is potentially tedious and time-consuming, particularly for computing choroidal volumes from dense three-dimensional scans, and ultimately limits the application of these approaches to large-scale studies. With the broad availability of SD-OCT capable of volumetric scanning, an objective approach to identify the choroid and define its changes during disease progression and follow-up treatment is desirable and necessary. 

In 2006, Li et al. ¹⁵ presented a graph search framework for the multiple layer segmentation of mutually interacting surfaces in three-dimensional volumetric images. It was later adapted for the multiple retinal layer segmentation in SD-OCT volumes and has demonstrated a great suitability in several applications. ^16–20 For instance, Garvin et al. ¹⁹ adapted it for seven retinal layers' segmentation in SD-OCT volumes. Lee et al. ²⁰ applied a fast multiscale scheme to segment four retinal layers. However, none of those has identified the choroidal layer, probably because the choroid is sometimes not as well visualized given its depth and distance from the zero delay, which challenges the segmentation. 

The purpose of this study was to adapt the graph search algorithm to semiautomatically identify the choroidal layer in SD-OCT volume scans in normal and diseased eyes, to quantify the choroidal thickness, and to compare the algorithm's performance to manual delineation by expert graders. 

Materials and Methods

Subject Recruitment

Twenty healthy subjects with healthy eyes and 10 subjects with bilateral non-neovascular AMD were enrolled in this study. For the healthy subjects, the absence of any ocular disease in either eye was confirmed by ophthalmoscopic examination. Subjects with non-neovascular AMD (all with large drusen, with or without pigment epithelial changes, but no atrophy) were recruited from Medical Retina Clinics at the Doheny Eye Institute of the University of Southern California. Patients with non-neovascular AMD were chosen because they have known abnormalities in choroidal thickness and have a high prevalence in a retina clinic population. All subjects provided written informed consent. The study was approved by the Institutional Review Board of the University of Southern California and adhered to the tenets set forth in the Declaration of Helsinki. 

OCT Imaging

For each subject, both eyes underwent volume OCT imaging using a Heidelberg Spectralis (Heidelberg Engineering, Heidelberg, Germany) SD-OCT in accordance with the existing standardized image acquisition protocol used by the Doheny Imaging Unit. All scans consisted of a macular cube scan pattern of 1024 (A-scans) × 37 (B-scans) × 496 voxels. The physical scan dimensions varied slightly between cases, but were on average 5.88 × 4.93 × 1.92 mm. Scans were obtained with nine times averaging, with the scan oriented for vitreous zero delay. Choroidal zero delay was not selected, as that has generally been used only for individual B-scans and not volume scans by our imaging unit. However, scanning was performed close to the zero delay line to optimize choroidal sensitivity. The voxel depth was 8 bits in gray scale. For each subject, one eye was randomly chosen for subsequent segmentation analysis. 

Multistage Multisurface Retinal Layer Segmentation

Overall, we approached the segmentation of the choroidal band - the layer between the outer RPE and choroid-sclera junction, using a graph-based ^15–20 multistage ²⁰ multisurface segmentation approach. An additional three surfaces, specifically the internal limiting membrane (ILM), inner-outer segment junction (IS-OS), and the inner RPE surface were also segmented to facilitate the choroidal segmentation. 

The graph search approach used in this study was inspired by the strategy previously described by Li et al. ¹⁵ Multiple surface segmentation could be considered as an optimization problem with the goal being to find a set of surfaces with the minimum cost such that the found surface set was feasible. To find a set of surfaces with the minimum cost, a graph with a subset of graphs corresponding to each individual surface was constructed. The cost function was a signed edge-based term, favoring a dark-to-bright or bright-to-dark intensity transition based on different surfaces. Surface feasibility constraints (i.e., smoothness constraints within a particular surface and interaction constraints between different surfaces) were applied to limit the neighborhood searching. In Li et al.'s ¹⁵ previous approach, both the smoothness and interaction constraints were of a constant value. In this study, however, the varying smoothness and interaction constraints were applied to allow more flexibility. ¹⁹  

