Purpose To evaluate the reproducibility and interuser agreement of measurements of choroidal neovascularisation in optical coherence tomography angiography (OCTA).
Design Prospective non-interventional study.
Methods Consecutive patients, presenting with neovascular age-related macular degeneration (AMD), underwent two sequential OCTA examinations (AngioVue, Optovue, Fremont, California, USA), performed by the same trained examiner. Neovascular lesion area was then measured on both examinations in the choriocapillaris automatic segmentation by two masked readers, using the semiautomated measuring software embedded in the instrument. Two measuring features were used: the first corresponding to the total manually contoured lesion area with the flow draw tool (select area) and the second to the total area of solely vessels with high flow within the lesion (vessel area). These measurements were then compared in order to assess both the reproducibility of OCTA examination and the interuser agreement with the embedded software.
Results Forty-eight eyes of 46 patients (77.4 mean age,+/-8.2 SD, range from 62 to 95 years old, eight men, 38 women) were included in our study. Mean choroidal neovascularisation area was of 0.72+/-0.7 mm2 for the first measurement and 0.75+/-0.76 mm2 for the second measurement; difference between the first and the second measurement was 0.03 mm2. Intrauser agreement was of 0.98 (CI 0.98 to 0.99) for both ‘vessel area’ and ‘select area’ features. Interuser agreement was of 0.98 (CI 0.97 to 0.99) for ‘select area’ and ‘vessel area’ features.
Conclusion Our data suggest that OCTA provide reproducible imaging for evaluation of the neovascular size in the setting of AMD.
- OCT angiography
- neovascular AMD.
Statistics from Altmetric.com
Age-related macular degeneration (AMD) is one of the leading causes of blindness among individuals ≥55 years in developed countries.1 Two forms of AMD, neovascular and atrophic, have been described, both potentially causing severe visual loss consequent to either choroidal neovascularisation (CNV) or geographic atrophy.2 According to their anatomic location, Gass classified choroidal neovascular membranes in two types: type 1, also known as ‘occult’ neovascularisation, which occurs beneath the retinal pigment epithelium (RPE) monolayer, and type 2, known as ‘classic’ neovascularisation3 4 which occurs above the RPE. Following Gass’s anatomical description, Freund and Yanuzzi defined type 3 neovascularisation as a proliferation in the neurosensory retina harbouring either a retinal or a choroidal origin.5 6
The abnormal growth of newly formed vessels within the macula causes exudation, bleeding and disciform scar formation.2 Clinical aspects of different phenotypes of neovascular AMD have been widely described in the literature, along with the typical characteristics at fundus examination, red-free and infrared photographs, leakage patterns in fluorescein angiography (FA) and indocyanine green angiography (ICGA) and typical features in optical coherence tomography (OCT).7–10
Reliable instrumental tests, as Spectral Domain OCT (SD-OCT), are essential for the diagnosis and follow-up of patients affected by wet AMD during antiangiogenic treatment.
Optical coherence tomography angiography (OCTA) is a novel imaging technique derived from ‘En face’ OCT and based on split-spectrum amplitude decorrelation angiography (SSADA) algorithm that allows flow detection within retinal microvasculature without dye injection.11 Thus, it allows the visualisation of dynamic erythrocyte motion, by the acquisition at the same location on the retina of several sequential OCT cross-sectional scans. This new non-invasive tool may be used in daily clinical ophthalmology practice and may replace invasive angiography techniques in the future.12
Methods for automated detection and semiautomated quantification of CNV so far mostly relied on manual area contour measurement.4 5
Several studies reported the suitability of OCTA for the detection of CNV13 and, more recently, the reproducibility of the measurements of choroidal neovascular membranes area has been afforded.14 15 OCTA reflects the flow features without direct anatomical view of the CNV lesion and SSADA algorithm grants the reproducibility of the flow measurements; it could be therefore considered as a viable method for diagnosis and quantification of CNV all along the course of disease. Because the visualisation of the flow does not directly reflect the anatomical structures, the reliability of CNV measurement could be questioned.
Thus, the aim of the study was to assess the reproducibility and the reliability of vessel area measurements by OCTA and its usefulness in the follow-up of AMD patients.
All consecutive patients diagnosed with neovascular AMD, both treatment-naive and treated by anti-vascular epithelial growth factor, presenting at the University Eye Clinic of Créteil (France) on 1 March 2016, were prospectively included. These patients presented a type 1 neovascularisation, with a retinal pigment epithelial detachment of height inferior to 300 microns, as well as type 2 or type 3 neovascularisation.
