Repeatability and Reproducibility of Choroidal Vessel Layer Measurements in Diabetic Retinopathy Using Enhanced Depth Optical Coherence Tomography

Abstract

Purpose.: To describe novel segmentation protocols for choroidal layers, Sattler's medium and Haller's large vessel layers, using enhanced depth imaging optical coherence tomography (EDI-OCT), and to examine the repeatability and reproducibility of these measurements in eyes with diabetic retinopathy.

Methods.: Fifty-one patients with Type 2 diabetes mellitus were imaged using custom EDI scanning protocols. Detailed segmentation was performed to quantify the retina, choroid, Haller's large, and Sattler's medium vessel layers in the total macular circle (TMC) and foveal central subfield (FCS). The coefficient of repeatability (CR) and intraclass correlation coefficient (ICC) were used as a measure of repeatability and relative reliability within graders. Reproducibility or interobserver variability was assessed using Bland-Altman plots and 95% limits of agreement (LoA).

Results.: Intragrader CR of the retina, choroid, Sattler's, and Haller's layers for thickness measurements were 19.2, 26.9, 35.2, and 29.2 μm, respectively. Intergrader 95% LoA were 27.9, 41.5, 38.6, 31.1 μm (thickness), respectively. Choroidal sublayer measurements showed good intraobserver reliability (ICC 0.78–0.98). Interobserver variability for retinal and choroidal measurements was not significantly different (P > 0.45). Measurements from the TMC showed slightly better repeatability and agreement compared with the FCS alone. Mean intergrader differences were reduced after training, and were most apparent in choroidal sublayers.

Conclusions.: The choroidal vascular sublayers can be quantified with good reliability, repeatability, and reproducibility. Accurate quantitative assessment of these sublayers may provide new insights into the role of the choroid in visual loss in patients with diabetic retinopathy, and prove useful for future clinical trials.

Introduction

The choroidal circulation forms an integral part of metabolic exchange in the outer retina. ¹ This is of particular significance in the macula, due to the lack of retinal vasculature, the foveal avascular zone, and a high metabolic demand from an increased photoreceptor density. Dysfunction of the choroidal circulation has been long implicated in diabetic retinopathy and maculopathy. ² Since the advent of enhanced-depth optical coherence tomography (EDI-OCT) imaging, allowing noninvasive examination of the choroid in vivo, ³ there has been renewed interested in diabetic choroidopathy as a disease entity. ^4–7  

Evidence for diabetic choroidopathy was first noted in histopathological studies, where abnormalities, such as arteriosclerosis, choriocapillaris degeneration, focal scarring, and neovascularization, were observed. ^8–10 This was followed by indocyanine green angiographic findings in the diabetic choroid, showing hyper- and hypofluorescent spots at the level of the choriocapillaris, suggested to represent aneurysms or deficits in the choroidal vasculature. ^11,12  

A number of preliminary studies have used OCT, either EDI-OCT or long-wavelength OCT research prototype systems, to examine the choroid in patients with diabetic retinopathy. These studies have reported a thinner choroid compared with healthy eyes, with an increased disparity in eyes with a greater severity of diabetic retinopathy, the presence of macular edema, and prior pan-retinal laser photocoagulation. ^4–7 However, the thinning observed in eyes of patients with diabetic retinopathy has been described to exceed the magnitude of possible choriocapillaris atrophy in isolation. ⁴ It is therefore conceivable that choroidal vascular changes observed in patients with diabetes occur not only in the choriocapillaris, but also within its medium (Sattler's) and large (Haller's) vessel layers. To date, analysis of these choroidal vascular sublayers using OCT has been restricted to small studies using long-wavelength OCT in healthy subjects. ^13,14  

In this study, we describe novel segmentation protocols for the choroidal medium and large vessel layers in EDI-OCT image sets, and examine the repeatability and reproducibility of thickness and volume measurements in the diabetic macula. 

