Article Text

Download PDFPDF

Characterisation of choroidal morphological and vascular features in diabetes and diabetic retinopathy
  1. Preeti Gupta1,2,
  2. Sri Gowtham Thakku1,2,
  3. Charumathi Sabanayagam1,2,3,
  4. Gavin Tan1,2,3,
  5. Rupesh Agrawal1,4,5,
  6. Chui Ming Gemmy Cheung1,2,3,
  7. Ecosse L Lamoureux1,2,3,
  8. Tien-Yin Wong1,2,3,
  9. Ching-Yu Cheng1,2,3
  1. 1Singapore Eye Research Institute and Singapore National Eye Centre, Singapore, Singapore
  2. 2Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore and National University Health System, Singapore, Singapore
  3. 3Duke-NUS Medical School, Singapore, Singapore
  4. 4Tan Tock Seng Hospital, Singapore
  5. 5School of Material Science and Engineering, Nanyang Technological University, Singapore
  1. Correspondence to Dr Cheng-Yu Cheng, Academic Medicine Research Institute, Duke-NUS Medical School, Head, Ocular Epidemiology Research Group & Statistics Unit, Singapore Eye Research Institute, 20 College Road, The Academia, Level 6, Discovery Tower, Singapore 169856, Singapore; chingyu.cheng{at}


Aim We aimed to characterise specific morphological and vascular features of the choroid in Indian adults with diabetes and diabetic retinopathy (DR).

Methods Consecutive participants from the Singapore Indian Eye Study's 6-year follow-up examination underwent choroidal imaging using spectral domain optical coherence tomography (OCT) with enhanced depth imaging. Raw OCT images were loaded on a custom-written application on MATLAB that enabled delineation for detailed morphological and vascular analyses. Multiple linear regression analyses were performed to assess differences in choroidal characteristics by diabetes DR.

Results Of the 462 recruited participants, 273 had no diabetes (mean age was 60.1±6.8 years), 100 had diabetes but no DR (61.8±7.4 years) and 89 had DR (62.4±6.0 years). In multiple regression analysis, after accounting for relevant confounders, compared with those without diabetes, participants with diabetes had significantly thinner mean choroidal thickness (CT; mean difference (MD)=−25.19 µm, p=0.001), smaller choroidal volume (MD=−0.23 mm3, p=0.003), more inflection points (MD=1.78, p<0.001) and lesser choroidal vascular area within the foveal (MD=−0.024 mm2, p=0.001) and macular (MD=−0.095 mm2, p<0.001) regions. Among the diabetic group, subjects with DR had significantly thicker mean CT (MD=25.91 µm, p=0.001), greater choroidal volume (MD=0.24 mm3, p=0.009), lesser inflection points (MD=−0.478, p=0.045) and greater choroidal vascular area at foveal (MD=0.016 mm2, p=0.019) and macular (MD=0.057 mm2, p=0.016) regions, compared with those without DR.

Conclusions Choroidal morphology and vasculature were altered in Indian adults with diabetes and DR. These findings may provide insights into choroidal changes in diabetes and DR.

  • Choroid
  • Epidemiology
  • Imaging
  • Diagnostic tests/Investigation

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Diabetes mellitus is a metabolic disease characterised by microvascular complications.1 Diabetic retinopathy (DR) is the most common microvascular complication, primarily caused by chronic hyperglycaemia leading to loss of pericytes of retinal capillaries and subsequent ischaemia.2 The pathogenesis and clinical features of DR are primarily attributed to retinal vascular damage;3 however, there is increasing evidence of choroidal involvement as persons with diabetes have been found to have various choroidal abnormalities including choroidal vascular degeneration, choroidal aneurysms, choroidal neovascularisation, obstruction of the choriocapillaris, and increased tortuosity and narrowing of the choroidal vessels.4–6

The recent availability of enhanced depth imaging (EDI) optical coherence tomography (OCT) has allowed the non-invasive quantitative assessment of the choroid.7 Since then various studies have proposed choroidal thickness (CT) as a marker to assess different diseases including diabetes. However, reports on CT in diabetes have been contradictory with studies reporting increase,8 decrease9 ,10 or even no change.11 This could possibly be due to confounding effects from factors such as age, gender, refractive error, axial length, diurnal variation, etc which have been shown to influence CT.12–15 Thus measurements of CT alone may not be sufficient to reflect specific changes within the choroid, which is a complex, three-dimensional structure predominantly composed of interconnected network of blood vessels surrounded by stromal tissue.16 Therefore, a more detailed investigation of choroidal structure and vasculature is essential to provide a better understanding of the choroid in diseases such as diabetes and DR where to date there has been no consensus as to whether there is an increase or decrease in CT.

