Optical coherence tomography angiography in diabetic retinopathy

1Diabetic retinopathy (DR) is the most common complication of diabetes mellitus and a leading cause of vision loss in people of working age, with an estimated global prevalence of 27–35% among all patients with diabetes. DR has a high estimated global burden: in 2021, it was projected that patients with DR would exceed 160 million by 2045.

2Although there is a long history of research into DR, there remain many unanswered questions about how to assess and treat the pathology and complications that arise from this disease, including how DR may differentially affect different layers of the retina, how current (and future) treatments affect the retinal microvasculature and therefore impact retinal blood flow, and how functional measures link to anatomical and vascular metrics. Optical coherence tomography angiography (OCTA) is an emergent imaging modality that may be able to provide information that answers some of these questions.

3The early clinical features of DR occur because of chronic hyperglycemia leading to changes in the vascular wall such as pericyte loss, as well as microvascular damage, which includes occlusion of the capillaries as well as increased leakage from these vessels into surrounding tissue and subsequent abnormalities in blood flow. These events lead to progressive microvascular injury, resulting in microaneurysms, hard exudates, and retinal hemorrhages. As DR progresses, a range of complications may manifest, including cotton wool spots, venous beading, and intraretinal microvascular abnormalities (IRMAs). Furthermore, with the onset of proliferative DR, retinal neovascularization can emerge, leading to serious complications such as preretinal and vitreous hemorrhages, as well as tractional retinal detachment. These complications can cause severe vision loss and may necessitate vitrectomy surgery. Patients may develop additional complications that exacerbate vision loss, including diabetic macular edema (DME) and diabetic macular ischemia (DMI). Advanced, proliferative DR is typically treated with panretinal photocoagulation, alongside anti-vascular endothelial growth factor (VEGF) therapies, depending on specific comorbidities.

4DME is the result of a build-up of intracellular and/or extracellular fluid (edema) in the macula, originating from leaky blood vessels and the breakdown of the blood-retinal barrier. If left unmanaged, DME can result in up to 8% of vision being lost per year over a 3-year period; this is equivalent to 15 Early Treatment Diabetic Retinopathy Study (ETDRS) letters. The standard of care for DME has historically been focal laser photocoagulation. However, anti-VEGF therapies are now the gold standard for center-involved DME, leading to a more rapid resolution of DME and results in a mean visual gain that is superior to laser treatment.

5Macular ischemia is a key feature of DR and another cause of vision loss associated with DR. It is associated with decreased macular retinal vessel density and/or non-perfusion of the superficial vascular complex (SVC) and deep vascular complex (DVC). It often results in enlargement and disruption of the foveal avascular zone (FAZ), the integrity of which is central to maintaining normal visual acuity. Irreversible vision loss in DMI occurs due to damage to neuronal tissues from macular capillary loss or capillary non-perfusion. Currently, there is no formal clinical definition for DMI and no approved treatment. Treatments used to treat DME do not appear to improve ischemia.

6Retinal non-perfusion in eyes with DR occurs mostly in the mid and far periphery, with up to 70% of non-perfusion occurring outside the posterior pole of diabetic eyes. Analysis of Protocol AA data showed that increasing baseline retinal non-perfusion index on ultra-widefield fluorescein angiography (UWF-FA) is associated with worsening of Diabetic Retinopathy Severity Scale (DRSS) score, development of proliferative diabetic retinopathy (PDR), and vitreous hemorrhage. These associations are driven by non-perfusion located primarily in the posterior pole. Ofuji et al. showed that a higher baseline retinal non-perfusion index (NPI) on UWF-FA is associated with later worsening of DRSS score, development of PDR, and vitreous hemorrhage. This is supported by data from a large, prospective longitudinal study, which confirmed that greater baseline peripheral non-perfusion and the presence of predominantly peripheral lesions (PPLs; which refer to visible abnormalities such as microaneurysms and hemorrhages) on UWF-FA are associated with a significantly higher risk of disease worsening; this included areas of both retinal non-perfusion and neovascularization. However, no association between PPLs was found in color fundus photos and an increased risk of disease worsening could be detected. Ofuji et al. also found that the NPI in the posterior pole was most strongly associated with progression to PDR. A greater extent of peripheral non-perfusion is associated with greater DR severity and worse visual acuity. Furthermore, a large peripheral area of capillary non-perfusion at baseline is predictive of non-response to panretinal photocoagulation, the standard treatment for proliferative DR. Whereas central ischemia tends to result in DMI, peripheral ischemia can lead to reactive neovascularization and further ischemia in a vicious cycle. Furthermore, neovascularization of the optic disc is linked to larger overall areas of retinal non-perfusion. Thus, depending on the location and severity of microvascular damage to the retina, a range of different comorbidities may arise in eyes with DR.

7OCTA is a novel and non-invasive three-dimensional (3D) imaging method that can visualize capillaries in all retinal layers. Detailed descriptions of the principles underpinning OCTA are beyond the scope of the current review but have been thoroughly described in other papers. In short, an OCTA image is a map constructed from blood cell motion across the retina revealed by changes in optical coherence tomography (OCT) signal intensity.