Patients and Methods

Inclusion Criteria and Data Collection

Clinical and imaging data were collected retrospectively from patients attending medical retinal clinics at Moorfields Eye Hospital, London, United Kingdom. All data were collected over a 6-month time period. This study was part of a separate ongoing clinical study, examining the OCT features of diabetic retinopathy; in particular, the relationships between retinal and choroidal thickness measurements to the presence of angiographic ischemia in diabetic eyes. Approval for data collection and analysis was obtained from the local ethics committee and adhered to the tenets set forth in the Declaration of Helsinki. 

Patient demographic data, visual acuities, and retinopathy/maculopathy grades, were obtained from standardized electronic reports in the United Kingdom National Screening Committee (UK NSC)–Diabetic Eye Screening Programme, a grading system that has been described in more detail elsewhere. ¹⁵ Patient age at time of attendance, and presence of ocular comorbidity, were obtained from electronic patient records. 

Patients attending the clinic with a diagnosis of Type 2 diabetes mellitus and OCT images acquired using the Spectralis imaging system (Heidelberg Engineering, Heidelberg, Germany) were included. Patients with ocular comorbidities, including retinal arterial or venous occlusion, epiretinal membrane, neovascular age-related macular degeneration, inherited macular disease, or macular scarring of any etiology, were excluded. 

OCT image sets had to be of sufficient quality to allow segmentation of retinal and choroidal layers. A single eye was initially selected using permuted-block randomization for inclusion in the study. Where the image quality was not of sufficient quality for grading, the fellow eye was included. Six eyes were excluded based on these criteria. Boundaries were segmented by two blinded graders (SF and HM). HM was initially the “untrained” grader, and underwent training on OCT manual segmentation of retinal and choroidal layers during the course of this study. 

Acquisition and Analysis of OCT Image Sets

Grading Methods.

Spectral domain OCT images were obtained using the Spectralis imaging system (Heidelberg Engineering). All OCT image sets contained a minimum of 13 B-scans distributed in a horizontal raster pattern overlying the area covered by the nine subfields of the Early Treatment Diabetic Retinopathy Study (ETDRS) grid. Custom image analysis software (OCTOR; Doheny Image Reading Center, Los Angeles, CA) was used for quantitative analysis of OCT image sets. OCTOR has been described and validated in previous reports. ^16,17  

Segmentation Protocol for Retinal Spaces.

Boundaries were manually segmented in accordance with standardized OCT grading protocols. ¹⁸ The retinal space was defined as the space lying between the inner aspect of the internal limiting membrane and the outer border of the photoreceptor outer segments. ^19,20  

Segmentation Protocol for Choroidal Spaces.

The choroid was defined as the space between the outer border of the retinal pigment epithelium and choroidoscleral junction. The choroid was further subdivided into Haller's large vessel and Sattler's medium vessel layers (Fig. 1). Haller's large vessel layer was defined as the outer choroid, consisting of large hypointense spaces representing large vascular luminal spaces. Sattler's medium vessel layer consisted of small- to medium-sized hypointense spaces, surrounded by hyperintense stroma (increase scattering by high density of melanocytes), giving a mottled appearance on scans. ¹⁴ This layer also included the choriocapillaris, which at 10 μm is not easily distinguished from the Sattler's medium vessel layer on OCT images. The detailed choroidal segmentation protocol is illustrated in Figure 1 (Figs. 1a, 1b). This protocol was used in the training of graders and was used before all grading sessions. 

Figure 1

 

An example of a foveal-centered B-scan, acquired using the EDI-OCT protocol. Annotations illustrate the instructions from the choroidal segmentation protocol, and demonstrate identification and segmentation of choroidal layers by examining changes in light reflectance at the different tissue interfaces. (a) Black arrows indicate the outer border of the RPE, and white arrowheads the outer choroidal border (OCB), which represents the choroidoscleral interface. (b) Segmentation lines on the outer border of the RPE (red) and OCB (green) demarcating the total choroidal area/volume. (c) Black arrowheads indicate the junction between Sattler's medium and Haller's large vessel layers. Black stars indicate difficulty areas where the interface between both Sattler's and Haller's layers is ambiguous. Segmentation of these areas may be aided by clues from adjacent B-scans. (d) Segmentation line on the Sattler's and Haller's vessel layer interface (yellow).