There have been few attempts to evaluate a few morphological and vascular features of the choroid such as the shape of the choroid scleral interface (CSI) and choroidal vessel layer (large and medium) thickness measurements in DR using OCT.17 ,18 However, these studies have not had detailed description of the morphological features such as curvature of the CSI, number of inflection points (local variations) and additional architectural features of the stromal and vascular parameters in the choroid. Given choroidal vascular dysfunction are associated with many pathological conditions including DR and in view of paucity of data on these features as well as the rising trend in the prevalence of both diabetes and DR among Asian,19 ,20 it is imperative to have detailed in vivo baseline measurements of the choroidal morphometric and vascular features in persons with diabetes and DR.

The aim of this study was to characterise the morphological and vascular features of the choroid in a sample of Indian adults with diabetes and DR using novel segmentation methods on choroidal images obtained by EDI spectral domain (SD)-OCT.

Materials and methods

Study subjects

In this cross-sectional study, 500 consecutive subjects were recruited during their 6-year follow-up examination from a total of 2200 subjects from Singapore Indian Eye Study-2 (SINDI-2, 2013–2015), a population-based study conducted at the Singapore Eye Research Institute. Details of the baseline study design, sampling plan and methodology (SINDI-1, 2007–2009) have been reported elsewhere.21

Each subject underwent an interview and detailed systemic and ocular examination according to the SINDI study protocol.21 The study protocol was approved by the SingHealth Centralised Institutional Review Board and adhered to the ethical principles outlined in the Declaration of Helsinki, 2008. Written voluntary informed consent was obtained from each subject.

Exclusion criteria

Among non-diabetic and diabetic group (including those with and without DR), subjects with refractive error <−6.00 dioptres, age-related macular degeneration (AMD), glaucoma and those whose choroidal images were of insufficient quality (quality index of <18 dB) were excluded. However, subjects with DR that had received treatments such as intravitreal injections or laser photocoagulation (more than 6 months of their participation into the study) were not excluded.

OCT image acquisition

The choroid was imaged using EDI mode of Spectralis SD-OCT (Heidelberg Engineering, Heidelberg, Germany) after pupil dilation. OCT raster scans with EDI were acquired at the macular region of both eyes of each subject. Each set of images comprised seven serial horizontal B-scans (each composed of 1536 A-scans) covering a rectangular region of 30°×5° centred on the fovea. Distance between consecutive scans was on average about 240 μm and each scan was 8.9 mm in length. To reduce speckle noise, each B-scan was averaged 75 times during acquisition.

Image delineation

Although choroidal images of both eyes of each study subject were obtained, due to strong intereye correlation only the right eye of each subject was delineated for detailed morphological and vascular analyses. Raw OCT images were loaded on a custom-written application on MATLAB that enabled delineation. Two structures were delineated: the Bruch's membrane and the CSI. The foveal centre and the Bruch's membrane were automatically delineated by the Spectralis OCT device whereas three to four most prominent points corresponding to the CSI were manually marked by a single author using our custom application.

Measurement of choroidal morphological and vascular parameters

For each eye, the horizontal cross-sectional scan passing through the fovea was identified. Based on this cross-sectional scan, we defined and calculated (in MATLAB) the following choroidal morphological and vascular parameters (figure 1).

Figure 1

(A) Visualisation of choroid morphological and vascular parameters obtained by our custom-written application on MATLAB on choroidal images acquired by enhanced depth imaging, spectral domain optical coherence tomography. (B) Validation of Niblack binarisation against manual delineation of vascular and stromal areas. Identification of choroid vasculature by automated binarisation with overlayed manual delineation of four blood vessels.

CSI curvature

When compared with a healthy choroid, eyes with DR may have irregularities in the choroidal boundary, including changes in CT. To quantify these irregularities, we used parameters such as curvature and number of inflection points at the CSI. Curvature of the CSI gives an overall measure of the CSI shape (flat, ‘U’ shaped, inverted ‘U’ shape). To measure overall curvature, the best-fitting circular arc to the marked CSI points was identified. The reciprocal of the radius of this best-fitting circular arc was reported as CSI curvature. The sign of the curvature indicated the direction of the arc (positive denotes an anteriorly curved, and negative a posteriorly curved arc).

Number of inflection points

The number of inflection point measures the extent of local irregularities. A more irregular CSI would have greater local variations and hence more inflection points. Mathematically, an inflection point is a point on a curve at which the curve changes from being concave to convex, or vice versa. An inflection point represents the contour of the CSI, with >1 point signalling irregular shape.

