Three-dimensional modelling of the choroidal angioarchitecture in a multi-ethnic Asian population

This study aimed to describe the topographic variation of the macula’s choroidal angioarchitecture using three-dimensional (3D) choroidal vascularity index (CVI) of healthy eyes from an Asian population and to investigate the associations of CVI. 50 participants were recruited via stratified randomisation based on subfoveal choroidal thickness from the Singapore Epidemiology of Eye Diseases Study. Macular volume scans were acquired using spectral-domain optical coherence tomography with enhanced depth imaging. CVI was assessed based on B-scan binarisation and choroid segmentation. The 3D CVI of the whole, superior, central, and inferior macula were 62.92 ± 1.57%, 62.75 ± 1.93%, 63.35 ± 1.72%, and 62.66 ± 1.70%, respectively, pairwise comparisons P all > 0.05). 3D CVI (Whole Macula) and 2D CVI (Subfoveal) were associated only with each other and not with other ocular and systemic factors. 2D CVI (Subfoveal) had a moderate agreement with 3D CVI (Central Macula) [intraclass corelation coefficient (ICC) = 0.719], and had poorer agreement with 3D CVI of the whole macula, superior, and inferior macula (ICC = 0.591, 0.483, and 0.394, respectively). Scanning volume did not influence 3D CVI measurements. In conclusion, 3D CVI demonstrated no significant topographic variation. CVI was not correlated with demographic or ocular structural features. 2D CVI of the fovea is partially representative of 3D CVI of the macula.

To the best of our knowledge, there has not been a head-to-head comparison of 3D and 2D CVI measurements. There is also a need to see whether scanning volume influences 3D CVI measurements. While it has been reported that 3D CVI of the macula does not exhibit significant topographic variation, it is unknown if this holds true across a range of CT, especially in thicker choroids 18,20,21 . Whether 3D CVI of the macula in eyes with thicker choroids is different has not been investigated. This is important considering the prevalence of the pachychoroid phenotype among Asians 22,23 .
To these ends, we sought to describe the topographic variation of choroidal angioarchitecture of the macula using 3D CVI across a range of CT in healthy eyes from an Asian population and to investigate the associations of CVI.

Methods
Study population. The Singapore Epidemiology of Eye Diseases (SEED) Studies comprise prospective population-based cohort studies of Malays, Indians, and Chinese in Singapore [24][25][26][27] . The study design has been described in detail previously. The studies are still ongoing. The data for this study was derived from the SEED-2 Study, which is a six-year follow-up study of SEED-1 14 . Stratified sampling by mean subfield subfoveal choroidal thickness (SFCT) was performed by dividing the SFCT data into quintiles and by randomly selecting 10 participants from each quintile. This produced 50 participants who were selected for analysis in this study. This study adhered to the tenets of the Declaration of Helsinki. Ethics approval was obtained from the Institutional Review Board of the Singapore Eye Research Institute. Written informed consent was obtained from all participants.
OCT. The Spectralis (Heidelberg Engineering, Heidelberg, Germany) was used to image the choroid under standardised mesopic lighting conditions. After pupil dilation with tropicamide 1% and phenylephrine hydrochloride 2.5%, spectral-domain OCT (SD-OCT) raster scans with enhanced depth imaging (EDI) were acquired on a 30° × 30° macular region centred on the fovea in the High Speed mode, with 31 B-scans per volume scan. Each B-scan was averaged [Automatic Real-time Tracking (ART) mode] using 75 frames, and the distance between consecutive B-scans ranged from 233 to 244 μm. All SD-OCT scans were reviewed by trained graders from the centralised grading centre to ensure that the scans were of sufficient clarity. OCT scans with a signal strength of ≥ 20 were eligible for the analysis. Images with poor focus, motion artifacts, and/or obscure choroidscleral interface were excluded. CVI measurement. To measure CVI, the public domain software ImageJ (Version 1.53c, National Institutes of Health, Bethesda, Maryland, USA) 28,29 was used to perform image binarisation using techniques that were described by Agrawal et al. 1,2,17 Image binarisation converts grey scale images into binarised images. An appropriate image binarisation technique is essential to accurately apply a threshold to an image and takes into account illumination, contrast, and resolution of the image pixels.