The mean retinal and choroidal thicknesses (μm) and volumes (mm³) were then calculated for the Early Treatment Diabetic Retinopathy Study (ETDRS) areas 1 to 9 or the “total macular circle” (TMC), and ETDRS area 9 which corresponds to the foveal central subfield (FCS). 

Assessment of Repeatability of Segmentation.

A subset of OCT images (n = 20) was segmented three times by a single grader (HM): once before training, and twice after training. This OCT image set was also segmented twice by the experienced grader (SF) to assess repeatability. The time interval between each episode of segmentation was greater than 2 weeks for both graders. 

Assessment of Reproducibility of Segmentation.

All OCT images (n = 51) were segmented by both graders in a masked fashion. 

Statistical Analysis

Patient demographic and imaging data were analyzed with frequency and descriptive statistics. 

The coefficient of repeatability (CR) of retinal and choroidal thickness measurements was calculated using the within-grader standard deviation (S_w) derived from the intragrader mean square of differences. The CR, as defined by Bland and Altman, ²¹ was calculated as 1.96 times the SD of the differences between two measurements: CR = 1.96 × √(2S² _w) or 2.77S_w. To allow comparison with other studies, we also expressed the CR as a percentage of the mean measurement for all retinal and choroidal layers (CR/Mean), with a lower the CR/Mean percentage representing greater repeatability within graders. The intraclass correlation coefficient (ICC) was used as a measure of relative reliability of measures within graders. 

Reproducibility or interobserver variability was assessed using Bland-Altman plots, using the mean thickness and volume measurements, segmented by each grader. ²² The mean difference and confidence intervals were calculated. Agreement between graders was also examined using Bland-Altman analysis, with the 95% limits of agreement (μm) (LoA) between graders calculated, after confirming that the measurement differences were normally distributed using histograms. 

Statistical analyses were performed using SPSS software version 16 (SPSS, Inc., Chicago, IL) and MedCalc software version 10.3.2 (MedCalc Software, Mariakerke, Belgium). 

Results

A total of 51 eyes of 51 patients with a diagnosis of type 2 diabetes with a mean age of 60.1 years were included in this study. Patient demographic data and mean retinal and choroidal thickness and volume measurements from each grader area are summarized in Table 1. 

Table 1. 

 

Demographic Data of Study Cohort

Total Cohort, n = 51	Thickness Measurements, μm			Volume Measurements, mm3
	Mean, n (%)	SD	Range	Mean	SD	Range
Type 2 diabetes, n	51 (100)
DR
None	2 (3.9)
Nonproliferative DR	28 (54.9)
Proliferative DR active and treated	21 (41.2)
Diabetic maculopathy
No DME	21 (41.2)
DME	15 (29.4)
CSME active and treated	15 (29.4)
Male, n	33
Left eye, n	16
Age, y	60.1	13.6	28.0−84.0
VA, logMAR	0.27	0.26	−0.2−1.3
Retina
TMC
Grader 1	290.9	43.3	218.7−493.1	8.03	1.32	4.22 13.78
Grader 2	282.4	47.0	171.3−497.1	7.68	1.69	5.85−13.75
Grader 1	271.8	102.0	166.5−737.3	0.214	0.08	0.130−0.580
Grader 2	270.5	102.1	160.4−739.4	0.203	0.12	0.102−0.671
Choroid
TMS
Grader 1	232.7	66.5	86.6−389.7	6.42	1.90	2.38−10.98
Grader 2	224.7	62.2	97.5−359.5	6.12	1.82	2.69−10.26
FCS
Grader 1	251.1	73.0	88.1−409.0	0.198	0.057	0.070−0.320
Grader 2	242.9	67.9	87.1−407.4	0.191	0.045	0.061−0.304
Haller's layer
TMS
Grader 1	115.2	39.3	17.7−210.3	3.18	1.11	0.47−5.92
Grader 2	122.8	35.6	50.2−196.3	3.33	1.22	1.35−5.58
FCS
Grader 1	124.7	47.1	7.3−256.2	0.098	0.037	0.010−0.200
Grader 2	133.2	41.1	47.4−239.7	0.109	0.023	0.040−0.190
Sattler's layer
TMS
Grader 1	117.5	32.2	55.5−221.0	3.24	0.93	1.49−6.31
Grader 2	105.9	28.7	48.2−164.3	2.84	0.91	1.32−4.69
FCS
Grader 1	126.2	38.5	50.0−239.0	0.100	0.030	0.040−0.190
Grader 2	115.7	33.5	39.7−191.3	0.091	0.024	0.032−0.150