Thickest point distance from the fovea

CT varies within the macular region. To quantify the position of the region with the thickest choroid, the distance of this point from the foveal centre was measured.

Choroidal thickness and volume

CT was defined as the distance between the external limit of the Bruch's membrane and the CSI. The average thickness within the macular (6 mm centred on fovea) and foveal (1.5 mm centred on fovea) regions were reported as mean CT. Choroidal volume was measured by the summation the area enclosed between the Bruch's membrane and the CSI for each of the seven horizontal scans multiplied by the distance between consecutive scans.

Choroidal vascular area within foveal and macular regions

Based on the marking of the choroidal region in the horizontal scan passing through the fovea, image binarisation was used to distinguish between the vascular and stromal areas of the choroid. The Niblack local thresholding technique was used, similar to that described in previous papers.22 Black spaces within the choroid after binarisation were assumed to represent vascular regions. The area of this region was measured within the subfovea as well as the macula, and defined as vascular area.

Validation of image binarisation

The Niblack local thresholding technique used to identify stromal and luminal areas within the choroid was validated against manual delineations performed by a single person. On a set of eight randomly chosen cross-sectional scans of the choroid passing through the fovea, the three or four best visible blood vessels were delineated. Four similarly sized regions corresponding to the stroma were also marked. The automatically binarised images were compared with these manual markings. In the binarised image, black regions correspond to the lumen and white regions to the stroma. Agreement between the manual and automated delineations was assessed by overlaying manually identified blood vessels over the binarised image. Using the manual delineation as the gold standard, the area of the white pixels falling within manually delineated blood vessels was divided by the area of the manually delineated blood vessel. This percentage was reported as incorrectly identified luminal area. Likewise, the area of the black pixels outside the manually delineated blood vessels (but continuous with the inside black region) was measured and divided by the area of the manually delineated blood vessel. This percentage was reported as incorrectly identified stromal area (figure 1). We found that 2.5% of the luminal area and 4.1% of the stromal area were incorrectly identified by our automatically binarised images.

Systemic factors and diagnostic definitions

A detailed interviewer-administered questionnaire was used to collect demographic data, medical history and medication use from subjects. Non-fasting venous blood samples were analysed for biochemical testing of serum total cholesterol, glycosylated haemoglobin (HbA1c) and serum glucose level. Diabetes mellitus was defined as random glucose of 11.1 mmol/L or more, haemoglobin A1c (HbA1c) of 6.5% or more, use of diabetic medication or a physician diagnosis of diabetes.23 DR was defined according to the Early Treatment Diabetic Retinopathy Study severity scale.24 Hypertension was defined as systolic blood pressure (SBP) 140 mm Hg or more, diastolic blood pressure (DBP) 90 mm Hg or more at the time of the examination, or a reported history of physician diagnosed hypertension or self-reported history of antihypertensive medication use, or both.

Statistical analyses

To compare the characteristics of participants among groups, the independent t-test was performed for continuous variables and χ2 tests or Fisher's exact test for categorical variables. Data were expressed as mean±SD. Multiple linear regression analysis was used to assess differences in choroidal characteristics by diabetes (in reference to no diabetes) and DR (in reference to no DR) after adjusting for factors with p<0.1 in the univariate analysis as well as biologically plausible factors such as age, sex and axial length. Statistical significance was set at p<0.05 unless otherwise indicated. All statistical analyses were carried out using SPSS (V.24.0; Chicago, Illinois, USA).


Of the 500 subjects initially enrolled, 37 were excluded based on our clinical exclusion criteria; myopia <−6.00 D (n=6), AMD (n=4), glaucoma (n=13) and poor OCT image quality (n=14), leaving 463 subjects for analysis. Of the 463 subjects included, 273 had no diabetes, 100 had diabetes but no DR and 89 had DR.

The demographic, clinical and choroidal characteristics of the subjects with and without diabetes are shown in table 1. There were significant differences in characteristics (clinical and choroidal) between the two groups. Total choroidal area, vascular area and choroidal vascularity were lower in subjects with diabetes compared with non-diabetics (all p<0.001). There was no significant difference in choroidal stromal area between the two groups. Of the subjects, 86.4% without diabetes had one inflection point versus 24% of those with diabetes.

Table 1

Clinical and choroidal characteristics of study subjects in the non-diabetic and diabetic groups

In the multiple regression analysis after accounting for relevant confounders such as age, gender, SBP, DBP, ocular perfusion pressure, HbA1c and presence of hypertension (model 2), subjects with diabetes had significantly thinner CT at all the studied locations (all p<0.001), including thinner mean CT (mean difference (MD)=−25.19 µm, p=0.001), lesser choroidal volume (MD=−0.23 mm3 p=0.003) and more inflection points (MD=1.78, p<0.001) compared with those without diabetes (table 2). They also had lesser total choroidal area, choroidal vascular area and choroidal vascularity within the foveal and macular regions.