For 2D CVI (Subfoveal), the foveal OCT B-scan was used. Total choroidal area (TCA) was selected using the polygon tool with the upper border at the RPE-Bruch's membrane complex and the lower border at the choroid-scleral interface, and added into the Region of Interest (ROI) Manager. Niblack's auto local thresholding was applied after image conversion into eight bits. This generated the mean pixel value with standard deviation for all points. Image conversion to the Red, Green and Blue (RGB) format was performed to allow the colour threshold tool to select the dark pixels. The TCA and the area of dark pixels, which corresponded to the luminal area (LA), were calculated. To determine the LA within the selected polygon, both the areas in ROI Manager were selected and merged by the "AND" operation of ImageJ. The composite third area was added to the ROI Manager. The first area represents the TCA selected, and the third composite area is the LA. 2D CVI (Subfoveal) was then calculated by dividing LA by the TCA.
3D CVI was subsequently calculated by integrating the 2D CVI across the scan volume. TCA and LA were measured for every OCT B-scan in the volume scan which comprised 31 B-scans. Essentially, choroidal volume (be it the total volume, or that of the stromal or luminal components) in mm 3 is the summation of the mean area in each B-scan in mm 2 multiplied by the distance between two consecutive B-scans in μm. 3D CVI was calculated by dividing the luminal volume by the total choroidal volume. This was separately applied for the whole macula (all 31 B-scans), superior macula (superior 10 B-scans), central macula (central 11 B-scans), and inferior macula (inferior 10 B-scans). See Fig. 1 for a pictorial representation of the locations in the macula where the various CVI measurements were taken.
The following steps were applied: 1. Area in the n th OCT B-scan (Area n ) (for both TCA and LA) in pixel 2 in each B-scan were converted to mm 2 by multiplying the area by the scaling factors in the x and y axes:  CT measurement. CT was automatically measured as the intervening distance between the RPE-Bruch's membrane complex and the choroid-scleral interface. The slab was segmented on every B-scan of the Spectralis volume scan using the manufacturer's own algorithm, Heyex SP-X version 6.4.8.116 (exclusively available in the specific research; Heidelberg Engineering). The Heyex software generated a regional CT map with a Early Treatment of Diabetic Retinopathy Study (ETDRS) grid that described the mean subfield CT in all nine ETDRS grid subfields from the 31 horizontal raster B-scans. The B-scans were checked for segmentation artifacts and errors. Misalignment was manually corrected as needed.
Systemic risk factors. Trained interviewers used standardised questionnaires to collect demographic information, medical history, and lifestyle factors 30 . Body mass index (BMI) was computed by dividing the body weight (kg) by the square of body height (m). Blood pressure was measured with a digital automatic monitor (Dinamap model Pro Series DP110X-RW, 100V2; GE Medical Systems Information Technologies Inc., Milwaukee, Wisconsin, USA) after five minutes of rest with the participants seated. Hypertension was defined as systolic blood pressure (SBP) of ≥ 140 mmHg, diastolic blood pressure (DBP) of ≥ 90 mm Hg, physician-diagnosed hypertension, or a self-reported history of hypertension.
Non-fasting venous blood was collected for glycosylated hemoglobin (HbA1c), lipid levels, including total cholesterol, high-density lipoprotein (HDL), and low-density lipoprotein (LDL), and serum creatinine. Hyperlipidemia was defined as total cholesterol of ≥ 6.2 mmol/L, or a self-reported use of lipid lowering drugs. Diabetes mellitus was defined as a random glucose level of ≥ 11.1 mmol/L, HbA1c of ≥ 6.5%, use of diabetes mellitus medication, or a physician-diagnosed history of diabetes mellitus. Renal function was assessed using estimated glomerular filtration rate (eGFR) from serum creatinine using the Chronic Kidney Disease-Epidemiology Collaboration equation 31,32 . Chronic kidney disease was defined as an eGFR of < 60 ml/min/1.73 m 2 . Cardiovascular disease was based on a self-reported history of myocardial infarction, angina, or stroke.