 DME does not meet the ETDRS criteria for treatment; CSME meets ETDRS criteria for treatment. LogMAR, logarithm of the minimum angle of resolution.

Repeatability of Measurements

The repeatability of retinal and choroidal measurements was assessed in subset of 20 eyes randomly selected from the study cohort. Each image set was manually segmented twice by each “trained” grader. (The junior grader, HM, received training before both episodes of segmentation.) The CR (μm or mm³), CR expressed as a percentage of mean thickness or volume (CR/Mean), and reliability (ICC) for all indices measured in the TMC and FCS are presented in Table 2. 

Table 2. 

 

Repeatability and Reliability of Retinal and Choroidal Thickness Measurements in Patients With Type 2 Diabetes

	CR, μm	CR/Mean, %	ICC	95% CI
Retina
TMC	19.2	6.7	0.98	0.95−0.99
FCS	49.0	18.1	0.98	0.95−0.99
Choroid
TMC	26.9	11.8	0.97	0.94−0.99
FCS	48.3	19.6	0.93	0.83−0.97
Haller's layer
TMC	35.2	29.6	0.86	0.68−0.94
FCS	38.2	29.7	0.78	0.53−0.91
Sattler's layer
TMC	29.2	26.0	0.89	0.73−0.95
FCS	39.0	32.2	0.88	0.73−0.95

 TMC: ETDRS areas 1 to 9; FCS: ETDRS area 9.

Repeatability of Thickness and Volume Measurements.

Thickness measurements of the TMC showed better repeatability than those at the FCS (Figs. 2, 3; Tables 2, 3). The CR of both retinal and choroidal thicknesses ranged from 19.2 to 35.2 μm in the TMC, and 26.9 to 49.0 μm in the FCS. TMC thickness measurements were also more reliable in retinal, total choroidal, and Sattler's medium vessel layers. The lowest CR/Mean percentages (greatest repeatability) were observed in the TMC of the retina and total choroidal thickness measurements (6.7 and 11.8, respectively). Segmentation for choroidal sublayers, Haller's large vessel, and Sattler's medium vessel layers showed less repeatability, with higher CR/Mean percentages. Repeatability of volume measurements were similar to that of thickness measurements, as observed in the CR/Mean percentages. 

Figure 2

 

Choroidal segmentation protocol, including detailed instructions on segmentation of the choroidal sublayers Haller's large vessel (HLV) and Sattler's medium vessel (SMV) layers. This was used in the training of graders, and read before all grading sessions.

Figure 3

 

Bar chart illustrating the interobserver mean difference in thickness measurements (μm) of the retina, total choroid, Haller's larger vessel, and Sattler's medium vessel layers. Measurements in the TMC, gray bars, and FCS, white bars, are presented.

Table 3. 

 

Repeatability and Reliability of Retinal and Choroidal Volume Measurements in Patients With Type 2 Diabetes

	CR, mm3	CR/Mean, %	ICC	95% CI
Retina
TMC	0.64	8.1	0.97	0.94−0.99
FCS	0.04	18.7	0.98	0.95−0.99
Choroid
TMC	0.83	13.1	0.97	0.92−0.99
FCS	0.04	20.6	0.92	0.80−0.97
Haller's layer
TMC	1.03	31.4	0.85	0.66−0.94
FCS	0.03	29.7	0.76	0.48−0.90
Sattler's layer
TMC	0.80	25.8	0.89	0.74−0.95
FCS	0.03	31.4	0.88	0.71−0.95

 TMC: ETDRS areas 1 to 9; FCS: ETDRS area 9.