For the choroidal band segmentation, an estimated morphological model was employed to the interaction constraints. Based on a recent work from our group ¹⁴ that studied topographical changes in posterior pole choroidal thickness in a cohort of 55 normal eyes, we were able to infer normal, expected regional changes in the choroidal thickness. Specifically, relative to the foveal center, the choroidal thickness shows a significant reduction nasally (−15%) and temporally (−14%). In contrast, it shows a slight increase superiorly (+4%) and is relatively stable/consistent inferiorly (−1% decrease). To summarize, the choroidal thickness varies markedly relative to the foveal center in the nasal-temporal direction (x-direction in the OCT images) at a range of −14% to approximately −15%, but is relatively stable in the superior-inferior direction (y-direction in the OCT images) at a range of +4% to approximately −1%. The normal and drusen datasets included in our present study demonstrated a similar regional trend in choroidal thickness. In designing the interaction constraints, using the A-scan (y-direction) located at the foveal center as a reference, we applied a mathematical model on the interaction constraints, which had a greatest value at the central foveal A-scan and linearly decreased bilaterally at each B-scan (x-direction). In each A-scan (y-direction), the interaction constraints remained the same. 

In this study, two OCT image datasets consisting of normal eyes and eyes with non-neovascular AMD containing drusen were used. We first segmented the multiple surfaces in the OCT images from normal eyes. To facilitate the surface segmentation, we also applied the multiscale graph search presented by Lee et al. ²⁰ More specifically, as shown in Figure 1, the algorithm sequentially downsampled the original three-dimensional SD-OCT image (Fig. 1a) to four, two, and one times and performed the multilayer segmentation in the three different stages. For stage 1 (Fig. 1b), the graph search approach ^15–20 was applied to first simultaneously segment the ILM (magenta arrow) and IS-OS junction (red arrow) and a penalty was applied to the region above the surface of the IS-OS junction. The double-surface graph search was then performed to simultaneously identify the outer RPE (blue arrow) and the choroid-sclera junction (orange arrow). Based on the segmented surfaces from stage 1, the cost function at the low probability positions for the search of the corresponding surfaces at stage 2 (Fig. 1c) was penalized. The same four surfaces segmented in stage 1 were refined in stage 2 based on the penalized cost function. In addition, in stage 2, an additional surface, namely the inner RPE (green arrow), was also defined using a single surface graph search. In stage 3 (Fig. 1d), the same penalizing strategy was applied in the original image space, and the segmentation of the five surfaces from stage 2 was defined more accurately in the high resolution (non-downsampled) image. 

Figure 1. 

 

Example illustration of the multistage layer segmentation in a B-scan of a healthy eye. (a) Original SD-OCT slice. (b–d) Layer segmentation at stages 1, 2, and 3, respectively. In (b), the four surfaces from the top to bottom are ILM (magenta arrow), IS-OS junction (red arrow), outer RPE (blue arrow), and choroid-sclera junction (orange arrow), respectively. In (c) and (d), the five surfaces from the top to bottom are ILM (magenta arrow), IS-OS junction (red arrow), inner RPE (green arrow), outer RPE (blue arrow), and choroid-sclera junction (orange arrow), respectively.

To perform the segmentation in the non-neovascular AMD images containing drusen, the algorithm for the four-surface segmentation at stage 1 was the same as that in the normal eyes. At stages 2 and 3, the segmentation method for the surfaces of ILM, IS-OS junction, outer RPE, and choroid-sclera junction remained the same. However, for the segmentation of the inner RPE, possible drusen positions were estimated based on the combined presegmented layer of the IS-OS junction and outer RPE, and a tuned cost penalty and smoothness constraints allowing more flexibility were applied on the positions when  

 

where t is the thickness of the IS-OS junction and outer RPE at that position, and μ and σ are the mean thickness and SD of the IS-OS junction and outer RPE, respectively. For both the normal eyes and the eyes with AMD, a thin-plate spline fitting ^20–22 was applied to smooth the segmented surfaces.  