Exclusion criteria. Eyes with BCVA of worse than 20/60, evidence of ocular diseases, or history of ocular surgery were excluded. Eyes with other ocular conditions, such as refractive errors, were not excluded if the OCT scan quality was sufficient to be evaluated.

Statistical analysis.
Statistical analyses were performed using R version 4.1.0 33 . Continuous variables were summarised using means, standard deviations (SD), and 95% confidence intervals (CI), and categorical variables were summarised with counts and percentages. Since CVI and SFCT have different measurement units, we used the coefficient of variation (COV) (%) to compare variability between CVI and SFCT. The normality of the data was assessed with the Shapiro-Wilk test. The comparison of ocular and systemic characteristics among groups www.nature.com/scientificreports/ was performed using analysis of variance for continuous variables and chi-square tests for categorical variables. Univariate linear regression analyses were performed to examine the associations among ocular and systemic factors with SFCT and CVI. Age and factors that were significant in the univariate analyses (P < 0.05) were included in the multivariable linear regression model. Only one eye from each participant was used. Agreement was assessed using intraclass correlation coefficients (ICC) (two-way random effects, absolute agreement, and single rater/measurement) 34 , Bland-Altman analysis, and F-test for equality of variances. For ICC, values less than 0.5 indicate poor reliability, values between 0.5 and 0.75 indicate moderate reliability, values between 0.75 and 0.9 indicate good reliability, and values greater than 0.90 indicate excellent reliability 34 .

Results
The ocular and systemic characteristics of the 50 participants are described in Table 1. CVI was similar across the SFCT quintiles. Overall, the 3D CVI of the whole macula was 62.92 ± 1.57%. The 3D CVI of the superior, central, and inferior macula were 62.75 ± 1.93%, 63.35 ± 1.72%, and 62.66 ± 1.70%, respectively. The 2D CVI (Subfoveal) was 63.53 ± 2.09%. The mean SFCT was 288.3 ± 85.6 µm. There were no significant differences in any characteristics among the SFCT quintiles except for proportion of smokers (P = 0.037).
Topographically, there were no significant differences among the 3D CVI of the superior, central, and inferior macula (pairwise comparisons P all > 0.05).
Lastly, the influence of scanning volumes on CVI was assessed. See Supplementary Fig. 1 for a pictorial representation of the regions in the macula that were compared. The ICC, Bland-Altman analysis, and F-test for equality of variances are shown in Table 4. The 3D CVI measurements showed good to excellent agreement regardless of the number of B-scans used from the volume scan. The ICC was more than 0.85 for all three comparisons (     Notably, the adjusted R-squared values of the multivariable analyses were higher for 3D CVI (Whole Macula) and 2D CVI (Subfoveal), compared with that for SFCT. This indicated that the assessed factors accounted for more of the variation for CVI compared with SFCT.
Both 3D and 2D CVI were not significantly related to age. This agrees with previous studies by Agrawal et al. 17 19 .
To the best of our knowledge, there has not been a head-to-head comparison of 3D and 2D CVI measurements in the posterior pole of healthy eyes. 2D CVI (Subfoveal) had only a moderate agreement with 3D CVI in the central macular region of 11 B-scans. The agreement of 2D CVI (Subfoveal) with 3D CVI (Whole Macula), 3D CVI (Superior), and 3D CVI (Inferior) were poorer. The Bland-Altman analysis showed a similar picture, www.nature.com/scientificreports/ with a smaller mean difference and a narrower 95% limit of agreement for 2D CVI (Subfoveal) versus 3D CVI (Central), compared with 2D CVI (Subfoveal) versus 3D CVI of the whole, superior, and inferior macula. In this regard, 2D CVI of the fovea is only partially representative of 3D CVI of the central macula in healthy eyes. Our results reflect differences in 2D and 3D CVI measurements. The latter involve incorporation of interscan distances and scaling factors in the x and y axes. Also, that the 2D CVI (Subfoveal) showed a better agreement with the 3D CVI of the central macula region as opposed to that of the whole macula and of the other regions reflects subtle topographic differences across the macula. Therefore, CVI of a single 2D B-scan may not be a representative indicator of the choroidal vasculature. The choroid vascular status should be assessed using 3D information of the choroid rather than using a 2D scan. The assessment of 3D CVI may be particularly useful in understanding the changes in the overall and regional choroidal vasculature in diseases involving part of the choroid or of the entire choroid, e.g., age-related macular degeneration, diabetic retinopathy, inherited retinal diseases, posterior uveitis, etc. The regional topographical variations in the 3D CVI can potentially allow novel pathomechanisms of the disease, and accordingly, selective pharmacotherapy, to be identified.
