Article Text
PDF
Abstract
Purpose To describe optical coherence tomography angiography (OCTA) image artefacts in eyes with and without ocular pathologies.
Methods The OCTA images of healthy subjects and patients with age-related macular degeneration, diabetic retinopathy and retinal vascular occlusions were retrospectively reviewed. All OCTA images were obtained using a swept-source OCTA instrument (Triton, Topcon). The frequency of various image artefacts including segmentation, banding, motion, projection, masking, unmasking, doubling of the retinal vessels, blink, stretching, out-of-window and crisscross artefacts was assessed. The impact of the artefact on the grading of the images for the foveal avascular zone in deep and superficial retinal layers, capillary non-perfusion and choroidal neovascularisation (CNV) was evaluated.
Results OCTA images of 57 eyes of 48 subjects including 23 eyes (40.3%) with CNV, 13 eyes (22.8%) with dry age-related macular degeneration, 9 eyes (15.7%) with cystoid macular oedema due to diabetic retinopathy or retinal vein occlusion and 12 normal eyes (21.1%) were available for evaluation. At least one type of artefact was present in the images from 51 eyes (89.4%). Banding artefact, segmentation, motion, unmasking, blink, vessel doubling, masking and out-of-window artefacts were found in 51 (89.4%), 35 (61.4%), 28 (49.1%), 9 (15.8%), 5 (8.8%), 1 (1.7%), 1 eye (1.7%) and 1 eye (1.7%), respectively. Projection artefact, stretch artefact or crisscross artefact was not observed. Banding, motion and segmentation artefacts were statistically significantly more frequent in eyes with ocular pathology compared with control eyes (all p<0.001). Eyes with choroidal diseases had significantly higher rate of segmentation error in the choriocapillaris slab compared with eyes with only retinal disease (p=0.02). In nine eyes (17.6%), the artefacts were deemed severe enough by the graders to preclude accurate grading of the image.
Conclusions Image artefacts occur frequently in OCTA images. The artefacts are more frequent in eyes with pathology.
- Imaging
- Retina
Statistics from Altmetric.com
Introduction
Optical coherence tomography angiography (OCTA) is a non-invasive technology that uses motion contrast imaging to produce high-resolution angiographic images of the retinal and choroidal vasculature. OCTA provides a three-dimensional vascular mapping at the microcirculation level and has shown its ability to identify important vascular changes in different retinal and choroidal diseases including diabetic retinopathy, retinal vascular occlusions and choroidal neovasculartisation (CNV).1 ,2
An essential step in the interpretation of any image from any retinal imaging modality is to ensure that the information or finding is real and not due to noise or artefact. This is a critical element in the standard ‘reading centre’ approach to image analysis. Artefacts, unfortunately, may be present with any imaging modality; the type of artefacts may be unique to individual imaging modalities. Failure to properly recognise these image artefacts may lead to incorrect diagnosis and management of the disease. Recent studies have described different types of artefacts in OCTA images.3–5 However, the frequency and impact of these artefacts have not been clearly established. This is important to determine as OCTA imaging begins to be incorporated into clinical trials and analysed in centralised reading centres. In addition, artefacts in the setting of swept-source OCTA have not been defined. The aim of this study was to quantitatively analyse the artefact types and their prevalence in a routine retinal clinical practice using a swept-source OCTA technology.
Methods
In this retrospective observational study, the OCTA images of all patients who underwent OCTA imaging in the Retina Unit of the Doheny Eye Center between January and April 2016 were reviewed. OCTA images of a group of healthy volunteers were selected as controls. The study was approved by the Institutional Review Board of the University of California—Los Angeles. The research adhered to the tenets of the Declaration of Helsinki and the Health Insurance Portability and Accountability Act, and informed consents were obtained from the participants.