8Numerous studies have confirmed that OCTA can be used to identify the earliest evidence of microvascular change and provide quantitative assessment of the extent of diseases such as DR and its common complications, including DMI. OCTA has been shown to be more sensitive than physician examination or fundus imaging and detects capillary loss earlier than when the clinical lesions of DR become apparent, indicating its sensitivity and promise as a tool for assessing DMI. However, it is worth noting that certain microvascular changes, such as microaneurysms associated with DR, may be detected more effectively with FA than OCTA, despite the latter’s ability to identify areas of nonperfusion. In addition, unlike FA, OCTA does not detect vascular leakage, an important feature of DR pathology. These considerations highlight the complementary nature of these imaging modalities in providing a comprehensive assessment of DR.

9Here, we will discuss in detail several OCTA metrics that have potential for use in clinical trials of DR and its complications, how OCTA can assess anatomical changes resulting from DR and relate them to visual function, and how to interpret OCTA images, as well as the challenges and drawbacks of using OCTA in DR. We intend this article to be a definitive point of reference to guide future research using OCTA in DR, supporting the development of new treatments and improving the understanding of how DR can lead to vision loss.

10The FAZ is an area surrounding the fovea that is free of retinal capillaries, limiting light scattering within the foveal pit and resulting in enhanced vision quality. Loss or non-perfusion of the small capillaries at the margins of the FAZ is associated with reduced visual function. In this section, we will discuss using FAZ metrics in OCTA to stratify by disease severity and using change in FAZ metrics over time to evaluate for worsening disease.

11FAZ size, whether measured by OCTA or by fluorescein angiography (FA), is negatively correlated with visual acuity. In one study of 246 patients who had DMI, FAZ sizes were measured using FA and ranged from 0.27 to 0.78 mm2. The study found that patients with larger FAZ sizes, indicating more severe DMI, were significantly more likely to have decreased visual acuity (p < 0.001). Specifically, those with “moderate” or “severe” DMI, defined as FAZ ≥0.32 mm2, exhibited a greater likelihood of reduced visual acuity. That said, despite numerous studies correlating DMI with vision loss, the precise stage at which DMI affects visual function is unclear. Furthermore, other studies have shown a poor correlation between FAZ size and BCVA; this is confounded by the large physiological variability in FAZ size (discussed in more detail further on in this section). The lack of correlation between change in FAZ size and vision loss may be due to a time lag between enlargement and resultant vision loss or to the fact that a certain threshold of capillary dropout needs to be met before visual function declines. However, BCVA is not solely dependent on FAZ enlargement. It is also affected by other factors such as retinal neurodegeneration that results from diabetes itself, as well as associated inflammation. Concurrent pathologies in the macula, such as a history of DME or choroidal ischemia caused by diabetes, may also contribute to BCVA, further complicating the relationship between FAZ size and visual acuity. Establishing a threshold of FAZ size at which visual impairment occurs is difficult. Microperimetry used in conjunction with OCTA may be one option to detect impaired visual sensitivity resulting from macular ischemia at an earlier stage than BCVA.

12Increased FAZ size can be used as an indicator of DMI severity, and FAZ size positively correlates with overall DR severity and progression. In a study including 205 eyes, an increase in FAZ size significantly predicted the progression of DR (p < 0.001). In addition, the FAZ shows a higher degree of irregularity in patients with severe DME compared with those with earlier-stage DME. These data suggest that measurements of the FAZ may provide insight into the status and progression of DR, DME, and DMI. It is important to note that FAZ size can exhibit variability between the superficial and deep plexus and may vary with segmentation even within the same eye, depending on the OCTA machine used. This variability in FAZ size measurement presents an additional layer of complexity when assessing extent of DR. The ETDRS grading system is the gold standard for assessing DR severity and predicting DR progression. Retrospective comparison of OCTA images demonstrated that superficial and deep-retinal OCTA slabs correlate significantly with ETDRS grades (p < 0.001).

13FAZ size can have high inter-person variability (up to ~3-fold; Fig. 1), limiting its utility for confirming DR-induced capillary nonperfusion unless baseline images are taken before the onset of the pathology. Several studies have examined FAZ area in normal eyes, which can vary substantially between individuals (e.g., 0.110–0.861 mm2). In addition, other pathological features, such as epiretinal membrane or the presence of center-involving DME, may alter the size and contour of the FAZ, complicating comparisons. Furthermore, the structure of the FAZ, composed of a convergence of multiple vessel complexes, complicates the quantification of border changes assessed using conventional two-dimensional imaging. Although it is difficult to establish an accurate baseline size from fundoscopy or FA, it is possible to compare an OCTA image with its matching OCT reflectance image to reveal the extent of an individual’s FAZ enlargement. This is achieved by comparing blood vessels between the en face reflectance and angiographic perfusion images, thus generating a differential image.

14There are only a limited number of studies examining change in FAZ size over time, but some data indicate that FAZ size and appearance can change significantly even between just two clinic visits (p < 0.05; mean visit interval 3.2 months), though this could be due to differences in image segmentation or receipt of treatment by some patients between visits. FAZ size changed in eyes either with and without treatment, making it difficult to assess which changes were the result of an intervention, disease progression, or segmentation variability. In any case, because of the variability in FAZ size in the healthy populations, the use of FAZ area alone to stratify patients by DR severity is highly challenging, limiting its value both in the clinic and as a stratification factor for studies. Changes in FAZ size, however, may be a potential study endpoint.