Comparison of the Algorithm-Defined Borders with Manual Segmentation

Although the graph search-based algorithm permitted segmentation of multiple surfaces, only two (outer RPE band and the choroid-sclera junction) were relevant for choroidal segmentation and subsequent computation of choroidal thickness. Thus, to evaluate the accuracy of the algorithm segmentation, the two relevant surfaces of these same cases were manually segmented by two trained, certified OCT graders from the Doheny Image Reading Center who were masked to the algorithm-defined results. Because we desired a high level of precision, even a single pixel discrepancy in the position of the border at any location was deemed to constitute a discrepancy. The traditional approach in our image-reading center has been to determine one consensus result rather than average the results of two graders. For that reason, when both graders met to review the gradings, they were forced to come to agreement on the position of the boundaries at every A-scan location. In some cases, one or the other grader's assessment was accepted; in other cases, the two graders came up with a new consensus location during the adjudication process, and finally, in some cases the graders could not agree and the reading center medical director (SRS) had to make the final determination. 

Twenty normal eyes and 10 eyes with AMD were used in this study. The algorithm segmented all the B-scans in all the 30 images of the normal eyes and the eyes with AMD. The manual segmentation was performed for all the 10 eyes with AMD. However, for the 20 normal eyes, in some B-scans of eight cases, neither the graders nor the adjudicator could define the full extent of the outer border of the choroid (i.e., the choroid-sclera border) due to poor visibility. The eight eyes with such B-scans were still separately considered for comparative analyses with the algorithm-segmented boundaries, but the individual B-scans in which the full extent of the choroid (average number of B-scans with nonvisible outer boundary: eight B-scans/case) could not be seen were excluded from the comparative analyses. The 12 normal eyes with the full extent of the choroid in all the B-scans were all manually segmented and included in the comparative analyses. For the convenience of comparison, the 12 normal eyes with the full extent of the choroid in all the B-scans were labeled as “normal group 1” and the remaining eight normal eyes in which the outer choroid was not visible in a portion of some B-scans was labeled as “normal group 2.” 

The mean differences in choroidal thickness (at each z position) between the algorithm and the manual segmentation were calculated for each case, and compared using Pearson's correlation coefficients. In addition, the mean and absolute differences in the z position of the two boundaries (for each case) were also computed. To allow consistent statistical analysis, all the left eyes were horizontally flipped in the x-direction. 

Results

Table 1 provides the demographics for the healthy subjects and subjects with AMD. Table 2 shows the mean and absolute mean border positioning differences between the algorithm-defined and manual-delineated choroidal borders, the thickness, and the thickness differences of the choroidal layers by the algorithm and manual segmentation for “normal group 1.” Table 3 provides the mean and absolute mean border positioning differences, thickness, and the thickness differences between the algorithm-defined and manual-delineated choroidal borders for “normal group 2.” Table 4 provides the mean and absolute mean border positioning differences between the algorithm-defined and manual-delineated choroidal borders, the thickness, and the thickness differences of the choroidal layers of the algorithm and manual segmentation for the 10 eyes with AMD. Although not the main focus of this study, Table 5 provides a summary of the mean age and the mean choroid thickness from the two normal groups and the group with AMD. 

Table 1. 

 

Demographics of Subjects

Normal Group 1 and Normal Group 2
Eyes (subjects)	20 (20)
Age, y, mean ± SD (range)	29.6 ± 7.16 (20∼40)
Refractive error, diopters, mean ± SD (range)
	–0.83 ± 2.33 (−6.38∼+2.00)
Axial length, mm, mean ± SD (range)
	23.58 ± 1.12 (21.17∼25.49)
Eyes with Non-Vascular AMD
Eyes (subjects)	10 (10)
Age, y, mean ± SD (range)	75.9 ± 9.0 (62∼88)

Table 2. 

