Review of OCT Angiography Findings in Diabetic Retinopathy: Insights and Perspectives

Diabetes mellitus (DM), a disorder rapidly growing in prevalence, is linked to the retinal microvasculature complication diabetic retinopathy (DR). As one of the leading global causes of vision impairment and loss, imaging techniques to detect and monitor DR must continue to improve in order to address this growing burden. Optical coherence tomography angiography (OCTA) is a nascent imaging modality that generates three-dimensional visualizations of the retinal and choroidal microvasculature. Compared to fluorescein angiography, the gold-standard imaging modality for retinal vessels, OCTA offers the advantages of being non-invasive, quick, and able to resolve the multiple plexuses within the retina. Quantitative OCTA studies have explored parameters such as vessel density (VD), foveal avascular zone (FAZ), acircularity index, vessel tortuosity (VT), and fractal dimension (FD) amongst DR patients. This review synthesizes the main trends emerging from quantitative OCTA-based studies of DR and interrogates them within the context of DR pathophysiology. We offer a glimpse into how analysis techniques have shifted in the years since OCTA came into existence, while speculating on its future role in clinical practice.


Introduction
Diabetes mellitus (DM) is a substantial public health crisis that is projected to rise in prevalence, mortality, and burden throughout the next century. Current studies estimate that anywhere from 463 to 476 million individuals live with DM today, representing over 6% of the global population [1][2][3]. Forecasts from the International Diabetes Federation predict that global DM prevalence will grow to nearly 700 million individuals by the year 2045 [4]. DM is a major cause of morbidity and mortality, coming in as the 15th leading cause of years of life lost in 2015, demonstrating an increase of 25.4% from 2005 [5]. Costs of DM also represent a significant economic burden at 1.8% of the global gross domestic product (GDP) in 2015, with projections for DM-related costs reaching an estimated 2.2% of global GDP by the year 2030 [6,7]. With the global DM landscape becoming increasingly dire, so too are DM-related complications.
Poor glycemic control amongst individuals with DM results in damage to the systemic macro-and micro-vasculature, including the condition diabetic retinopathy (DR). DR can be divided into non-proliferative (NPDR) and proliferative (PDR) stages. Characteristic features of NPDR include intraretinal hemorrhages, hard exudates, microaneurysms, and capillary non-perfusion [8]. Hypoxic changes can cause progression to PDR, an advanced stage of the disease hallmarked by neovascularization, which is the abnormal growth of new vessels [9]. Neovascularization precipitates severe vision impairments when new vessels cause retinal detachment or vitreous hemorrhage [10]. DR affected 103 million adults in 2020, a number that is expected to grow to 160 million by the year 2045 [11]. Worldwide, DR was the 5th leading cause of moderate to severe vision impairment and the 5th leading cause of blindness in 2020 [12]. Diabetic macular edema, which can occur at any stage of DR, is characterized by fluid accumulation due to the occurrence of blood-retinal barrier breakdown and represents the leading cause of blindness amongst the diabetic population [13]. Given the rising prevalence of DM and DR and that vision impairment and blindness stemming from DR can be prevented with early detection and treatment in a vast majority of patients, the development of an efficient and cost-effective DR screening program represents a promising opportunity in the years to come.
Optical coherence tomography angiography (OCTA) has, in just a few short years, revolutionized the field of ophthalmic imaging. OCTA offers a quick, non-invasive, and depth-resolved method of imaging the retinal and choroidal micro-circulation. OCTA is an extension of optical coherence tomography (OCT), which is built on the principle of light interference [14]. A beam of light is applied to the retina and subjected to backscattering, reflection, or absorption by structures within the posterior segment of the eye. A reference beam of light is compared to a beam of light that "samples" the retina, with differences between these two light paths yielding a depth profile of the retina [15]. In OCTA imaging, a series of B-scans are captured at the same position within the retina, and differences in these successive B-scans are analyzed. Given that the retina is a static tissue, any differences between B-scans are expected to arise from the movement of red blood cells within the circulation only. These differences are termed the decorrelation signal, and a map of the decorrelation signal represents the vascular anatomy of the retina [16]. A complete three-dimensional view of the vasculature is captured by performing B-scans in a raster pattern that fully covers the retina. Since their introduction into the commercial market in 2016, OCTA instruments have been used in numerous studies that have furthered our understanding of a wide range of chorioretinal diseases [17,18]. Its application in the study of DR holds particular promise.
In this review, we evaluate the clinical utility of OCTA quantitative metrics as they relate to the diagnosis, staging, and the pathophysiology of diabetic retinopathy.