15An alternative measure to FAZ area size is FAZ circularity index (4πA/P2; where A is FAZ area and P is FAZ perimeter). The normal shape of the FAZ approximates a circle, with a continuous loop of capillary around the perimeter. The FAZ is known to become more tortuous or acircular in patients with DR, as the dropout of capillaries at various locations increases the perimeter reducing its circularity. This suggests that FAZ circularity is a more sensitive metric in patients with DR than FAZ area alone. However, this also suggests that the FAZ circularity is more sensitive to factors associated with OCTA image acquisition variability than FAZ area. The inter-person variability of FAZ circularity is lower than that of FAZ area, which may make circularity more suited for the assessment of DR severity than FAZ area when no longitudinal data are available. Often, patients with severe DME, which commonly occurs in severe DR, have greater FAZ acircularity. Thus, the presence of DME may confound the utility of FAZ circularity to grade DR severity. That said, changes in FAZ parameters over time (including FAZ circularity) measured using OCTA may be useful as a meaningful parameter in clinical trials. Some work has shown that there are significant changes in FAZ circularity for patients with moderate-to-severe DR over a 1-year period, regardless of the presence of DME, but not in patients with no or mild DR. This suggests that changes in the FAZ over 12 months for patients with DR versus controls could be used to assess DR severity, with some evidence indicating that long-term changes in FAZ may be used to predict DR progression. Finally, it should be noted that, although the immediate repeatability of FAZ size measurement is high, with correlation coefficients >0.9, little is known about the repeatability over a longer period of time.

16Another alternative measure of FAZ-related macular ischemia may be 3D para-FAZ vessel density (Fig. 2). This FAZ measurement is based on an estimate of FAZ size before the onset of DR, termed theoretical baseline FAZ (tbFAZ). The boundary of tbFAZ is defined based on the typical relationship between ganglion cell complex thickness (from structural OCT) and FAZ boundaries (from OCTA) in the normal population. With the addition of deep-learning-based macular fluid segmentation, tbFAZ can be estimated in eyes with DME. Estimation of tbFAZ is followed by the construction of a para-FAZ volume and calculation of projection artifact-resolved 3D vessel density. Compared with conventional FAZ measurements, 3D para-FAZ vessel density has demonstrated greater correlation with DR severity and higher diagnostic sensitivity in differentiating DR of various severities.

17High-resolution OCT and OCTA volumes are required to develop robust parafoveal or para-FAZ ischemia metrics. With further improvement of resolution, OCTA may better reveal anatomical and vascular details, while also allowing the development of sensitive biomarkers. Ideally, FAZ should be assessed using a full-retinal slab to limit segmentation errors and confounding factors, such as center-involved DME. Furthermore, it is critical that all OCT images of the FAZ be checked not only for comorbid conditions, such as epiretinal membrane and DME, but also for low image quality, low resolution, artifacts, and segmentation errors. The inclusion of additional parameters when measuring FAZ area, such as 3D para-FAZ vessel density and FAZ circularity, would enrich the data and enhance the accuracy and insightfulness of the assessment of patients with DR.

18The relatively low density of capillaries in the primate retina is the result of a balance between vascularization, to support the metabolic needs of visual information processing, and limitation of the number of blood vessels, to avoid hindrance of the passage of photons to the retina. This means that there is little capacity for a reduction in retinal vessel density without a direct impact on vision; therefore, the density of the retinal vascularity is a useful metric for estimating visual function in the macular area.

19Vessel density is a quantitative metric that approximates the number of vessels occupying a particular region of the retina. This can be evaluated in various ways. It may be assessed as the proportion of vessel area with blood flow over the total measured area (vessel area density or flow density) or as the total vessel length in an area of interest (vessel length or skeletonized vessel density). As retinal vasculature is known to change with retinal pathology, measuring vessel density may provide insight into microvascular diseases like DR and, in particular, ischemic complications such as DMI. Furthermore, unlike FAZ area, vessel density is not associated with age or sex, making it a potentially more reliable measure.

20There are a number of ways to measure vessel density, but the most common metrics are vessel area density and vessel length density. Skeletonization is a process in which the width of each detected vessel is reduced to 1 pixel in size; following this, total vessel length is calculated. In theory, this reduces the impact of image resolution on the apparent vessel caliber, but images with low signal strength can lead to discontinuity in the skeletonized vascular network. There are advantages to both skeletonized and non-skeletonized methods. Skeletonization reduces the weighting of large blood vessels and may therefore increase sensitivity to microvascular changes. Conversely, non-skeletonized methods provide more physiologically relevant information about the overall vasculature as vessel size is accounted for.

21Several studies have shown that decreased vessel density across the capillary plexuses correlates with increasing severity of nonproliferative diabetic retinopathy (NPDR). In patients with PDR, there is likewise a clear reduction in vessel density across all capillary plexuses (Fig. 3, previously unpublished data; the published method can be found in). Interestingly, some research has shown that patients with diabetes but without DR have higher capillary density than non-diabetic controls, potentially as a response to increased metabolic demand. Taken together, these specific data suggest that a decrease in capillary density can indicate disease progression, while an increase in density may indicate disease onset. However, this remains an area of debate, as other work has shown an increase in capillary density upon progression from moderate to severe DR, which has been hypothesized to be correlated with an increase in VEGF burden resulting in vessel dilation. More research is needed to establish the relationship between DR stage and capillary density, with the understanding that, in general, worsening disease is associated with capillary dropout and decrease in density.

22Decreased capillary perfusion and vessel density, including in the SVC and DVC, are also specifically linked to ischemia, disorganization of the retinal inner layers (DRIL), and neurodegeneration visible as ganglion cell layer thinning. Furthermore, decreased vessel density in the SVC may be associated with the development of DME. Importantly, changes in SVC vessel density may be more reliable than measures in the DVC, as they are less susceptible to noise and projection artifacts, even when artifact-removal algorithms are used.