OCTA was performed with a Topcon OCT instrument (DRI OCT Triton plus, Topcon, Tokyo, Japan). The Triton swept-source OCT uses a wavelength of 1050 nm with a scan speed of 100 000 A-scans per second. The instrument employs an active eye tracker that follows the eye movement, detects blinking and adjusts the scan position accordingly, thereby reducing motion artefact during OCTA imaging. All eyes were scanned using a 3×3 mm protocol. An artefact removal option is also available in the software and was used for this study. The macular scans were automatically segmented by the OCTA software into four ‘en face’ OCT slabs; (1) superficial retinal slab (SRL) from 2.6 µm beneath the internal limiting membrane to the 15.6 µm beneath the interface of the inner plexiform layer and inner nuclear layer (IPL/INL), (2) deep retinal slab (DRL) from the 15.6 µm beneath the IPL/INL to 70.2 µm beneath the IPL/INL, (3) outer retinal slab from 70.2 µm beneath the IPL/INL to the Bruch's membrane (BM) and (4) choriocapillaris from the BM to 10.4 µm beneath the BM.
Two independent, certified Doheny Image Reading Center graders (KGF and MA) reviewed the images on the instrument software (IMAGEnet 6 V.1.14.8538) and recorded the artefacts based on previously described categories.3 ,4 In all four slabs, all horizontal OCT B-scans were reviewed to ensure accurate segmentation of the layers. OCTA images were reviewed for motion, projection of superficial vessels, masking, unmasking, doubling of the retinal vessels, blink, stretching, out of screen and crisscross artefacts (figure 1A–G). A brief description of each artefact is provided in table 1. A very strict definition of segmentation artefact was used; any detectable deviation from the expected boundary for any length of B-scan, on any B-scan was deemed to be evidence of a segmentation artefact. During the assessment of the OCTA images, the graders noted a new artefact type that had not been specifically reported in OCTA images, which was termed a ‘banding’ artefact (figure 2A–D). In this artefact, bands (composed of multiple adjacent B-scans) on the en face structural OCT or OCTA image showed a different intensity or brightness to adjacent regions. Since the ‘negative’ (dark outline of the removed vessel) projections of the superficial vessels is considered an unavoidable and expected consequence of artefact removal software and is always present in all DRL, outer retinal and choriocapillaris slabs, we did not consider it as an additional artefact for this analysis. The impact of artefacts on the ability to grade the margin of the fovea avascular zone (FAZ) in the DRL and SRL was assessed in all images. Also, in eyes with CNV, the ability to grade the margin of the neovascularisation, and in eyes with cystoid macular oedema (CMO) due to diabetic retinopathy or retinal vein occlusion, the ability to grade the margin of the non-perfusion areas was evaluated. Independent evaluation allowed us to assess the repeatability of the artefact determinations. However, to generate one final answer for data analysis, discrepancies between the two graders were resolved by open post hoc adjudication. The reading centre medical director (SRS) served as the final arbiter in case that the two graders were not able to resolve a disagreement.
Definitions for the artefacts assessed in the images obtained using a swept-source optical coherence tomography angiography instrument
Artefacts in optical coherence tomography angiography (OCTA) images. Correct segmentation in a normal eye (A). The error in detecting inner plexiform layer/inner nuclear layer interface (long arrow) and Bruch's membrane (short arrow) leads to segmentation error in superficial retinal layer and outer retinal slabs (B and C). Motion artefact (long arrow) and blink artefact (short arrow) in superficial retinal layer slab (D). Unmasking artefact (short arrows) in choriocapillaris slab may resemble a choroidal neovascularisation (E). Severe motion artefact preventing proper visualisation of the macular microvasculature and grading of the foveal avascular zone (F). Doubling of the retinal vessels (arrows) in superficial retinal layer slab (G).
Banding artefact in a normal eye (A and B) and in diabetic macular oedema (C and D). Bands (multiple adjacent B-scans) of lighter or darker intensity can be seen in the en face optical coherence tomography angiography (long black arrow) or structural optical coherence tomography (short black arrow) images. Some banding artefacts (long white arrows) coincide with the motion artefact (short white arrows).