General Comparison of OCT Angiography and Fluorescein Angiography
Fluorescein angiography is considered the gold standard in evaluating the retinal microvasculature in cases of DR [19]. However, one of the main drawbacks of fluorescein angiography is that it requires the intravenous injection of dye, which causes anaphylactic reactions in rare cases [20,21]. OCTA, on the other hand, is completely non-invasive and acquires angiograms of the retinal and choroidal vasculature within seconds. This makes OCTA an appealing option to longitudinally track the progression of DR and its response to treatment. OCTA is also depth-resolved, giving it the ability to visualize the vasculature of the middle (MCP) and deep capillary plexuses (DCP) (Figure 1), while fluorescein angiography is mainly limited to the superficial capillary plexus (SCP) [22]. The SCP supplies the ganglion cell layer and inner plexiform layer, while the MCP is located at the border of the inner plexiform layer and inner nuclear layer, and the DCP is found between the border of the inner nuclear layer and outer plexiform layers [23,24]. Commercial OCTA machines utilize both spectral domain (SD-OCTA) and swept source (SS-OCTA), with the latter enabling better visualization of the choroidal circulation. Hence, OCTA is able to precisely locate changes occurring at each of the capillary plexuses as well as the choriocapillaris due to DR.
One of the limitations of OCTA is its comparably smaller field of view. Several studies have demonstrated the use of SS-OCTA, which has faster scanning speeds compared to SD-OCTA, to capture 12 × 12 mm single scan sizes with up to a 100-degree field of view [25][26][27]. However, this does not compare to the 200-degree field of view that can be captured with ultra-widefield fluorescein angiography [28,29]. This is an important limitation of OCTA devices, as it confines their applicability to diseases primarily affecting Figure 1. Optical coherence tomography angiography (OCTA) custom segmentation of the three capillary plexuses as seen by 3 × 3 mm en-face angiograms (top row) and cross sectional B-scans (bottom row). Green and red lines on B-scans correspond to segmentation boundaries used. (A) The superficial capillary plexus (SCP) was segmented from the internal limiting membrane to 10 µm above the inner plexiform layer; (B) The middle capillary plexus (MCP) was segmented from 10 µm above to 30 µm below the inner plexiform layer; (C) The deep capillary plexus (DCP) was segmented from 30 µm below the inner plexiform layer to 10 µm below the outer plexiform layer.

OCT Angiography Artifacts
OCTA is also susceptible to a variety of artifacts. Motion artifacts stem from the movement of a patient's eye, head, or body during the image acquisition process. These movements create a false decorrelation signal that is not representative of moving erythrocytes. While programs exist to correct for these movements, they too can generate artifacts that stretch or double vessels, which presumably artificially inflates quantitative OCTA metrics [30].
Segmentation artifacts, a consequence of errors in separating the vascular layers, have also been reported to occur in higher rates amongst patients with DM, NPDR, or PDR compared to healthy controls, representing another potential source of confounding [31,32]. These artifacts likely arise from the distorted or abnormal retinal architecture that can arise in states of ocular disease. While these segmentation artifacts can be corrected manually, this is an intensely time-consuming process [33].
Projection artifacts describe the appearance of superficially situated vessels in the deeper vascular layers of the retina. A variety of programs have been implemented to remove these projections occurring in the deeper capillary plexuses [34,35]. One would presume that the removal of projection artifacts would consistently decrease measurements of vessel density (VD) at the DCP. However, projection removal has had a varying effect on VD measurements at the DCP depending on the severity of DR, suggesting a likely interplay with the pathophysiology of the disease state [36]. Optical coherence tomography angiography (OCTA) custom segmentation of the three capillary plexuses as seen by 3 × 3 mm en-face angiograms (top row) and cross sectional B-scans (bottom row). Green and red lines on B-scans correspond to segmentation boundaries used. (A) The superficial capillary plexus (SCP) was segmented from the internal limiting membrane to 10 µm above the inner plexiform layer; (B) The middle capillary plexus (MCP) was segmented from 10 µm above to 30 µm below the inner plexiform layer; (C) The deep capillary plexus (DCP) was segmented from 30 µm below the inner plexiform layer to 10 µm below the outer plexiform layer.

OCT Angiography Artifacts
OCTA is also susceptible to a variety of artifacts. Motion artifacts stem from the movement of a patient's eye, head, or body during the image acquisition process. These movements create a false decorrelation signal that is not representative of moving erythrocytes. While programs exist to correct for these movements, they too can generate artifacts that stretch or double vessels, which presumably artificially inflates quantitative OCTA metrics [30].
Segmentation artifacts, a consequence of errors in separating the vascular layers, have also been reported to occur in higher rates amongst patients with DM, NPDR, or PDR compared to healthy controls, representing another potential source of confounding [31,32]. These artifacts likely arise from the distorted or abnormal retinal architecture that can arise in states of ocular disease. While these segmentation artifacts can be corrected manually, this is an intensely time-consuming process [33].
Projection artifacts describe the appearance of superficially situated vessels in the deeper vascular layers of the retina. A variety of programs have been implemented to remove these projections occurring in the deeper capillary plexuses [34,35]. One would presume that the removal of projection artifacts would consistently decrease measurements of vessel density (VD) at the DCP. However, projection removal has had a varying effect on VD measurements at the DCP depending on the severity of DR, suggesting a likely interplay with the pathophysiology of the disease state [36].