23To quantify vessel density in intermediate and deep capillary plexuses, a reliable algorithm should be applied to remove the projection artifacts. To generate a reliable vessel density map, it is also necessary to compensate for the effect of reflectance variation and remove any invalid areas caused by shadow artifacts. Furthermore, if there is a strong bulk motion presence, inclusion of a preprocessing step to subtract the bulk motion is a prerequisite. With the assistance of artificial intelligence, high-resolution capillary maps can be reconstructed and enhanced to facilitate reliable vessel density measurements, although this is very much an active work in progress. Using this technology, definite and clear capillary plexuses can be reconstructed from large field of view scans (e.g. 6 × 6-, 9 × 9- or 12 × 12-mm scans) that have been under-sampled.

24One major issue in the reliability of vessel density metrics is the ability to accurately segment the individual capillary plexuses in eyes that have structural abnormalities related to DR, such as the presence of edema or ischemic-related structural changes. Therefore, the most repeatable and reproducible metric may be the evaluation of full-thickness retinal vessel density, which provides data for all retinal layers. This facilitates the alignment of longitudinal scans within patients and the distinction of each retinal layer even in the presence of pathology. Unfortunately, the inbuilt software of OCTA devices produced by some OCT manufacturers does not support the assessment of full-thickness retinal vessel density, which means that non-standardized custom software is typically required. Another factor of increasing importance when assessing vessel density using OCTA is the image field of view. Capillary density in the DVC decreases toward the periphery of the retina. As OCTA technology has developed, a number of commercially available devices now offer wide-field and ultrawide-field OCTA (12 × 12 mm; 12 × 9 mm; 15 × 9 mm; 23 × 20 mm). Research has also shown that OCTA metrics in the retinal periphery have high repeatability and are therefore potentially more reliable between visits than other imaging methods. As wide-field technology becomes more commonplace, the assessment of peripheral vessel density may become a more important focus for research investigating DR.

25Another metric, the intercapillary area, has been evaluated in patients with DR and may be more sensitive to vessel dropout than vessel density. However, it is also more dependent on image quality and therefore may show more variability across visits.

26Retinal vessel density and blood flow are closely related metrics. As with vessel density, the retinal blood supply operates within a narrow margin of error and must locally adjust for the high metabolic demands of the retinal neurons. Therefore, measures of blood flow can indicate when and where the retina is under metabolic stress or lacks sufficient perfusion to carry out its function. OCTA images indicate either the presence or possible absence of or slowing of blood flow. Changes in blood flow are central to the pathology of complications of retinal diseases, such as DME and DMI as well as microaneurysms. The detection of microaneurysms using OCTA via adaptive optics scanning laser ophthalmoscopy (AOSLO) may be influenced by intramicroaneurysmal turbulence. Recent studies highlighted the value of an en face image-averaging process in improving image quality and reducing background noise. This process has been found to enhance the visualization of microaneurysms with intermittent blood flow and increase the detection ratio in eyes with DR. In addition, utilizing extended interscan times allows for the detection of slower capillary flows that can enhance the detection of microaneurysms in OCTA scans. In terms of morphological assessment, there is a correlation between the morphological appearance of microaneurysms on OCTA and FA leakage. However, a single OCTA scan may depict microaneurysms differently for each acquisition. This variability can be mitigated by OCTA averaging, which has been shown to stabilize the visualization of retinal microvasculature. A recent analysis used these averaged OCTA images to demonstrate a correlation between microaneurysm morphology, DR stage, and FA leakage. One study suggested that OCT-identified microaneurysms may be able to be classified according to co-localized OCTA flow signal into active and inactive types. This study suggested that active microaneurysms are more likely to be associated with diabetic macular edema than those without flow.

27Variable interscan time analysis (VISTA) produces maps of relative blood flow, coded by color, that can indicate where in a specific image blood flow is relatively altered. Although the technique does not yet provide absolute measures of flow, cannot quantify flow, and can be affected by shadowing projection artifacts and motion artifacts, the values retrieved are repeatable between scans. Recent work confirms that VISTA can be used to detect relative blood-flow speeds in microvascular changes associated with DR. Changes in adjusted flow index (the average decorrelation value of all pixels in the OCTA image that are above the noise threshold) in the capillary plexuses correlate with DR progression, with measurements at the DVC typically declining with increasing DR severity, while the SVC remains stable. Interestingly, some OCTA data indicate that, following panretinal photocoagulation, capillary-layer flow metrics increase in patients with high-risk PDR. It is also important to note that the repeatability of vascular density measurements in the three plexus layers, especially the ICP and DCP, decreases with an increase in the size of the OCTA scan and the presence of macular edema. A study using ultrahigh speed, swept-source optical coherence tomography (SS-OCT) angiography demonstrated the occurrence of retinal and choriocapillaris microvascular abnormalities in all stages of diabetic retinopathy, even in diabetic patients without retinopathy. An analysis of the association between DR severity and macular choriocapillaris flow deficit percentage found that choriocapillaris flow impairment corresponds to DR severity, with all regions of the choriocapillaris significantly affected. To evaluate and quantify this accurately, algorithms need to be developed that can distinguish choriocapillaris flow deficit from shadowing. Overall, measurement of flow remains an emerging field that may show promise in patients with macular changes related to DR. However, these metrics are currently not available on commercial OCTA devices.

