Article Text

Download PDFPDF

Identification of imaging features that determine quality and repeatability of retinal capillary plexus density measurements in OCT angiography
  1. Beau J Fenner1,
  2. Gavin S W Tan1,2,
  3. Anna C S Tan1,2,
  4. Ian Y S Yeo1,3,
  5. Tien Yin Wong1,2,3,4,
  6. Gemmy C M Cheung1,2
  1. 1 Vitreo-Retinal Service, Singapore National Eye Centre, Singapore
  2. 2 Retina Research Group, Singapore Eye Research Institute, Singapore
  3. 3 Ophthalmology and Visual Sciences Academic Clinical Research Program, Duke-NUS Medical School, National University of Singapore, Singapore
  4. 4 Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore
  1. Correspondence to Dr Gemmy C M Cheung, Singapore National Eye Centre, 11 Third Hospital Ave, 168751 Singapore; gemmy.cheung.c.m{at}


Purpose Optical coherence tomography angiography (OCT-A) potentially allows for rapid and non-invasive quantification of retinal capillary plexus density in various disease states. This study aims to identify the key features that influenced the repeatability of OCT-A measurements.

Methods We obtained OCT-A images on two separate visits in 44 healthy eyes from 44 subjects, each imaged with using the Topcon DRI OCT Triton imaging system. The parafoveal vessel density within a 1.5 mm radius centred over the fovea was obtained with the built-in tool for the superficial and deep retinal plexuses. Repeatability of vessel density was determined by intraclass correlation (ICC) and mean variation. We evaluated several image parameters to determine their influence on the repeatability of vessel density measurement in each of the two capillary plexuses.

Results The mean age of the subjects was 70.2±9.2 years, with 64% males. Mean parafoveal vessel density measurements for the first and second visits were 53.3±11.1 and 53.3±10.3 for the superficial plexus and 27.3±8.59 and 27.0±8.78 for the deep plexus. ICC analyses demonstrated that high fine vessel visibility, the absence of motion artefact and software-derived image quality score of 60 or above were necessary to obtain a good (ICC>0.6) or excellent (ICC>0.75) repeatability.

Conclusions Our study identified the imaging parameters that determined the repeatability of quantitative retinal vessel density measurements. These findings have implications in determining if OCT-A images can be used to accurately evaluate serial changes in retinal vessel density.

  • Retina
  • Macula

Statistics from


Optical coherence tomography angiography (OCT-A) is a novel, non-invasive method for examining the microvasculature within the eye.1–4 The utility of this technique as an alternative to conventional fluorescein and indocyanine green angiography has been explored for numerous retinal and optic nerve diseases in the past few years with encouraging results.5–14 In addition to qualitative features (eg, subjective assessment of foveal ischaemia), several studies have now reported the use of quantitative measures (eg, retinal capillary plexus density measurements) to evaluate severity and prognosis of conditions such as diabetic retinopathy,15 age-related macular degeneration,16 Stargardt disease17 and retinitis pigmentosa.18

As with any new instrument, the interpretation of OCT-A is affected by variations and artefacts from several sources, including physiological and anatomical variation of the eye, media opacities, image acquisition, eye movement, data processing and image display.19 Key types of artefacts include motion artefacts such as displacement, stretch and gap defects, projection artefacts from vessel reflections and image saturation.19 The accuracy of quantitative parameters, such as retinal capillary plexus density, can be affected adversely by images of suboptimal resolution. For instance, the presence of media opacities or poor ocular surface can reduce the resolution of retinal vascular structures, while tilting or decentration of the scanned area around the fovea can similarly interfere with accurate capillary density measurements.20

In the assessment scan quality, most recent work reporting on the use of OCT-A to quantify non-perfusion or vascular density21–24 use qualitative parameters to exclude suboptimal scans from analyses.5 22 25 These include removal or scans with artefacts, media opacities and segmentation errors,22 25 although the impact that these measures have on OCT variability has not previously been reported.

The repeatability of foveal avascular zone (FAZ) area measurements using OCT-A has been addressed by several groups recently with impressive results. Intraclass correlation coefficients (ICCs) exceeding 0.9 for FAZ area measurements in healthy subjects were reported by several groups.22 26 27 Fewer studies have evaluated the repeatability of retinal capillary density measurements. Agemy et al 28 evaluated retinal capillary density in a group of five healthy subjects with ages ranging from 52 to 58 and found a coefficient of variation of less than 7%. Iafe et al 29 found that capillary density declined while the FAZ area expanded with age.

In the current study, we sought to identify features of OCT-A images that significantly influence the repeatability of retinal capillary plexus density measurements. As OCT-A becomes more widely incorporated into clinical practice, such information will become important for deciding if multiple scans from a single patient can be reliably compared over time.