There was also an indication that in the pairwise comparison between 2D CVI (Subfoveal) and 3D CVI (Whole Macula), the proportionate difference [given by 2D CVI (Subfoveal)-3D CVI (Whole Macula)] increased as the 2D CVI (Subfoveal) increased. This is an interesting observation. For choroids with a greater CVI, one should be mindful that 3D CVI measurements of the whole macula may diverge from 2D CVI of the subfoveal region.
Lastly, the comparison of 11, 21, and 31 B-scans showed that there was no significant influence of the scanning volume on the 3D CVI measurements of the macula. The ICC was more than 0.85 (good to excellent) for all three comparisons. The Bland-Altman analysis showed low mean differences and narrow 95% limits of agreement. Reducing the number of OCT B-scans is appealing to increase cost-effectiveness and increase patient compliance during the OCT scan.
However, the agreement was still better for 21 versus 31 B-scans compared with 11 versus 21 B-scans. It is logical that measuring 3D CVI over a large macular area will produce measurements closer to the true measurement. This is also more essential in diseased eyes if it has a localised pattern. Nonetheless, this comes at the expense of a longer scanning time and more effort being spent on data and image processing.
The key strength of this study lies in the assessment of 3D CVI from the entire volume scan, which is a more representative and relevant approach for the assessment of the choroidal vasculature and architecture, as compared with an assessment of a single subfoveal scan or of the average of selected B-scans across the macula. The novel analyses include the head-to-head comparison of 2D and 3D CVI measurements and the assessment of scanning volume on 3D CVI measurements. The data set was obtained via stratified sampling from the SFCT quintiles of a population-based study and has a wide representation of CT profiles. Standardised clinical examination protocols were used in our study.
This study has limitations. Firstly, a limited number of healthy eyes were studied, and even fewer in each SFCT quintile. The study is underpowered to perform comparisons of 3D CVI among the SFCT quintiles and among the macular regions. We were not able to perform sample size calculations a priori for this study. Using the means and standard deviations of the differences in 3D CVI (Whole Macula) among the different SFCT quintiles, approximately 100 eyes in each quintile will be required to achieve a power of 80% and a level of significance of 5% in two-sided tests. Nonetheless, we hope that this will help guide sample size considerations in future studies. Secondly, the topographic variation and the associations of 3D CVI may not apply to diseased eyes, especially if the disease process is localised. It will be interesting to assess 3D CVI in diseased eyes. Thirdly, the CVI measurements were time-consuming and laborious. Therefore, clinical utility may be limited. For 50 participants with 31 B-scans in each volume scan, a total of 1550 B-scans were manually segmented. By applying automated algorithms, these processes can be performed automatically and expediently. Zhou et al. have described that automated assessments of the choroid have good agreement with manual segmentations 19 .
In summary, this study has described the topographic variation of choroidal angioarchitecture and associations of 3D CVI in healthy eyes. CVI was not correlated with demographic or ocular structural features. 2D CVI of the fovea is partially representative of 3D CVI of the macula in healthy eyes, thus the choroid vascular status should be assessed using 3D information of the choroid rather than by a 2D scan. Lastly, there is no significant influence of scanning volume on 3D CVI measurements of the macula.