28Extrafoveal avascular areas are defined as any region of avascularity outside the foveal region (e.g., Fig. 4). Within the macula (Fig. 5), these are usually defined as being outside the central 1 mm circle of the SVC/DVC, which defines the fovea. Avascular areas outside the fovea but within the macula appear to be associated with risk of progression and vision loss.

29Detecting avascular areas has been traditionally done using FA, but leakage from the microvasculature can lead to inaccurate evaluations. OCTA, on the other hand, allows for the quantitative detection of non-perfused areas without the confounding effect of leakage. Measurement of such areas is relatively straightforward and can provide a repeatable and reproducible automated metric for the evaluation of retinal ischemia. Compared with analyses within non-segmented slabs that comprise the entire inner retina, extrafoveal avascular areas in the macula can be better assessed by dividing the retina into three slabs: SVC, ICP, and DCP (Fig. 5). In addition, when avascular area is measured in three separate slabs, it is superior at detecting DR at different severities than the measurements from a commercially available vessel density in 2-layer scheme. However, such metrics, while more sensitive, are also more dependent on image quality and therefore may be prone to issues with repeatability and reproducibility.

30In patients with severe DR, areas of capillary dropout are clearly visible in OCTA images (Fig. 5). Research has demonstrated that extrafoveal avascular areas detected using OCTA are significantly associated with DR severity (p < 0.0001), progression (p = 0.04), and treatment requirement (p = 0.002) over a 1-year period. As more OCTA data on extrafoveal avascular areas both within and outside the macula are collected, it may be possible to collate a database of pathology-stratified extrafoveal avascular area measurements. This database could then be used to enhance the definitions of disease severity in DR.

31Outside the macula, the retinal plexuses merge as the retina becomes thinner with greater eccentricity, and so capillary dropout is more reliably visualized using en face OCTA of the entire inner-retinal slab. Wide-field OCTA montages capturing extrafoveal avascular areas outside the macula show that the size of these areas correlates with DR severity and visual acuity.

32With wide-field OCTA becoming more accessible, the assessment of extrafoveal avascular areas is likely to become a focus of research in DR. Areas of midperipheral non-perfusion may prove more sensitive than macular non-perfusion for stratifying and staging DR. Recent research examining deep-learning approaches to distinguish avascular areas from signal-reduction artifacts have shown promise and can provide reliable measurements across a range of different scan qualities. However, although it is possible to analyze three layers of the macula with confidence, at present we are able to assess only the superficial layer of the periphery, making it difficult to evaluate the relative sensitivity of intra- and extramacular regions.

33IRMAs are “torturous intraretinal vascular segments” (Fig. 6) that vary in size and can be almost undetectable. They are associated with severe NPDR (ETDRS ). Therefore, the classification and detection of IRMAs may provide insight into the disease status of patients with DR.

34Results from a secondary analysis of the CLARITY study using widefield fundus imaging suggested that IRMAs are improved by intravitreal anti-VEGF. However, this treatment response will need further validation using OCTA. On en face OCTA images, IRMAs present as dilated, irregularly structured vessels, surrounded by capillary dropout. OCTA is as sensitive as FA and more sensitive than color fundus imaging in detecting the features of IRMA; approximately 50% more IRMAs are detected on OCTA than on color fundus imaging. Furthermore, it is now possible to see differently shaped IRMAs using OCTA. Importantly, as a 3D imaging method, OCTA allows accurate layer-specific localization of IRMAs within the retina, improving their differentiation from neovascularization, which would be detected on the retinal surface and breaking through the internal limiting membrane.

35Pathological neovascularization is a hallmark of PDR and can be examined using OCTA; treatment-naïve patients with PDR typically have irregular proliferation of fine new vessels, and this can be used as an imaging marker for active neovascularization. The detection rate of neovascularization of the disc (NVD) and neovascularization “elsewhere in the retina” (NVE; classed as neovascularization occurring outside the optic disc) by OCTA is superior to color photos and FA. An OCTA-based classification of NVD into types 1–4 has been established: however, types 1 and 2 are clinically undetectable. NVE can also be examined in patients with DR using wide-field OCTA. Subclinical or suspected NVE can be detected by wide-field OCTA either by using a high-speed system or a montaged wide field of view from commercial OCTA devices, suggesting that OCTA may improve the clinical evaluation of DR.

36One study found that an increase in NVE is significantly associated with the degree of ischemia (p =0.009); furthermore, the type of growth pattern of NVE (“round” vs “ramified”) correlates with ischemia, with round growth resulting in a larger ischemic index than that for ramified growth. In addition, NVE measured by OCTA has been used to assess treatment response in patients with PDR. One limitation to the assessment of NVE by OCTA may be the field of view; even montaged OCTA scans do not consistently capture peripheral lesions. That said, the assessment of neovascularization patterns using OCTA can provide insight into not only the degree of pathological neovascularization in patients with DR but also the development of ischemia.

37Neurovascular coupling refers to changes in blood flow in response to localized neuronal activity. As OCTA reflects retinal hemodynamics, it may be a useful tool for assessing neurovascular coupling in the retina. Induced hyperglycemia has been shown to disrupt neurovascular coupling in healthy eyes; the evaluation of the status of neurovascular coupling in the retina could therefore provide insight into the degree of metabolic disturbance as a measure of diabetic-induced dysfunction.