Diagnosis of types 1, 2 and 3 neovascularisation had been either previously made on multimodal imaging (FA, ICGA, SD-OCT) at first examination or, if the patient was treatment-naive, we performed FA and ICGA, alongside with SD-OCT and OCTA, the latter having been performed for all the patients included in this series.
Actually, FA and ICGA were performed before OCTA. Conventional angiography was only analysed in the context of the present study in order to confirm the type of CNV.
The study was performed in agreement with the Declaration of Helsinki for research involving human subjects and the French legislation. Approval from Université Paris Est Créteil Institutional Review Board was obtained for this study.
Patients with one or more of the following clinical condition were not included in the study: low visual acuity (<20/250 because of poor fixation), media opacities, subretinal fibrosis, pathological myopia ≥6 D, any previous treatment (retinal surgery, laser photocoagulation) and hereditary retinal dystrophy. Patients with poor quality images in OCTA, due to motion artefacts or low signal strength (<55), were excluded from the analysis.
All patients underwent a complete ophthalmic examination including monocular best-corrected visual acuity (BCVA) using standard ETDRS charts, slit-lamp examination, fundus biomicroscopy as well as serial infrared and simultaneous SD-OCT imaging using a Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany).
All patients also underwent OCTA (AngioVue, RTVue XR Avanti, Optovue, Fremont, California, USA, V.2015.1.0.90) to obtain amplitude decorrelation angiography images. Two trained examiners (AM, FA) ran OCTA examinations after pupil dilatation of patients: each examiner separately performed two sequential OCTA examinations on each patient of the study. AngioVue has an A-scan rate of 70 000 scans per second, using a light source centred on 840 nm and a bandwidth of 50 nm. Each OCTA volume contains 304×304 A-scans with two consecutive B-scans captured at each fixed position before proceeding to the next sampling location. Split-spectrum amplitude-decorrelation angiography was used to extract the OCTA information. Each OCTA volume is acquired in 3 s and two orthogonal OCTA volumes were acquired in order to perform motion correction to minimise motion artefacts arising from microsaccades and fixation changes. Angiography information displayed is the average of the decorrelation values when viewed perpendicularly through the thickness being evaluated.
To evaluate the morphology and flow within the neovascular membranes, we used a 3×3 mm scanning area. Only CNV lesions comprised in the 3×3 mm frame were analysed. The segmentation provided by the machine software underwent minor adjustments, to ensure correct visualisation of the capillary plexuses, outer retinal layers and choriocapillaris layer.
Moreover, we coined a 30 microns manual segmentation, having its inner boundary at the Bruch’s membrane level and its outer boundary 30 microns underneath the Bruch’s membrane.16
Manual measurements of CNV area were performed independently by the two examiners using measuring tool embedded in the machine software: two parameters were identified, the first corresponding to the total manually contoured lesion area with the flow draw tool (select area) and the second to the total area of solely vessels with high flow within the lesion (vessel area).15 The choriocapillaris segmentation was used for measurements for type 1 and type 2 neovascularisation, while for type 3 neovascularisation, given its predominantly intraretinal localisation, we used a customised 30 micron manual segmentation around Bruch’s membrane. To evaluate the reproducibility of this examination, two masked trained users (AM, FA) independently performed two consecutive measurements of the lesion area for all patients included in this study. A total of four vessel and select area measurements of the same neovascular membrane were achieved.
Discordance between each independent measurement apart was defined as a more than 10% variation in lesion area, and in such case, a third reader performed the measurement (EHS).
Qualitative variables were described in percentages and quantitative variables were described by their mean value with SD or by their median with IQR, depending on their distribution (STATA SE/13.1, StatCorp, Texas, USA).
Bland and Altman test was used to evaluate the agreement between the readers in vessel and select area measurements. The mean of the measurements were calculated and compared with their differences. For each subject, the mean of the measurements were plot along the X axis and the differences between the measures of the readers along the Y axis. The measurements were considered to be in agreement when most of the plots were within the range d+/-2 SD, where d was the mean of the differences between the reader’s measurements. p<0.05 were retained as significant.
Demographics and main clinical data
Forty-eight eyes from 46 consecutive patients (eight men, 38 women, mean age 77.4 years+/-8.2 SD, ranging from 62 to 95 years) with exudative AMD were included.
Cases included in our study presented with type 1 CNV in 26/48 cases (54.16%), type 2 CNV in 14/48 cases (29.16%), mixed CNV in 2/48 (4.16%), while type 3 CNV was present in 6/48 eyes (12.5%), as assessed by fluorescein and ICGA examination. All demographic and clinical characteristics of patients are summarised in table 1.