The reliability of both thickness and volume measurements, as expressed by the ICC, showed good reliability (ICC > 0.8) across all layers, apart from measurements in Haller's large vessel layer at the FCS (thickness: ICC 0.78, volume: ICC 0.76). 

Reproducibility of Measurements

Reproducibility of Retinal and Choroidal Measurements.

The reproducibility of thickness and volume measurements between both trained graders are displayed in Tables 4 (thickness) and 5 (volume) for all retinal and choroidal layers. Bland-Altman plots were used to illustrate agreement between graders (Figs. 4–7; Tables 4–6). No significant changes were observed with interobserver variability for the range of retinal and choroidal measurements. 

Figure 4

 

Bland-Altman plot illustrating the reproducibility of retinal thickness measurements. Mean difference between graders, 8.5 μm; 95% confidence intervals (CIs), −19.4 to 36.5 μm. The variance ratio (F statistic) of intergrader differences showed no significant difference between graders (P = 0.91 TMC, P = 0.94 FCS).

Figure 5

 

Bland-Altman plot illustrating the reproducibility of total choroidal thickness measurements. Mean difference between graders, 8.0 μm; 95% CIs, −33.5 to 49.4 μm. The variance ratio (F statistic) of intergrader differences showed no significant difference between graders (P = 0.94 TMC, P = 0.68 FCS).

Figure 6

 

Bland-Altman plot illustrating the reproducibility of Haller's large vessel layer thickness measurements. Mean difference between graders, −4.1 μm; 95% CIs, −42.6 to 34.5 μm. The variance ratio (F statistic) of intergrader differences showed no significant difference between graders (P = 0.98 TMC, P = 0.49 FCS).

Figure 7

 

Bland-Altman plot illustrating the reproducibility of Haller's large vessel layer thickness measurements. Mean difference between graders, 13.7 μm; 95% CIs, −17.4 to 44.8 μm. The variance ratio (F statistic) of intergrader differences showed no significant difference between graders (P = 0.88 TMC, P = 0.45 FCS).

Table 4. 

 

Reproducibility of Retinal and Choroidal Thickness Measurements in Patients With Type 2 Diabetes

	Interobserver Mean Difference, μm	95% CI, μm	95% Limits of Agreement, μm
Retina
TMC	8.52	4.51−12.5	27.9
FCS	2.80	−4.26−9.87	48.8
Choroid
TMC	7.99	2.04−13.9	41.5
FCS	9.56	1.85−17.3	53.1
Haller's layer
TMC	−4.05	−9.96−1.86	38.6
FCS	−2.12	−10.9−6.63	56.5
Sattler's layer
TMC	13.7	8.95−18.5	31.1
FCS	14.0	7.16−20.8	44.0

Table 5. 

 

Reproducibility of Retinal and Choroidal Volume Measurements in Patients With Type 2 Diabetes

	Interobserver Mean Difference, mm3	95% CI, mm3	95% Limits of Agreement, mm3
Retina
TMC	0.36	0.18−0.54	1.26
FCS	0.00	−0.0035−0.0071	0.04
Choroid
TMC	0.31	0.10−0.52	0.43
FCS	0.01	0.0013−0.014	0.04
Haller's layer
TMC	−0.06	−0.24−0.13	1.21
FCS	−0.11	−0.14 to −0.09	0.17
Sattler's layer
TMC	0.42	0.28−0.56	0.92
FCS	0.01	0.0057−0.017	0.03

Table 6. 

 

The Variance Ratio (F Statistic) of Intergrader Differences in Retinal and Choroidal Measurements

	Thickness Measurements		Volume Measurements
	F Test	P Value	F Test	P Value
Retina
TMC	1.18	0.56	1.03	0.91
FCS	1.00	1.00	1.02	0.94
Choroid
TMC	1.14	0.64	1.02	0.94
FCS	1.15	0.62	1.12	0.68
Haller's layer
TMC	1.22	0.50	1.01	0.98
FCS	1.32	0.36	1.23	0.49
Sattler's layer
TMC	1.26	0.44	1.05	0.88
FCS	1.32	0.35	1.26	0.45

 TMC: ETDRS areas 1 to 9; FCS: ETDRS area 9.