38In 2020, OCTA was used to examine the extent of neurovascular coupling impairment present in patients with diabetes prior to the onset of visible anatomical changes associated with DR. In patients with diabetes without anatomical evidence of DR, altered neurovascular responses were similar to those recorded in eyes with mild NPDR (Fig. 7). This suggests that changes in neurovascular responses assessed using OCTA could detect changes that precipitate DR before the onset of identifiable structural alterations.

39BCVA is a commonly used, regulator-accepted clinical trial endpoint of visual function. Establishing anatomical metrics that correlate with BCVA could be useful as an objective surrogate endpoint for clinical trials as well as a proxy for predicting changes in vision, especially those that are not measured with conventional visual function testing methods. Although there are limited data examining the correlation of OCTA metrics with visual acuity, preliminary data indicate that there may be potential avenues worth further exploring, although some of the data are conflicting. In one study, OCTA 3D para-FAZ measurements were shown to significantly correlate with measures of BCVA (p < 0.01), as opposed to measurements such as FAZ area and perimeter, which in some studies do not. In another study, combining FAZ area with FAZ shape irregularity was shown to correlate with both DR grade and BCVA. Large baseline FAZ areas in the DVC revealed by OCTA have also been shown in one study to be predictive of worsening visual outcomes over the course of a year in patients with diabetes. Interestingly, a stronger correlation was found between visual acuity loss and changes in the vascular density of the deep plexus as compared to the superficial or intermediate plexus in eyes with DR. This suggests that different plexus layers might have different impacts on visual acuity loss in DR. Further research examining the correlation of OCTA with measures of visual acuity is necessary to validate the relationship.

40DR is also associated with a reduction in contrast sensitivity. Some data suggest that contrast sensitivity may be a more appropriate measure for tracking progressive vision loss than traditional visual acuity metrics. This suggests that an imaging measure that correlates with contrast sensitivity could provide more useful information than visual acuity alone.

41Microperimetry is a method that can measure the retinal function of specific locations in the retina. Given the similarity of en face microperimetry and OCTA, the two measures could be overlaid to look for relationships between vascular features and functional performance. To date, it is unclear whether there is a correlation between OCT metrics and microperimetry outputs in patients with diabetic eye disease; recent work identified a correlation between OCTA-assessed vessel density (but not FAZ area) and microperimetry-assessed retinal sensitivity in eyes with advanced DR, but not in eyes with lower grades of severity and in eyes with retinitis pigmentosa. By using microperimetry and OCTA quantitative flow density, Arima et al. showed that vascular hyperpermeability may inhibit the retinal sensitivity reduction in the non-edematous ischemic diabetic retina. These findings highlight the potential value of combining these innovative tools for a more comprehensive understanding of the functional and structural implications of diabetic eye diseases.

42Despite OCTA being a relatively recent innovation, a number of clinical trials have begun to incorporate OCTA metrics into assessments of DR as primary endpoints (Table 1). Additionally, many large ongoing trials in DR and DME have incorporated OCTA metrics as exploratory endpoints, including FAZ size, vessel density, and vessel perfusion; although this highlights the increasing interest in OCTA, it should be noted that it is often not mandatory for the sites to capture exploratory endpoint images, resulting in fewer data overall.

43Change in FAZ size is one of the most commonly selected OCTA-assessed measurements of DR progression (Table 1). Vessel density and perfusion have begun to gain traction as metrics and clinical trial endpoints to assess retinal disease progression (Table 1). As most of these trials are currently ongoing, the results are not yet available; however, when the data are released, they may reveal how useful the OCTA is in clinical studies.

44One potential issue with OCTA images taken during clinical trials is the degree of unevaluable data. One non-interventional study found that 53.5% of OCTA scans in clinical trials contain severe artifacts, substantially affecting the reliability of any quantitative assessment, while an analysis of three multicenter clinical studies of patients with DR found that 39% of scans lacked the required signal strength index to be considered useable images. However, another study found that over 90% of images obtained in a clinical trial setting were evaluable. Clinical trials in ophthalmology may often operate under sub-optimal retinal imaging conditions, as enrolled patients may have vision loss that leaves them unable to fixate or otherwise follow ideal protocols, resulting in poor-quality images. It is therefore critical that clinical trials are carefully planned to account for this during data collection and that additional time is allowed to assess OCTA quality and to re-image where necessary. In contrast to the findings of Lujan et al. and Holmen et al. the percentage of OCTAs that were excluded from the TIME-2b study was low; quality metrics for OCTA images were pre-stipulated and carefully monitored over time.

45Although FAZ measurements are relatively robust within individual patients, as discussed earlier there is considerable variability at the population level (Fig. 1). Some evidence suggests that this variability can be overcome by using metrics such as 3D para-FAZ vessel density.

46Using different OCTA devices can introduce considerable variability in results from quantitative metrics. While intra-individual repeatability is quite high when using the same device, there is generally only moderate-to-poor agreement for OCTA metrics between different device models. In one study, the repeatability of peripapillary capillary quantification differed significantly between four different OCTA device models (Spectralis, Optovue, Cirrus, and Triton) (p < 0.001), and significant differences in the evaluation of vessel visibility (p < 0.001) and motion artifacts (p < 0.001) have been found in a separate study examining the AngioVue (Optovue), AngioPlex (Zeiss), AngioScan (Nidek), and Spectralis (Heidelberg Engineering) (both 3 × 3- and 6 × 6-mm scans). Choice of commercial OCTA instruments should be carefully considered to ensure that the approach and format of vascular density and FAZ measurements is appropriate to the needs of clinicians. This is mainly due to the varying methodologies employed by different devices for segmenting the retinal plexus layers and detecting vascular flow on OCTA. Each commercial device utilizes a unique software and algorithm, leading to potential variations in the output measurements. That said, one study found no significant difference in measured vessel density between four OCTA device models (Zeiss, Optovue, Topcon, and Heidelberg; p = 0.2).