Table 2

Differences in choroidal characteristics (dependent variables) in subjects with diabetes (n=189)and those without diabetes (n=273)

Of the diabetics, subjects with DR had significantly thicker choroid, greater choroidal volume, more posteriorly curved CSI and lesser inflection points (all p<0.05). Total choroidal, vascular and stromal areas within foveal and macular regions were significantly greater in those with the DR compared with those without (table 3, figure 2).

Table 3

Clinical and choroidal characteristics of study diabetic subjects without and with DR

Figure 2

Comparison of choroidal thickness, vascular and stromal areas across eyes with and without diabetic retinopathy. (A) Diabetic eye with diabetic retinopathy with thick choroid and high vascular and stromal areas. Choroidal thickness: 370 µm; vascular area: 0.99 mm2; stromal area: 1.02 mm2. (B) Diabetic eye without diabetic retinopathy with thin choroid and low vascular and stromal areas. Choroidal thickness: 175 µm; vascular area: 0.45 mm2; stromal area: 0.55 mm2.

In the multiple regression analysis after adjusting for potential confounders such as age, gender, axial length, SBP, HbA1c, duration of diabetes and insulin use (model 2), compared with diabetic subjects with no DR those with DR exhibited significantly thicker CT at all the studied locations (all p<0.001) including thicker mean CT (MD=25.91 µm, p=0.001), greater choroidal volume (MD=0.24 mm3, p=0.009) and lesser inflection points (MD=−0.478, p=0.045). In addition, they demonstrated greater total choroidal, vascular and stromal areas at foveal and macular regions (all p<0.05). However, at the macular region, choroid was less vascular in subjects with DR (table 4).

Table 4

Difference in choroidal characteristics (dependent variables) in diabetic subjects with DR (n=89) and those without DR (n=100)

DR subjects were classified according to their stage of DR. Of the 89 subjects with DR, 58 (65.1%) had minimal to mild DR, 24 (26.9%) had moderate non-proliferative DR, while none of our subjects had severe non-proliferative DR and 7 (7.8%) subjects had proliferative DR. Comparison of choroidal traits across different DR severity groups showed no significant trend in any of the choroidal attributes (p for trend >0.05 for all traits, see online supplementary table S1).

Given laser treatments in particular pan retinal photocoagulation (PRP) have been shown to result in choroidal thinning,25 we examined the effect of any previous laser treatments in diabetic subjects with DR. However, since two-thirds of our subjects had only minimal to mild DR, the number of subjects with laser treatment is also limited. Of the 89 subjects with DR, 74 (83.14%) had no previous laser treatment, 13 (14.6%) subjects had undergone focal laser and only 2 (2.24%) had PRP done. Comparison of choroidal features between subjects who had previous laser treatments (focal or PRP) with those with no laser treatment showed no significant difference in any of their choroidal traits (all p>0.05, see online supplementary table S2). Similar results were observed when further adjusted for DR stages.


This study characterised the detailed morphology and vasculature of the choroid in persons with diabetes and DR using the technique of image binarisation in choroidal images obtained using EDI SD-OCT. Our results demonstrated significant alterations in choroidal structural and vascular characteristics in both diabetes and DR. As speculated, subjects with diabetes showed significantly thinner choroid, lower vascular area and reduced vascularity, which is consistent with capillary drop-out in which decreased choroidal blood flow plays a key role in the pathogenesis. Conversely, among the diabetic group, subjects with DR displayed thicker choroid and significantly higher vascular area compared with no DR. These findings are useful in further understanding the role and contribution of choroidal measures in the pathogenesis of diabetes and DR.

We demonstrated that in persons with diabetes there is a significant increase in inflection points indicating more irregular contour of the CSI. This finding is supported by the clinical observation of focal changes in the choroid of diabetic persons.26 In terms of CT, our findings are consistent with the existing studies which reported choroidal thinning in diabetes.9 ,10 Distribution profiles of CT in persons with diabetes are similar to that in non-diabetic eyes (thickest: subfoveally; thinnest: nasally). Our results are congruent to what was observed in few other studies of horizontal variation in CT among subjects with diabetes.27 ,28

Interestingly, in comparison with non-diabetic eyes, choroidal vascular area was markedly reduced in persons with diabetes. In view of contradictory results of current studies on CT in diabetes8–11 it is still uncertain how diabetes affects the choroid. Because diabetes is primarily a vascular disease affecting both retinal and choroidal vasculature, knowledge of individual components of the choroid in particular vascular area may provide a better measure of the influence of diabetes on the choroid. We speculated reduction in luminal area in diabetes as there is narrowing of choroidal arterioles, choriocapillaris atrophy and capillary drop-out (shown by histological studies),5 ,29 which in turn lead to a decreased choroidal blood flow in diabetic eyes (shown by circulatory studies using Laser Doppler Flowmetry).30 The results of this study support our hypothesis of reduced choroidal vascularity in diabetic subjects.