To examine whether the 95% limits of agreement were significantly different between observers, the variance of the interobserver measurement differences were calculated and the F statistic (variance ratio) presented (Table 6). The variance ratio between graders was not statistically significant in all layers; however, the data showed a trend toward a higher variance ratio in the choroidal sublayers quantified at the FCS. TMC measurements showed similar variance ratios in all layers. 

Factors that may affect agreement were also analyzed. A univariate generalized linear model was fitted to the data, with the difference in thickness measurements between graders as the dependent variable, and retinopathy grade, maculopathy grade, age, sex, and visual acuity (VA) as covariates. VA was the only factor that showed significant association with retinal and choroidal thickness measurement difference between graders (Table 7). There was no significant association in the choroidal sublayers. Scatter plot analysis of VA and LoA did not reveal any association between these two factors (Fig. 8). 

Figure 8

 

Scatterplot showing the effect of VA on the thickness measurement differences (μm) between graders in the retina and choroid. No relationship was observed with visual acuity (regression coefficients: R ² = 0.029 retinal thickness, R ² = 0.093 choroidal thickness).

Table 7. 

 

Factors Affecting the Agreement Between Graders

	Difference Thickness Measurements, μm, Between Graders, F Statistic (P Value)
	Retina		Choroid		Haller's Layer		Sattler's Layer
	TMC	FCS	TMC	FCS	TMC	FCS	TMC	FCS
Retinopathy	1.45 (0.24)	0.03 (0.86)	3.24 (0.08)	1.29 (0.27)	2.48 (0.13)	0.33 (0.57)	0.01 (0.94)	4.34 (0.05)
Maculopathy	0.24 (0.87)	0.59 (0.63)	0.22 (0.88)	0.46 (0.71)	0.46 (0.71)	0.93 (0.44)	0.47 (0.71)	0.20 (0.90)
Age	1.77 (0.19)	0.40 (0.53)	1.40 (0.25)	0.81 (0.21)	1.64 (0.21)	0.01 (0.99)	0.01 (0.93)	0.71 (0.41)
Sex	3.23 (0.08)	2.64 (0.12)	0.10 (0.76)	1.01 (0.19)	1.76 (0.19)	0.16 (0.69)	1.78 (0.19)	1.81 (0.19)
VA	4.49* (0.04)	0.05 (0.82)	6.02* (0.02)	0.37 (0.55)	2.19 (0.15)	0.32 (0.57)	1.68 (0.20)	1.71 (0.20)

 Retinopathy: Eyes were classified as having no retinopathy, nonproliferative retinopathy, proliferative retinopathy, and treated retinopathy. Maculopathy: Eyes were classified as no maculopathy, nonclinically significant macular edema, and clinically significant macular edema (both treated and untreated).

*  Highlights factors that are significant at the P < 0.05 level.

The Effect of Training on Reproducibility of Measurements.

As expected, the mean intergrader differences were reduced after training. This was most apparent in measurements of the choroid and its sublayers (Table 8). 

Table 8. 

 

The Effect of Manual Segmentation Training on the Reproducibility of Retinal and Choroidal Thickness Measurements

	Before Training			After Training
	Intergrader Mean Difference, μ	95% CI, μm	95% Limits of Agreement, μm	Intergrader Mean Difference, μm	95% CI, μm	95% Limits of Agreement, μm
Retina
TMC	−5.98	−9.13−−2.83	8.63	4.07	−2.64−10.8	18.4
FCS	−4.44	−9.25−0.37	13.2	−8.96	−32.7−14.8	65.0
Choroid
TMC	23.7	11.3−36.1	34.0	12.3	0.29−24.2	32.8
FCS	33.1	9.09−57.13	65.8	21.3	0.34−42.2	57.4
Haller's layer
TMC	−14.3	−32.4−3.82	49.6	−4.74	−19.0−9.48	39.0
FCS	−6.74	−38.1−24.6	85.9	5.90	−17.9−29.7	65.3
Sattler's layer
TMC	37.1	24.0−50.1	35.7	16.7	5.31−28.2	31.3
FCS	39.1	23.6−54.7	42.5	14.7	3.20−26.3	31.6

The Effect of Clinically Significant Macular Edema on Reproducibility of Measurements.