47Until more definitive comparisons between OCTA devices are performed, it is difficult to definitively say which devices are the most sensitive or which are able to provide the most accurate measures of disease pathology. The data reported to date suggest that the choice of device model strongly affects the resulting OCTA metrics and image quality and may affect clinical trials in which different devices are used across different centers.

48Visualization of the retinal vasculature plexuses in 3D is one of the main advantages of OCTA. However, artifacts can cause misleading results; for example, motion artifacts caused by eye movement can result in the false appearance of decreased vascular density or “repetition” of anatomical landmarks, such as the fovea (Fig. 8). Common DR comorbidities, such as DME, DMI, epiretinal membrane, and vitreomacular traction, often overlap, resulting in vessel misidentification, segmentation errors, and noise artifacts in OCTA. Patient movements causing bulk motion artifacts or failure of the operator to focus and center the scan can further affect image clarity (Fig. 9), hampering retinal vasculature identification and reducing the range of data captured. Regardless of imaging system or scan protocol, the most common severe OCTA artifacts present in clinical trial data (406 images from 234 eyes) are shadow (26.9%), defocus (20.9%), and eye movement (16.0%).

49Use of full-vascular-layer OCT removes the risk of projection artifacts, which can occur when the vasculature of more superficial layers project onto deeper layers during imaging. Projection artifacts can limit the accuracy of deeper-layer measurements; one research group has used capillary density-deviation mapping and a normative reference group to show that there is a significant reduction in parafoveal capillary density in patients with retinal pathology (including DR) versus healthy controls. However, as full-layer processing that incorporates both the superficial and deep microvascular layers, it does not allow the separate assessments of capillary dropout (avascular area) in individual vascular plexuses. Projection-resolved (PR)-OCTA offers a solution to address the deep-layer image artifacts without the limitations of full-layer OCT scans. PR-OCTA presents all three retinal vascular plexuses distinctly, increasing the accuracy of the visualization of deeper-layer capillary abnormalities that may otherwise be obscured. In addition, algorithmic removal of decorrelation noise due to bulk motion in OCTA imaging appeared useful for improving the signal-to-noise ratio, contrast, vessel density repeatability, and noise in the FAZ.

50In low-quality scans that have confounding signal attenuation due to shadow or defocus, low-signal areas can appear quite similar to avascular areas and are challenging to differentiate. Guo et al. have demonstrated a deep-learning algorithm (Fig. 10) that can automatically differentiate avascular areas from shadow artifacts.

51Several OCTA artifacts and image distortions come from operator error; it is therefore critical to have an experienced and well-trained operator to minimize the incursion of artifacts at the point of data collection. Consistent protocols for imaging patients, including counting aloud the duration for which a participant must keep their eyes open, explaining where the patient must look, and carefully adjusting positioning equipment (such as chin rests) can ensure that bulk motion and blink artifacts are reduced.

52After data have been collected, the standard procedure should include projection artifact removal to allow the clean separation of retinal plexuses. By viewing retinal plexuses separately, non-perfusion areas (capillary dropout) can be recognized more accurately than otherwise possible. It is also useful to view en face structural OCT images together with the corresponding OCTA images so that areas of shadowing (low OCT signal) can be recognized and distinguished from nonperfusion.

53Current limitations in understanding the pathophysiology of DMI have hampered our ability to determine the predictive metrics for anticipating the progression of DMI. That said, some recent research has used a normative database of OCTA data to develop a parafoveal capillary density-deviation mapping technique.

54A recent Delphi consensus effort aimed to standardize OCTA nomenclature around retinal vascular diseases, specifically focusing on OCTA parameter definition and the assessment of DMI severity. Unfortunately, experts were not able to agree on a single metric in either the initial survey or the subsequent Delphi rounds; terms used to describe OCTA continue to vary notably between individuals. There remains a pressing need for more data and more consensus around OCTA in DR.

55In order for OCTA to have true clinical utility, it is critical to establish a consensus among experts and to validate OCTA metrics through research and clinical trials for its use in DR. Similar to the endpoints used in geographic atrophy for pivotal clinical trials, anatomical endpoints have many advantages over functional endpoints. The first step is for regulators to accept that OCTA, as a measurement of blood flow, is a surrogate for retinal function. FAZ area and vessel density are clear markers of DR progression, but so far there are not enough data to correlate OCTA metrics with standard measures of visual function, such as visual acuity. Therefore, they are not yet acceptable as surrogate markers of visual function by regulatory authorities. Studies that validate the correlation between OCTA metrics and functional measures of vision will broaden the use of OCTA in research and clinical settings. The ultimate goal is to establish an anatomical endpoint based on OCTA that can be used in pivotal trials of new therapies.