Recruitment and data collection

We performed a prospective study which utilised data collected from 44 eyes with no clinical evidence of retinal disease from 44 subjects who attended the Singapore National Eye Centre between 2015 and 2016. The study was approved by the SingHealth Centralised Institutional Review Board and was conducted as per the tenets of the Declaration of Helsinki. Written informed consent was obtained from each participant. OCT-A scans were acquired using the DRI OCT Triton system (Topcon, Tokyo, Japan) on two consecutive clinic visits. No evidence of retinal pathology was confirmed on dilated fundus examination between the two study visits. The DRI OCT Triton system uses a swept source laser with a centre wavelength of 1050 nm and scan speed of 100 000 A-scans per second. The OCT-A is based on Topcon OCT angiography ratio analysis (OCTARA) algorithm. An active eye tracker was used to reduce motion artefact during OCT-A imaging. Each 3×3 mm volume scan contains 320×320 pixels. For this study, we used the automated layer segmentation for superficial plexus (2.6 µm below internal limiting membrane to 15.6 µm below the junction between inner plexiform and inner nuclear layers (IPL/INL) and deep plexus (15.6 µm below IPL/INL to 70.2 µm below IPL/INL). Images were analysed using Triton software suite, V.10. Capillary density values were obtained by applying an ETDRS grid overlay containing the two inner rings of the ETDRS grid pattern to fovea-centred OCT-A images. Parafoveal retinal capillary densities from each of the four quadrants were averaged to obtain a mean vessel density used in subsequent analyses. Software-directed grid centration about the fovea was accurate by visual inspection for most patients. Scans that were not well centred were retained in the data set to enable analysis of this variable. The software generated TopQ image quality values for each OCT-A scan and vessel density for each layer.

For assessment of scan quality, we graded six prespecified scan parameters as good or poor per criteria defined in table 1. Images illustrating each of the scan parameters are shown in figure 1. Scan parameters were chosen based on previous work19 and the authors’ clinical experience with OCT-A. Each parameter was graded by two readers (BF and GC).

Table 1

Definitions for optical coherence tomography angiography (OCT-A) quality assessment

Figure 1

Representative images showing good or poor scan features for each of the optical coherence tomography angiography (OCT-A) scan parameters measured. Good visibility of the fine retinal capillary network is shown in (A), while poor visibility is shown in (B). Good scan centration is shown in (C) and poor centration in (D). Motion artefact is absent in (E), but present throughout more than 10% of the scanned area in (F). The retinal pigment epithelial and outer segment layers are visible throughout the length of the OCT image in (G), but are indistinct at several points in image (H). The OCT scan in (I) is tilted at 1° (ie, <5°) from the horizontal, while the scan in (J) is tilted 5° from the horizontal.

Statistical analysis

ICC values were calculated using MedCalc V.14 with the same rater for each variable and the absolute agreement option. ICC values were graded according to the methods outlined in Cicchetti.30 Bland-Altman plots were plotted using MedCalc. Two-tailed Student’s t-tests were performed using Microsoft Excel V.2016.


Characteristics of the study population are summarised in table 2. The average age of the participants was 70.2 years and 63.6% were males. There was a mean of 105±40 days between scans. Clinical examination confirmed there was no evidence of retinal disease on each of the two visits for all participants.

Table 2

Characteristics of the study population

Retinal capillary plexus density was greater in the superficial plexus compared with the deep plexus (figure 2), in agreement with previous works.31 However, we noted a marked variation in a normal distribution in the range of vessel densities between different individuals (figure 2). There was numerical decrease in the density of both superficial and deep plexuses with increasing age, although this did not reach statistical significance. There was also no significant association between capillary density and gender, hypertension or hyperlipidaemia (see online supplementary table 1). Bland-Altman plots were prepared to visualise the agreement between scans performed at the two different clinic visits (figure 3). We noted that variability of deep plexus density was particularly high when mean vessel density was <20%. The mean differences in superficial and deep plexus densities between clinic visits were 16.4%±15.1% and 31.7%±35.0%, respectively.

Supplementary Table 1

Figure 2

Variation in vessel density between scans for the superficial and deep retinal capillary plexus. Horizontal bars indicate the means and SD.

Figure 3

Bland-Altman plots showing the variation in parafoveal retinal capillary plexus density measurements between different optical coherence tomography angiography scan sessions for the superficial (A) and deep (B) plexus. The solid line indicates the mean difference and the dashed lines indicate the coefficient of repeatability.

We next assessed the effect of various OCT-A scan parameters on the between-visit variation in retinal capillary plexus density. For the superficial plexus, between-scan agreement in vessel density was excellent when motion artefact was absent (ICC 0.901) compared with scans with motion artefact (ICC 0.419) (table 3). For the deep plexus, between-scan agreement in vessel density measurements was also significantly better in eyes without significant motion artefacts (ICC 0.943 vs 0.344). In addition, lack of fine vessel visibility and low image quality signal also affected between-scan variation (table 3). A TopQ image quality cut-off value of ≥60 was chosen because it was associated with a dramatic improvement in the repeatability of deep plexus density measurements, although it did not affect the superficial plexus measurements (table 4). We observed that variations in centration and image tilt did not affect the scan repeatability of either the superficial or deep plexus. The significance of the improvements was assessed by performing Student’s t-tests for the mean differences between scans for those with good and poor quality parameters. Fine vessel visibility, B-scan quality and the TopQ scores all significantly impacted the mean differences in vessel density between good and poor scans (table 3). Sample scan images for patients with good and poor scan quality parameters are displayed in figure 4.