Data were analysed using SPSS software (V.16, SPSS, Chicago, Illinois, USA), and t-test, χ2 test and Fisher’s exact test were used for analysis and comparisons between groups. A p value <0.05 was considered significant.
Results
Overall, 57 eyes of 48 subjects including 25 females and 23 males with a mean age of 68.9±2.1 years were assessed. Clinical diagnoses among the cohort included CNV due to age-related macular degeneration (AMD) in 23 eyes (40.3%), non-neovascular AMD in 13 eyes (22.8%) and CMO due to diabetic retinopathy or retinal vein occlusion in 9 eyes (15.7%). Twelve eyes (21.1%) had no evidence of disease on clinical examination or imaging and served as healthy controls.
Fifty-one eyes (89.4%) had at least one type of artefact. All six eyes without artefact were healthy controls. The most prevalent artefacts were banding artefacts in 51 eyes (89.4%), followed by segmentation artefact in 35 eyes (61.4%) and motion artefact in 28 eyes (49.1%). The unmasking artefact, blink artefact, vessel doubling artefact, masking artefact and out-of-window artefact were found in nine eyes (15.8%), five eyes (8.8%), one eye (1.7%), one eye (1.7%) and one eye (1.7%), respectively. No eyes with projection artefact, stretch artefact or crisscross artefact were found. In eyes with segmentation artefact (figure 1B, C), SRL and DRL slab segmentation error was found in 32 eyes (91.4%), outer retinal slab segmentation error was found in 35 eyes (100%) and choriocapillaris slab segmentation error was found in 22 eyes (62.8%). It should be noted that we used a strict definition for a segmentation error and, in many cases, only a small segment of a given surface/layer in the B-scans was affected. Unmasking artefact was found in six eyes with geographic atrophy due to the AMD and three eyes with CNV associated with geographic atrophy. All unmasking artefacts were found in the outer retinal and choriocapillaris slabs.
Table 2 shows the characteristics of patients and artefact types in eyes with and without ocular pathology. Control subjects were significantly younger (p<0.001). Banding, motion and segmentation artefacts were statistically significantly higher in eyes with ocular pathology (all p<0.001).
Demographics and artefacts of eyes with and without ocular pathology
Eyes with ocular pathologies were further divided into retinal (diabetic retinopathy and retinal vein occlusion) and choroidal (CNV and dry AMD) groups (table 3). No significant difference was found between the two groups in terms of motion, blink and unmasking artefacts (p=0.4, p=1 and p=0.1, respectively). Segmentation error in SRL, DRL and outer retinal slabs was similar between the two groups (p=1, p=1 and p=0.6, respectively); however, eyes with choroidal disease had significantly higher rate of segmentation error in the choriocapillaris slab (p=0.02).
Frequency of artefacts in eyes with primary retinal (macular oedema due to diabetic retinopathy and retinal vein occlusion) and choroidal (exudative and non-exudative age-related macular degeneration) diseases
Among eyes with artefacts (51 eyes), grading of the FAZ (SRL or DRL), the area of non-perfusion (in eyes diabetic retinopathy or retinal vein occlusion) or the borders of CNV (in neovascular AMD eyes) was not possible in nine eyes (17.6%). All ungradable cases were eyes with ocular pathology. The primary artefact types that precluded grading were motion artefact in three eyes (33%) and a combination of motion, blink and segmentation artefacts in the other six eyes (66%). Overall, the FAZ was not gradable at the SRL and DRL in one (1.9%) and four (7.8%) eyes, respectively. In eyes with CNV (23 eyes), grading of the CNV border was not possible for five eyes (9.8%). In eyes with CMO due to diabetic retinopathy or retinal vein occlusion (nine eyes), the non-perfusion area was not gradable in one eye (1.9%).