56As OCTA remains a relatively novel technology, it is important that there is continued research focus on improving image quality, particularly in the presence of common complications of DR that can confound clear measurements, such as DME. At present, standardized commercial software specifically for the imaging of DR or for OCTA in general that can address these issues is yet to emerge. However, the standardization of analysis algorithms between machines and research centers will facilitate data sharing and enable comparisons to be made between techniques and analyses. A 1-day workshop promoting the adoption of ocular imaging standards among the National Eye Institute (NEI), US Food and Drug Administration (FDA), and Office of National Coordinator for Health Information Technology (ONC) was recently conducted, which highlights the current focus on this topic. Automated image grading and automated image quality checks may address some concerns about the use of OCTA in clinical settings, reducing the need for specialized individual training while also providing a standard set of recognizable parameters for disease assessment.

57DR is a microvasculopathy, and OCTA signal intensity is tied to blood flow; theoretically, this positions OCTA as the ideal method for assessing DR. An increasing body of research indicates that OCTA can be used to quantitatively assess DR and its complications, even before the onset of clinical lesions. Several features characteristic of DR pathology can be assessed using OCTA, including FAZ size, vessel density, changes in blood flow, and neovascularization. Furthermore, neurovascular coupling can be examined through OCTA assessment of retinal hemodynamics, and specific OCTA parameters have been linked to measures of visual function (e.g., BCVA). Therefore, OCTA presents a large and promising toolkit for clinical trials of patients with DR and DR-related complications. Indeed, several trials have already begun to incorporate OCTA metrics.

58Despite its promise, it is clear that several challenges remain in the use of OCTA. As a relatively new technology, there is still a lack of standardized tools and procedures to address common imaging problems like artifacts, segmentation errors, and image focus. Although the use of specific scan types and patterns (including whole-vascular-layer OCT or PR-OCTA) and techniques (deep-learning algorithms and image reconstruction) can ameliorate these problems, without a consensus on the best methodology, it can be difficult to compare results between studies, centers, and machines. However, since its advent, OCTA technology has been steadily improving; the emergence of higher-speed OCT platforms with a wider field of view is enabling more precise assessment of the peripheral retina for neovascularization and non-perfusion. Improved software that accurately delineates non-perfusion in the retinal plexuses without being affected by artifacts will make the assessment of ischemia more reliable. Assessment of macular ischemia remains a particular challenge because the FAZ size varies substantially even among the normal population. The development of algorithms that can accurately estimate the baseline FAZ area and loss of perfusion relative to that baseline will be essential to the clinically meaningful assessment of macular ischemia.

59Here, we propose the following methodology for the effective use of OCTA in clinical trials of DR. Using a full-retinal slab OCTA image can limit segmentation errors and confounding factors, such as center-involved DME. However, plexus-specific detection of avascular area (a.k.a. Non-perfusion area) in the macula is more sensitive than full-retinal slab detection of avascular area. Therefore, choice of OCTA slab thickness for patients with DR should be guided by a combination of both comorbidity extent and metric of interest. In addition, given emerging data suggesting the importance of the peripheral retinal vasculature in assessing and predicting DR progression, wide-field OCTA imaging should be used. As intra-individual repeatability is quite high when using the same device type, but there is generally poor agreement for OCTA metrics between different device models, the same OCTA device should be used to measure a patient over time, and, where possible and until better conversion metrics can be developed, across centers for clinical trials. Clinical trial protocols should account for the relatively high degree of poor-quality data likely to result from sub-optimal imaging conditions and incorporate time and processes for assessing OCTA image quality, re-training sites with an active dialogue between reading centers and sites, and re-imaging patients where absolutely necessary. Finally, the use of automated methods and algorithms for OCTA image analysis, such as those that can reconstruct images, distinguish between areas of true and false signals, and produce quantitative metrics, such as FAZ area, will greatly improve the efficiency and standardization of results between studies.

60A number of informative OCTA metrics could be used to assess DR in future clinical trials. Assessing multiple measurements of the FAZ (area, acircularity, 3D para-FAZ vessel density) would provide an informative mix of quantitative and qualitative information that remains robust across longitudinal studies. In addition, while the precise relationship between vessel density, DR severity, and DR onset is under debate, there is a reasonable volume of evidence supporting a link between vessel density and DR, and vessel density may therefore be included in clinical trials of DR. Finally, with the advent of wide-field OCTA, the examination of extrafoveal avascular zones may become even more informative when assessing patients with early-stage DR. However, wide-field images remain limited by the currently available fields of view, leading to reduced density of data which can degrade measurement accuracy. Therefore, collecting higher-density, higher-resolution OCTA scans of the macula in combination with wide-field OCTA images of extrafoveal regions may provide the best sensitivity to early changes. Collecting these metrics as standard across trials would help to generate a large OCTA database of DR images, ultimately facilitating the quick and noninvasive evaluation of patients with diabetes at risk of developing DR.

61Use of a standardized image format across OCTA devices is critical to establish the large OCTA datasets necessary to answer outstanding questions on OCTA metrics in DR. At the time of writing, a collaborative ocular imaging standardization initiative by the NEI, FDA, and ONC is underway. Although the outputs of this workshop are yet to be published, this ongoing initiative should assist in establishing a common foundation of imaging standards that can be applied across trials. Furthermore, the American Academy of Ophthalmology has recommended that imaging device manufacturers adhere to a set of standard formats. Adherence to these recommendations will facilitate collaborative research, improve data quality, and ultimately improve clinical care for patients.

62When we have solved these issues and established a common set of OCTA procedures for use in clinical trials, it will be possible to realize the full potential of OCTA as a robust and regulator-accepted clinical trial endpoint.