 

Differences in Choroidal Thickness and Boundary Positions between Algorithm and Manual Segmentation from “Normal Group 1”

	Difference in Choroidal Border Positions (Algorithm – Manual)				Mean Choroidal Thickness
	Absolute Mean		Mean
	Voxels	μm	Voxels	μm	Voxels	μm	r (95% CI)
Outer RPE border	0.90 ± 0.86	3.48 ± 3.32	−0.50 ± 0.81	−1.92 ± 3.14
Choroid-sclera junction	7.04 ± 1.92	27.23 ± 7.43	−0.53 ± 3.25	−2.05 ± 12.59
Algorithm					72.35 ± 7.79	280.00 ± 30.15	0.91 (0.80∼0.95)
Manual					72.38 ± 8.69	280.12 ± 33.64
Difference					−0.03 ± 3.65	−0.12 ± 14.12

 The negative value indicates the algorithm-segmented border is above the corresponding border from the manual delineation. CI, confidence interval; r, Pearson's correlation coefficient of algorithm- and manually defined choroidal thickness.

Table 3. 

 

Differences in Choroidal Thickness and Boundary Positions between Algorithm and Manual Segmentation from “Normal Group 2”

	Difference in Choroidal Border Positions (Algorithm – Manual)				Mean Choroidal Thickness
	Absolute Mean		Mean
	Voxels	μm	Voxels	μm	Voxels	μm	r (95% CI)
Outer RPE border	0.55 ± 0.54	2.13 ± 2.11	−0.27 ± 0.66	−1.04 ± 2.57
Choroid-sclera junction	1.86 ± 5.19	7.21 ± 20.09	−1.58 ± 5.36	−6.10 ± 20.74
Algorithm					87.23 ± 11.18	337.56 ± 43.26	0.93 (0.82∼0.97)
Manual					88.54 ± 14.80	342.64 ± 57.27
Difference					−1.31 ± 5.86	−5.08 ± 22.68

 The negative value indicates the algorithm-segmented border is above the corresponding border from the manual delineation.

Table 4. 

 

Differences in Choroidal Thickness and Boundary Positions between Algorithm and Manual Segmentation in Eyes with Non-neovascular AMD

	Difference in Choroidal Border Positions (Algorithm – Manual)				Mean Choroidal Thickness
	Absolute Mean		Mean
	Voxels	μm	Voxels	μm	Voxels	μm	r (95% CI)
Outer RPE border	0.92 ± 0.91	3.56 ± 3.53	0.24 ± 1.03	0.92 ± 3.98
Choroid-sclera junction	6.65 ± 1.84	25.72 ± 7.13	−1.13 ± 4.15	−4.36 ± 16.08
Algorithm					58.33 ± 15.04	225.74 ± 58.19	0.94 (0.83∼0.98)
Manual					59.70 ± 14.08	231.02 ± 54.48
Difference					−1.36 ± 4.85	−5.28 ± 18.78

 The negative value indicates the algorithm-segmented border is above the corresponding border from the manual delineation.

Table 5. 

 

Mean Age and Mean Choroid Thickness from Normal Eyes and Eyes with AMD

Mean Age, y	Mean Choroid Thickness from Algorithm
Normal group 2
28.3 ± 6.8	87.23 ± 11.18 (voxels)
Range (20∼40)	337.56 ± 43.26 (μm)
Normal group 1
30.4 ± 6.2	72.35 ± 7.79 (voxels)
Range (20∼40)	280.00 ± 30.15 (μm)
Eyes with AMD
75.9 ± 9.0	58.33 ± 15.04 (voxels)
Range (62∼88)	225.74 ± 58.19 (μm)