Discussion
In this study, a majority of OCTA images (89%) had at least one type of artefact, though the artefacts usually did not preclude the graders ability to make key assessments. In a minority of cases (17.6%), severe motion, blink and segmentation artefacts reduced the visibility of the vessels to the extent that at least one of three key quantitative parameters could not be assessed by the graders. In eyes with geographic atrophy, hypertransmission of the light through the area of retinal pigment epithelium (RPE) atrophy leads to augmented visibility of the choroidal vessels. This unmasking artefact could potentially be erroneously interpreted as CNV, but this was easy to recognise as an artefact with proper training.
Our results demonstrated that the prevalence of artefacts was higher in the presence of ocular pathology compared with the healthy eyes. This is similar to previous studies which have shown that OCT artefacts are more prevalent in the presence of retinal and choroidal pathologies.6–10 In particular, segmentation artefacts in this study was significantly higher in eyes with ocular pathology. The automatic segmentation of the swept-source OCTA image analysis software divides the OCTA images into four en face slabs. Three inner slabs use the IPL/INL interface as a reference. Consequently, every segmentation error in detecting the IPL/INL interface propagates the error to all three slabs. This was evident in our study as the frequency of segmentation error was similar in SRL and DRL slabs. Since the outer retinal slab is affected both by errors in the correct detection of the IPL/INL interface as well as the BM, segmentation errors were more frequent in this slab. Although the OCTA instrument interfaces allow the user to adjust the location of the slab by moving individual surfaces up or down, when the segmentation error affects only a portion of a B-scan, this requires correction on each individual section to produce an accurate slab—a task that may not be practical in clinical practice but may be feasible in the context of reading centres and clinical trials.
In our study, we did not detect certain types of artefacts that have been described in previous reports including the stretch and crisscross artefacts. Also, the doubling of vessel was detected only in one eye. We suspect this reflects differences in OCTA software and hardware. This report is the first to characterise artefacts with the Topcon swept-source OCTA instrument, which may behave differently than the previously described devices. The Topcon OCTA instrument has a real-time tracker that actively follows the fixation movements. As a result, subsequent software-based motion correction may not be necessary and may reduce the occurrence of stretch and crisscross artefacts. One artefact that was very frequent in our study was the banding artefact. Banding artefact can be observed in structural OCT en face images and is due to varying intensity/brightness between B-scans. This type of banding is not uncommon in the setting of spectral domain OCT devices in which the B-scan position within the scan window varies significantly (ie, the scan is not ‘locked’ in the z-axis) as there is significant sensitivity roll-off with depth in SD-OCT. However, with SS-OCT there is little sensitivity roll-off, and thus, the explanation for the banding artefact in this case may be different. Regardless, despite its frequency, the banding artefact did not seem to have much effect of the graders' ability to make qualitative assessments of interest. It is unclear, however, what impact such an artefact may have on quantitative assessments such as vessel density, and this may require further study.
Our study has several limitations. First, the sample size was relatively small, and thus we were underpowered for identifying smaller differences between groups, particularly among different types of diseases. In addition, the controls were significantly younger than the patients with pathology, and thus some of the artefacts may have been related to the aged eye rather than related to the pathology itself. Despite these limitations, this is the first study reporting the prevalence and relevance of these image artefacts using a SS-OCTA device, and using reading centre certified graders. Given the inclusion of OCTA technology in clinical trials, better defining these artefacts and assessing their impact or reading centre assessments would appear to be of importance.
Footnotes
Twitter Follow Handan Akil at @eyedrhandan
Contributors Concept and design; data collection; analysis and interpretation; writing the article; critical revision of the article; final approval of the article; provision of materials, patients or resources; literature search: KGF. Final approval of the article; data collection; provision of materials, patients or resources: MA-S. Data collection; critical revision of the article; final approval of the article; provision of materials, patients or resources; technical or logistic support: HA. Concept and design; critical revision of the article; final approval of the article; provision of materials, patients or resources; administrative, technical or logistic support: SRS.
Competing interests SRS is a consultant for Optos, Genentech and Allergan, and receives research support from Optos, Genentech, Allergan and Carl Zeiss Meditec.
Patient consent Obtained.
Ethics approval UCLA.
Provenance and peer review Not commissioned; externally peer reviewed.