Optical coherence tomography angiography images of the choriocapillaris segmentation revealed in all cases a high flow, convoluted neovascular network, which was measured in all the patients (figure 1).
Moreover, in the 14 cases of type 2 CNV, this network was visible on both the outer retinal and choriocapillaris segmentation. However, measurements were performed on the choriocapillaris segmentation.
As for type 3 neovascularisation (6/48 eyes), the mean lesion area was measured in the manually segmented 30 microns underneath Bruch’s membrane.
Overall, the first lesion area measurement was 0.72+/-0.7 mm2 for the first measure and 0.75+/-0.76 mm2 for the second measure; mean difference was 0.03 mm2.
Bland and Altman test revealed that, for both select area and vessel area, the intrareader agreement had an interclass correlation coefficient (ICC) of 0.98 (95% CI ranging from 0.98 to 0.99), while the inter-reader agreement had an ICC of 0.98 with a 95% CI ranging from 0.97 and 0.99 (figure 2). The summary of interuser and intrauser agreement for the entire cohort is presented in table 2.
Moreover, Bland and Altman and ICC were separately performed for the different types of neovascular membranes.
Thus, inter-reader ICC was very high between the two readers for the vessel area feature for type 1 CNV averaging 0.99, (95% CI 0.99 to 1), type 2 CNV averaging 0.96 (95% CI 0.93 to 0.99) and mixed CNV averaging 0.97 (95% CI 0.91 to 1.03). As for the select area feature, ICC was also very high for type 1 CNV averaging 0.99 (95% CI 0.99 to 1), for type 2 CNV averaging 0.97 (95% CI 0.95 to 0.99) and mixed CNV averaging 0.98 (95% CI 0.93 to 1.02). Concerning the six cases of type 3 neovascularisation, the ICC was 0.20 (95% CI 0 to 1.04) for the vessel area option and ICC averaged 0.25 (95% CI 0 to 1.04) for select area.
Intrareader ICC was high for all types of CNV included in our study. For type 1 CNV intrareader ICC averaged 0.99 (95% CI 0.98 to 0.99) for vessel area and 0.98 (95% CI 0.97 to 0.99) for select area, respectively. For type 2 CNV intrauser ICC averaged 0.97 (95% CI 0.95 to 0.99) for vessel area and ICC 0.97 (95% CI 0.95 to 0.99) for select area. Intrareader ICC was of 0.97 for vessel area of mixed CNV (95% CI 0.89 to 1.04) and 0.98 for select area of mixed CNV (95% CI 0.95 to 1.01). As for type 3 neovascularisation, intrareader ICC amounted to 0.99 (95% CI 0.97 to 1) for vessel area and 0.99 (95% CI 0.97 to 1) for select area, respectively. The summary of interuser and intrauser agreement for measurements regarding each type of neovascularisation is presented in table 3.
As regards the Bland and Altman, the d value was close to 0 for the vessel and select area measurements, it was considered that there were no systematic differences of measurements between the readers.
In order to validate a new imaging modality, there is a multistep procedure, starting with precision, as per repeatabilty and reproducibility. In our study, we aimed to describe these essential features for OCTA quantitative measurements, using the tool embedded in the AngioVue software, in a heterogeneous group of neovascular AMD patients. Agreement for the quantitative analysis of total lesion size (select area) and vessel density (vessel area) was very high: 0.98 (CI 0.98 to 0.99) for intrauser agreement and 0.98 (CI 0.97 to 0.99) for interuser agreement. Thus, our results have demonstrated that OCTA examination is a reproducible examination for neovascular AMD patients.
In table 3, we reported the intrareader and inter-reader agreement for each type of CNV: the concordance is very high for type 1 and 2 CNV for both intrareader and inter-reader agreement; on the contrary, for type 3, the intrareader agreement is good, but the inter-reader agreement is poor. Type 3 CNV is a peculiar type of neovascular AMD which can be visualised in three different segmentations: in deep capillary plexus, we noted the deep feeder vessel; in the outer retina segmentation, we noted the tuft-shaped lesion, and in the choriocapillary segmentation, we noted a clew-like lesion, which may be secondary to the projection artefact or the presence of a sub-RPE neovascularisation. Subsequently, the detection of type 3 CNV is more complex than types 1 and 2 CNV. This may explain the lower inter-reader agreement.