The variance ratio between eyes with clinically significant macular edema (CSME) and eyes with “no CSME” was calculated. As expected, this was significantly different between retinal measurements (by definition of the subgroup of eyes with CSME would have greater variance compared with eyes with “no CSME”), but not in the total choroid or its sublayer measurements (Table 9). 

Table 9. 

 

The Variance Ratio (F Statistic) of Retinal and Choroidal Measurements in Eyes With and Without Clinically Significant Macular Edema

	Thickness Measurements		Volume Measurements
	F Test	P Value	F Test	P Value
Retina
TMC	2.52	0.02*	1.91	0.11
FCS	2.37	0.03*	2.32	0.04*
Choroid
TMC	1.16	0.68	1.12	0.75
FCS	1.49	0.33	1.60	0.24
Haller's layer
TMC	1.48	0.32	1.46	0.35
FCS	1.32	0.36	1.23	0.49
Sattler's layer
TMC	1.75	0.17	1.66	0.21
FCS	1.74	0.18	1.76	0.16

*   P values significant at 5% level.

Discussion

In this study, we describe novel segmentation protocols for delineating the medium and large vessel layers of the choroidal vasculature. Using this approach, we present data examining the repeatability and reproducibility of measurements of the retina, choroid, and choroidal sublayers (Haller's large vessel and Sattler's medium vessel layers) in a cohort of patients with Type 2 diabetes mellitus. We also evaluate the effects of clinical parameters, such as diabetic retinopathy and maculopathy, and assess the effects of grader training on the precision and reliability of measurements. 

The mean subfoveal choroidal thickness measurement in our cohort of patients was 233 μm (Grader 1) and 225 μm (Grader 2). This is thinner than observed in healthy subjects, which has been reported to range from 287 to 332 μm. ^23,24 This is in agreement with current evidence on choroidal thickness measurements in diabetic eyes. ^4–7 The coefficient for intragrader repeatability of choroidal measurements in the TMC was 26.9 μm and was higher when measured at the FCS (48.3 μm). This represents 12% and 20% of the mean choroidal thickness, respectively. Intragrader repeatability of choroidal thickness measurements has not been assessed in diabetic eyes, but a study using manual caliper measurements in healthy eyes found a CR of 23 μm, lower than observed in our study. ²⁴ Several factors account for this difference. First, the segmentation algorithms used were different. The software used in our study differs from single manual caliper measurements, as it quantifies a continuous length of multiple OCT images. Second, the patient cohorts from both studies were not comparable, and last, the research and statistical methodology were different between studies. The method used in this study quantifies a more representative section of the OCT image set but is likely to bring about greater variability within measurements. 

This study also examined the intergrader variability, or reproducibility of retinal and choroidal measurements. The variance ratio, which compares the LoA of both graders, was not statistically significant in all retinal and choroidal layers (Table 6). However, TMC measurements had a consistently narrower LoA in all retinal and choroidal layers compared with FCS measurements. This was unexpected, as the TMC represents a larger area than the FCS, and segmentation over a larger area should show greater variability. A possible explanation for this observation is that the choroidoscleral interface at the FCS is indistinct in diabetic patients. This may be due to shadowing from macular edema in the retinal layers, or may be resultant of pathology within the choroidal large vessel layers obscuring this interface at the FCS. 

We observed an intergrader mean difference of 9.6 μm for the total choroidal thickness at the FCS, and 8 μm at the TMC. The corresponding LoAs were 41 μm and 53 μm, respectively. In other words, subfoveal or foveal center (EDTRS area 9) total choroidal thickness measurements greater than 41 μm are likely to represent a true difference. The LoAs at the FCS were comparable in the choroidal sublayers for both Haller's (56 μm) and Sattler's layers (44 μm). This suggests that segmentation of these layers can be done with a similar precision as for total choroidal thickness. 