Table 3

Impact of optical coherence tomography angiography scan features on variability of parafoveal capillary plexus density between scans performed during different clinic visits. Values shown are the intraclass correlations (ICCs) with the 95% CIs shown in parentheses. Significance levels were derived from two-tailed Student’s t-tests comparing the mean variation in plexus densities between scan sessions for good and poor scan parameters

Table 4

Impact of the TopQ image quality parameter on variability between scans during different clinic visits. Values shown are the intraclass correlations (ICCs) with the 95% CIs shown in parentheses

Figure 4

Sample optical coherence tomography angiography images taken from patients with good and poor scan quality parameters. Panel A–C images had good scan quality parameters, with a high level of fine vessel visibility, no or minimal motion artefact and good centration of the foveal avascular zone. The TopQ scores for these scans were over 65. Panel D shows poor fine vessel visibility, while panel E shows a scan with multiple motion artefacts. The scan in panel F suffers from poor fine vessel visibility, motion artefact and poor centration. TopQ scores for these scans were 45, 66 and 43 for panels D–F, respectively.

We additionally examined the relationship between repeatability of capillary vessel density and the variations in visual acuity noted between clinic visits. Correlation coefficient values were insignificant for both the superficial (r=0.01) and deep plexus (r<0.10), which was perhaps unsurprising given absence of clinical pathology in the eyes being examined.

Application of the quality control criteria to our data set improved the mean variation in plexus density measurements between clinic visits from 16.4±15.1% to 8.69±8.52% for the superficial plexus and from 31.7±35.0% to 6.56±6.48% for the deep plexus.


The non-invasive nature of OCT-A imaging makes it a potential breakthrough technology to collect data from large populations for screening and diagnostic purposes and to support the potential development of automated grading through deep learning technology by analysing large numbers of images.

In this regard, the assessment of quality and repeatability is critical for the use of OCT-A. Several previous studies have indicated that OCT-A has potential for excellent reproducibility and repeatability for quantifying retinal capillary density and foveal avascular zone,21 22 24 25 although few studies have clearly defined how the quality of scans was appraised. Shahlaee and others25 provided relatively comprehensive criteria for quality assessment in their study of macular vessel density using OCT-A and recommended exclusion of scans with motion and blink artefacts, media opacities and incorrect vascular network segmentation. While these criteria appear intuitive, their impact on repeatability between scanning sessions has not been formally tested. In the current work, we evaluated several image parameters and reported that some, but not all, have significant impact on the repeatability of retinal capillary plexus density measurements. In addition, these factors have different impact on the repeatability on the superficial and deep plexuses.

Overall, we observed that the repeatability of vessel density measurements was higher for the superficial plexus than the deep plexus, similar to previous work,21 and is likely the result of higher resolution of the superficial plexus compared with the deep plexus on the instrument used for this study. We also noted the presence of significant variation in vessel density in our study population, which was comparable to that reported previously, although our study included a group of patients at older age compared with previous studies.23 The mean differences in between scans for the superficial plexus was 16.4%±15.1% prior to the application of quality criteria, compared with 8.69%±8.52% after using our screening criteria. Variation in the deep plexus densities improved from 31.7±35.0% to 6.56±6.48% for the deep plexus. The unfiltered values may be useful for future studies to consider as a minimum value to reflect true change rather than intervisit variability. The only factors that significantly affected repeatability of superficial plexus density measurement were the presence of motion artefact and the TopQ score. More factors affected deep plexus density, including low visibility of fine vessels, the presence of motion artefact, B-scan quality and the TopQ score. Surprisingly, scan centration and tilt did not appear to significantly influence the repeatability of our scans. Our findings offer a standardised, yet simple parameters for future studies to screen the quality of OCT-A images and determine their acceptability for quantitative measurement analysis.

In terms of study limitations, we acknowledge the relatively small number of eyes used in this study (n=44) as an obvious limitation, as was the relative heterogeneity of the study cohort. Our study cohort also included an older age group than most similar works. While this might have introduced greater variability due to lower resolution and higher rates of motion artefacts, these data are particularly informative as they are not derived from only young and cooperative volunteers. An additional limitation is the use of automated retinal layer segmentation for our analyses. Lastly, the variability and influencing factors of capillary density measurements may be different according to individual OCT instruments.32 However, the factors we described (eg, visibility of fine vessels and motion artefact) are mostly qualitative and thus more likely to be generalisable.

In summary, our findings demonstrate the importance of several key OCT-A scan parameters including visibility of the fine vasculature of retinal capillary plexus, motion artefact and the TopQ score that can be used in the clinical setting to assess the quality of OCT-A scans to ensure reliability between clinic visits when making clinical decisions. Although this technology is still in its infancy, it will be important for ophthalmologists to be able to make these assessments when they become more widespread in day-to-day clinical practice.



  • Contributors BJF and GCMC were responsible for the original concept, data collection and analysis and manuscript writing. Drafting and critical revision were done by GSWT, ACST, IYSY and TYW.

  • Competing interests None declared.

  • Patient consent Obtained.

  • Ethics approval SingHealth Centralized Institutional Review Board.

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

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.