Furthermore, our group has previously used the 30-micron manual segmentation in multiple studies16; as we have observed, it allowed a more precise delineation of the neovascular membrane than automatic outer retinal or choriocapillaris segmentations. Precisely, the upper boundary of this segmentation is on the Bruch’s membrane, while its lower boundary is situated 30 microns underneath.
Recent papers have extensively focused on the qualitative evaluation of choroidal neovascularisation in OCTA.13 15 17 18 However, given that the use of OCTA imaging is widespread for the evaluation of neovascular AMD, the manual measurements of the select and vessel area have to be reproducible and repeatable, in order to ensure a correct follow-up of these patients. Carpineto et al 19 has recently shown that the reproducibility and repeatability of foveal avascular zone measurements on OCTA images in healthy subjects by optical coherence tomography angiography was excellent. Several studies reported the suitability of OCTA for the detection of CNV and provided quantitative measurements demonstrating also the reliability of OCTA as compared with ICGA.15
In the present study, conventional angiography (FA, ICGA) was performed before the acquisition of OCTA in the context of multimodality imaging; notwithstanding diagnosis was already achieved, each of these tests was performed independently to define specific features of the disease. This allowed us to avoid selection bias.
Furthermore, Gao et al 14 developed a level set method to segment neovascular component within the detected CNV membrane and a skeleton algorithm to determine vessel centrelines to quantify the vessel length of the CNV network. Nevertheless, in a clinical setting, an image analysis dedicated software usage is highly time-consuming, so the need of a validated, reproducible measurement of the measurement tool embedded in the instrument software is necessary. However, in our study, we have used for all measurements a customised 30-micron segmentation underneath the Bruch’s membrane. Moreover, we have used the highest resolution scanning area (3×3 mm) for lesion measurements, which allows a more detailed assessment of CNV morphology and size, as compared with 6×6 mm scanning area.
Our study demonstrates that OCTA measurements, as performed on the embedded software, are reproducible and repeatable. Indeed, statistical analysis as shown in table 2 and figure 2 showed a very strong intrauser and interuser reproducibility for both CNV select area and vessel area measurements, suggesting that OCTA is a reliable imaging tool, useful in the follow-up of patients affected by neovascular AMD.
Our study has several limitations. First, patient cooperation, a good fixation and no/few media opacities are required to obtain interpretable OCTA images. Thereafter, the scanning area for OCTA images with sufficient resolution is centred on the 10 central degrees, making the CNV area evaluation difficult in the case of large neovascular membranes.
Furthermore, we excluded patients with late AMD because of low VA and poor fixation.
Last but not least, we had a limited sample size, for mixed and type 3 CNV in particular. In fact, the vessel and select area measurements has proven to be, in our study, more useful in case of types 1 and 2 CNV, rather than in the other CNV types.
Finally, we included only pigment epithelial detachments (PEDs) with height of <300 microns due to technical limitations of OCTA instrument. Both from our experience and from recent literature,20 we have observed that the assessment of the CNV is difficult in PEDs of great height due to the flow void artefact.
Despite these limitations, OCTA showed a good reproducibility and repeatability, proving that OCTA has the potential of becoming a relevant biomarker for AMD patients. We retain that both lesion size and vessel area on OCTA could represent a valuable method to follow-up patients with CNV.
The possibility of analysing CNV has been definitely improved by the introduction of OCTA and built-in new softwares, that is, the flow draw tool to analyse the neovascular lesion area; these latter tools, above all, streamlined the understanding of retinal degenerative diseases, in particular neovascular AMD. This study indeed enters this thread and identifies the reproducibility and repeatability of OCTA inbuilt segmentation software. Although, this represents a preliminary report, we can assume that further valuable instruments have been added to help the diagnostic and management process of age-related macular degeneration.
Contributors FA and AM contributed to the conception and design of the study, writing, drafting and revising process. OS and VC were involved in execution of the study, revising and drafting process. CJ, FA and AM were involved in data collection, interpretation and statistical analysis. EHS participated in all the main aspects of the study, as well as revising and the final approval of the article.
Competing interests OS: consultant for Novartis (Basel, Switzerland), BayerShering-Pharma (Berlin, Germany), Allergan (Irvine, California, USA), Optovue (Freemont, California, USA). ES: consultant for Novartis (Basel, Switzerland), Bayer Shering-Pharma (Berlin, Germany), Allergan Inc (Irvine, California, USA), Farmila-Thea (Clermont-Ferrand, France).
Ethics approval Université Paris Est Créteil Institutional Review Board.
Provenance and peer review Not commissioned; externally peer reviewed.
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.