Only one other study has examined reproducibility of choroidal measurements in diabetic eyes. ⁷ The authors report a coefficient of repeatability (intergrader) of 28.8 μm (95% confidence interval, 24.8–32.8) for measurements at the FCS. The difficultly in making comparison between reproducibility is similar to that encountered with studies on repeatability, as discussed above. This study used manual caliper single-measurements to assess reproducibility, and although our study cohorts were similar, eyes with diabetic retinopathy, the statistical methods used to calculate reproducibility also differed. Interestingly, our measures of variability for choroidal thickness measurements, though larger at 53 μm (LoA), is less than the choroidal thinning observed in studies examining the diabetic choroid. Esmaeelpour et al. ⁴ report a mean choroidal thinning of 113 to 122 μm in patients with “no diabetic retinopathy” (No DR) and “clinically significant macular edema” (CSME); Vujosevic et al., ⁷ 99.5 μm in “proliferative diabetic retinopathy” (PDR); Regatieri et al., ⁶ 63.3 μm in “diabetic macular edema” (DME), and 69.6 μm in “treated PDR”; and Querques et al., ⁵ 118.2 μm and 71 μm in “CSME” and “No DR,” respectively. 

We also examined the effects of several factors, including severity of DR and maculopathy, age, sex, and visual acuity on the reproducibility of measurements (Table 7). Significant effects were noted only with VA. However, when the relationship VA with reproducibility was examined further on a scatter plot, no associations were identified (Fig. 8). We further investigated the effects that CSME may have in obscuring the location of foveal center, and consequently affecting measurements made at the FCS, compared with those at the TMC. The variance of choroidal thickness and volume measurements, in eyes with or without CSME, were similar in both areas (FCS and TMC), suggesting that CSME does not affect variability of measurements (Table 9). This was further verified by the Bland-Altman plots, which showed no obvious relationship between the magnitude of retinal thickness measurements, and differences between graders (Fig. 4). 

The strengths of this study include its sample size, and the methods of area and volume measurements, which quantify the entire length of scan, as compared with single-caliper measurements. One limitation of this study is that it used data from a separate ongoing clinical study. The consequences were (1) only eyes with DR were used to validate the segmentation protocol, hence, the LoA reported should be applied within the disease entity; and (2) the inclusion of eyes with more severe DR or maculopathy than would typically be observed in a clinical setting. However, this may better reflect a clinical trial setting, where patients who are receiving treatment would have a similar level of pathology. 

Research in “diabetic choroidopathy,” has historically been confined to histopathological and Doppler flow studies. ^8,25–27 Although current evidence (OCT studies) has suggested that the choroid is thinned in diabetes, the relationship of this finding to clinical parameters, such as DME and VA, is less clear. We speculate that the choroidal vasculature, especially in the foveal avascular zone may, in fact, play an important role in visual function in DME and ischemia, and serve as a useful parameter for monitoring and prognostic tool in diabetic eye disease. However, for reliable quantification of the choroid in diabetic eyes, it is important to establish the minimum difference that can be construed as “a true difference.” 

In summary, we describe segmentation protocols for Sattler's and Haller's vessel layers of the choroid, which showed good reliability, repeatability, and reproducibility within and between graders. With the increasing use of OCT in examining the choroid, the 95% limits of agreement presented in our study may be useful for facilitating studies investigating the underlying pathology in diabetic choroidopathy. 

Acknowledgments

Supported in part by the Department of Health's National Institute for Health Research Biomedical Research Centre for Ophthalmology at Moorfields Eye Hospital and University College London Institute of Ophthalmology (PAK, PJP, CAE, DAS, AT), and a grant from the Spanish Retina and Vitreous Society (Sociedad Española de Retina y Vítreo [SERV]) (JZ-V). The authors alone are responsible for the content and writing of the paper. 

Disclosure: D.A. Sim, Fight For Sight UK (F); P.A. Keane, None; H. Mehta, None; S. Fung, None; J. Zarranz-Ventura, None; M. Fruttiger, Fight For Sight UK (F); P.J. Patel, None; C.A. Egan, None; A. Tufail, None