Across all cases of the three groups, the mean and absolute border position difference for the outer RPE boundary was −0.19 ± 0.84 pixels (−0.74 ± 3.27 μm) and 0.81 ± 0.79 pixels (3.15 ± 3.07 μm) and for the choroid-sclera junction was −1.01 ± 4.11 pixels (−3.90 ± 15.93 μm) and 5.53 ± 2.77 pixels (21.39 ± 10.71 μm). The performance of the algorithm was better centrally when only the central 3-mm² circular region of the volume cube was considered. The mean and absolute border position differences (all groups combined) were −0.17 ± 0.63 pixels (−0.66 ± 2.45 μm) and 0.69 ± 0.59 pixels (2.67 ± 2.30 μm) for outer RPE boundary and −0.91 ± 3.08 pixels (−3.51 ± 11.95 μm) and 4.70 ± 2.07 pixels (18.18 ± 8.03 μm) for the choroid-sclera junction. The thickness measurements between the algorithm-defined and manual-delineated choroid demonstrated a good correlation with r = 0.91, 0.93, and 0.94 for the three groups, respectively, and r = 0.93 for all three groups combined. 

Figure 1 provides an illustration of the multistage layer segmentation approach. Figure 2 illustrates the layer segmentation result in a healthy subject and a patient with non-neovascular AMD (drusen). Figures 3 and 4 illustrate the mean choroidal thickness and thickness ratio (relative to the choroidal thickness at the fovea center) from the algorithm and manual segmentation of 12 healthy subjects and 10 AMD patients respectively. 

Figure 2. 

 

Example illustration of multilayer segmentation result in a healthy eye and diseased eye with non-neovascular AMD. Left column: B-scans from original SD-OCT volume. Right column: B-scans overlapping with the algorithm segmentation. Right column, top to bottom: A B-scan from a healthy eye, from an AMD eye with a large drusen under the foveola, and a smaller drusen in the region outside the foveal center. The five layers from the top to bottom are ILM, IS-OS junction, inner RPE, outer RPE, and choroid-sclera junction, respectively.

Figure 3. 

 

Mean choroidal thickness and thickness ratio (relative to the choroidal thickness at the foveal center) from the algorithm and manual segmentation of 12 healthy eyes.

Figure 4. 

 

Mean choroidal thickness and thickness ratio (relative to the choroidal thickness at the foveal center) of the algorithm and manual segmentation from 10 eyes with non-neovascular AMD.

Discussion

In this study, we adapted the graph search algorithm to semiautomatically identify the choroidal layer in SD-OCT volume scans and reported its performance in normal eyes and eyes with non-neovascular AMD. Good thickness agreement was observed between the algorithm and manual segmentation of the macular choroid, in both normal eyes (r = 0.91 for the thickness measurement for “normal group 1,” as shown in Table 2, and r = 0.93 for the thickness measurement for “normal group 2,” as shown in Table 3) and those with non-neovascular AMD (drusen) (r = 0.94 for the thickness measurement as shown in Table 4). 

Although this was not the primary purpose of this study, we observed that, compared with normal eyes, the choroid was thinner in the AMD eyes. The thickness measurement was lower in the AMD eyes, and this finding was consistent throughout the macular region sampled by the OCT. This observation of a thinner choroid in AMD is in agreement with previously published reports. ²³ However, the mean age of the healthy subjects was also considerably lower than the AMD patients, and choroidal thickness is also known to decrease with age. 

We hypothesized that the poor visibility of the choroid-sclera junction in some B-scans in “normal group 2” of the eight cases was due to a thicker choroid, limiting the penetration of light through its full extent. From the thickness measurement results in Tables 2 and 3, the average choroid thickness of “normal group 2” was thicker than that in “normal group 1,” which supported this hypothesis. 