On the other hand, compared with subjects without DR, persons with DR showed a decrease in inflection points indicating lesser local variations and therefore more regular CSI contour. Although there was a displacement of the thickest point of choroid from under the foveal centre towards temporal region, the displacement was lesser compared with that seen in subjects with diabetes and was not significant. Contrary to diabetes (without DR), there was a significant increase in CT (at all the locations), choroidal volume and vascular area within foveal and macular regions in DR. We hypothesise that the increase production of vascular endothelial growth factor (VEGF) or other cytokines in DR might result in choroidal vasodilation and elevation in choroidal blood flow, which subsequently increase the thickness and vascular area of the choroid. Nonetheless, recent innovations in technology such as longer wavelength OCT31 including swept source technology32 along with OCT angiography and en-face imaging33 may further improve delineation of microstructural changes and morphology of the retinal and choroidal vasculature to better understand choroidal involvement in retinal diseases.

There was no significant difference in any of the choroidal traits in DR subjects with or without laser treatment. We speculate the reason for no difference in choroidal features across the two groups to be the lesser number of subjects in treatment group compared with no treatment. It is more likely a power issue and we recommend future studies with more number of subjects in treatment group to study the likely effect of lasers on choroidal attributes.

The strengths of our study included a relatively large sample size with a single common ethnicity, standardised clinical examination protocols, reliable differentiation and quantifications of choroidal parameters. Furthermore, our analysis included a comprehensive list of potential confounding factors; choroidal characteristics in different groups were confirmed after adjusting for effect of these factors. However, this study has also limitations. Our OCT images were binarised at standard threshold, yet there was a possibility of over or under estimation of both vascular and avascular areas. Choroidal attributes were assessed from a horizontal cross-sectional scan passing through the fovea. Future studies using more dense volume scan protocols are warranted. Also future improved software will be able to perform binarisation of volume scans and will provide data not limited to single scan. We did not have information on prior intravitreal anti-VEGF injection for our diabetic subjects with DR. However, given that two-thirds of our subjects had only minimum to mild DR, it is unlikely to influence our findings. DR subjects were classified according to their stage of DR. However, the statistical power to perform subgroup analysis by stage may be limited. Future studies with large sample in different DR severity groups are warranted. Choroidal scans were only performed for one time point in the day. The effect of diurnal variation on choroidal features13 ,34 could not be assessed due to the inherent limitation of our study design.

In conclusion, we report significant alterations in choroidal morphometric and vascular attributes in persons with diabetes and DR. These findings will be useful in further understanding the role and contribution of choroidal measures in the pathogenesis of diabetes and DR.



  • Correction notice This article has been corrected since it was published Online First. Rupesh Agrawal has been added to the author list and his affiliations and Contributions have been added to the paper.

  • Contributors PG and C-YC had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. CS and C-YC: study concept and design. All authors: acquisition, analysis or interpretation of data; critical revision of the manuscript for important intellectual content; and administrative, technical, or material support. PG, SGT and C-YC: drafting of the manuscript. PG and C-YC: statistical analysis. CS, T-YW and C-YC: obtained funding. C-YC: study supervision. RA: analysis of data and technical support.

  • Funding This study was supported by a grant from National Medical Research Council, Singapore [grant number NMRC/CIRG/1371/2013]. The sponsor or funding organisation had no role in the design or conduct of this research.

  • Competing interests CMGC: independent consultant for Bayer and Novartis; T-YW: received grant support from National Medical Research Council (NMRC) and Biomedical Research Council, Singapore directed to Singapore Eye Research Institute, Advisory Board member for Abbot, Novartis, Pfizer, Allergan, and Bayer, independent consultant for Abbot, Novartis, Pfizer, Allergan and Bayer; C-YC: received grant support from NMRC and Biomedical Research Council, Singapore directed to Singapore Eye Research Institute.

  • Patient consent Obtained.

  • Ethics approval SingHealth Centralised Institutional Review Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Linked Articles