The algorithm performance was best centrally, with greater segmentation errors at the temporal and nasal edges of the scans. We suspect this is due to two reasons. One reason is that the image quality of the B-scans is poorer at the edges of the scan compared with the center, possibly due to the nature of the tracking technique. In this case, the confidence in the segmentation may not be as high for both the algorithm and manual delineation, leading to a greater apparent discrepancy. The second explanation is that the graph search of the surfaces was constrained by the neighborhood smoothness, but these same constraints could not be applied at the edge of the image. Despite these peripheral failures in some cases, the overall performance of the algorithm was good. Thus, it may be useful for large-scale quantitative studies, particularly for central choroidal thickness calculation. In addition, any errors or failures of the algorithm in selected B-scans could potentially be corrected manually. 

Despite the favorable performance of the choroidal segmentation algorithm, there are several limitations in this preliminary study. First, only anisotropic (1024 × 37 × 496 voxels) SD-OCT volumes were used in this study. In addition, only a limited number of B-scans were acquired for the SD-OCT volumes because of the additional time required to acquire volume scans when using a tracking OCT. As a result, the 37 B-scan volume cube (approximately 123 μm apart between adjacent B-scans) is the standard volume acquisition protocol used in the Doheny Imaging Unit. It is possible, however, that a more isotropic SD-OCT dataset could yield more accurate segmentation and resultant thickness measurements. A second limitation of the described approach is that a tuned cost penalty and smoothness constraints were applied for the drusen eye dataset. Application of these constraints could potentially affect the generalizability of this approach, as it will not be fully automated for all types of eyes and diseases. The true generalizability of the approach (or subsequent modifications) needs to be evaluated in much larger, heterogeneous datasets that will be the subject of our future studies. A third limitation of our study is the OCT scanning wavelength that was used. For the Spectralis OCT, the center wavelength is 870 nm. At this wavelength, there may be only partial penetration of the choroid even with enhanced depth imaging approaches, particularly in cases with thicker choroids. In fact, from the results presented in Tables 2, 3, and 4, the segmentation in the group with non-neovascular AMD eyes (Table 4) performed slightly better than that in the two normal groups (Tables 2, 3). The major reason was probably that the thinner choroid in the eyes with non-neovascular AMD allowed a deeper penetration of the light signal and better visualization of the choroid. However, as described previously, in the two normal groups, especially for the cases in “normal group 2,” the thicker choroid limited the light penetration. Although, the graders excluded B-scans in which the outer border of the choroid was not identifiable, there may have been other cases in which the visibility was poor but grading was still deemed to be possible. The accuracy of the manual segmentation may not have been as good in these cases. Similarly, the poor signal at the outer choroid in these cases, may have affected the performance of the algorithm. In these cases, the algorithm performed a best fit of the choroid-sclera junction, which may have undermined the accuracy of the segmentation. This problem, however, may be addressed in the near future, given the recent availability of Fourier Domain OCT devices with longer wavelengths (1050 nm), such as the Topcon swept-source OCT instrument (Topcon, Oakland, NJ). The final limitation of our algorithm is that it was evaluated only in normal and non-neovascular AMD eyes. The performance in a variety of other retinal and choroidal diseases remains to be determined. 

In summary, in this study, the graph-search algorithm was adapted to semiautomatically delineate and quantify the choroid in normal eyes and eyes with non-neovascular AMD. The algorithm showed excellent agreement with human expert manual segmentation, particularly in the central portions of the volume scans. The semiautomated choroidal thickness calculation may be useful for large-scale quantitative studies of the choroid in normal and diseased eyes in the future. 

Footnotes

 Supported by the Beckman Macular Degeneration Research Center and a Research to Prevent Blindness Physician Scientist Award.

Footnotes

 Disclosure: Z. Hu, None; X. Wu, None; Y. Ouyang, None; Y. Ouyang, None; S.R. Sadda, Carl Zeiss Meditec (F, C), Optos (F), Optovue, Inc. (F), Allergan (C), Genentech (C), Regeneron (C), Optos (C)