Comparison of Morphological Characteristics of Diabetic Retinopathy Seen by OCT and Fluorescein Angiography
In general, morphological features of DR that had been previously visualized by fluorescein angiography have been replicated by OCTA imaging. Microaneurysms are detected as focally dilated capillaries that appear as pinpoint hyperfluorescent lesions in fluorescein angiography and as saccular, pedunculated, or fusiform in OCTA imaging (Figure 2A,B) [37]. Multiple studies comparing the two imaging modalities have observed that fluorescein angiography detects a greater number of microaneurysms compared to OCTA [37][38][39][40][41]. Histopathological studies have identified subtypes of microaneurysms lacking blood flow [42]. Furthermore, measurements of relative blood speeds revealed slower flow in microaneurysms [43]. Given that OCTA relies on moving blood flow, these findings provide possible explanations for the lower detection rate of microaneurysms. An advantage of OCTA is that it enables the precise localization of microaneurysms. A custom algorithm has been developed that provides three-dimensional visualizations of microaneurysms, elaborating on their architecture and orientation [44].

Comparison of Morphological Characteristics of Diabetic Retinopathy Seen by OCT and Fluorescein Angiography
In general, morphological features of DR that had been previously visualized by fluorescein angiography have been replicated by OCTA imaging. Microaneurysms are detected as focally dilated capillaries that appear as pinpoint hyperfluorescent lesions in fluorescein angiography and as saccular, pedunculated, or fusiform in OCTA imaging (Figure 2A,B) [37]. Multiple studies comparing the two imaging modalities have observed that fluorescein angiography detects a greater number of microaneurysms compared to OCTA [37][38][39][40][41]. Histopathological studies have identified subtypes of microaneurysms lacking blood flow [42]. Furthermore, measurements of relative blood speeds revealed slower flow in microaneurysms [43]. Given that OCTA relies on moving blood flow, these findings provide possible explanations for the lower detection rate of microaneurysms. An advantage of OCTA is that it enables the precise localization of microaneurysms. A custom algorithm has been developed that provides three-dimensional visualizations of microaneurysms, elaborating on their architecture and orientation [44]. Neovascularization of the retina (NVE) or optic disc (NVD) are hallmarks of the proliferative stage of DR. In the later phases of fluorescein angiography, these vessels will demonstrate characteristic leakage of circulating dye [45]. However, leakage can mask clear resolution of neovascular complexes. OCTA's quick scan time prevents this effect from occurring, offering enhanced visualization of neovascularization. In OCTA, neovascularization appears as abnormally tangled vessels with flow above the optic disc (NVD) or above the internal limiting membrane in the retinal area (NVE) [46]. Numerous studies have demonstrated the effectiveness of OCTA to image both NVD and NVE [47][48][49][50][51][52][53][54]. Other studies have extended the use of OCTA as a tool to longitudinally track the therapeutic benefit of anti-VEGF agents or panretinal photocoagulation (PRP) in treating neovascularization in PDR [55][56][57][58][59]. Compared to ultra-widefield fluorescein angiography, a wideangle OCTA device demonstrated 100% sensitivity and 97% specificity in the detection of neovascularization, albeit in a limited number of eyes [26]. In a separate study of 37 eyes, Neovascularization of the retina (NVE) or optic disc (NVD) are hallmarks of the proliferative stage of DR. In the later phases of fluorescein angiography, these vessels will demonstrate characteristic leakage of circulating dye [45]. However, leakage can mask clear resolution of neovascular complexes. OCTA's quick scan time prevents this effect from occurring, offering enhanced visualization of neovascularization. In OCTA, neovascularization appears as abnormally tangled vessels with flow above the optic disc (NVD) or above the internal limiting membrane in the retinal area (NVE) [46]. Numerous studies have demonstrated the effectiveness of OCTA to image both NVD and NVE [47][48][49][50][51][52][53][54]. Other studies have extended the use of OCTA as a tool to longitudinally track the therapeutic benefit of anti-VEGF agents or panretinal photocoagulation (PRP) in treating neovascularization in PDR [55][56][57][58][59]. Compared to ultra-widefield fluorescein angiography, a wide-angle OCTA device demonstrated 100% sensitivity and 97% specificity in the detection of neovascularization, albeit in a limited number of eyes [26]. In a separate study of 37 eyes, widefield-OCTA was found to have 79% sensitivity and 96% specificity for neovasculariza-tion detection compared to fluorescein angiography [60]. Collectively, these studies suggest that OCTA is a viable alternative to fluorescein angiography for identifying and tracking neovascularization during PDR. In fact, it may be preferentially indicated in circumstances of repeated use due to its advantages in the fields of comfort and image acquisition speed.

Quantitative OCT Angiography Biomarkers in Diabetic Retinopathy
A wealth of quantitative OCTA metrics have been described as possible biomarkers for the detection and progression of both DM and DR. These parameters include measurements of vessel density (VD) or vessel length density (VLD); the area, perimeter, and acircularity of the foveal avascular zone (FAZ); fractal dimension (FD); vessel tortuosity (VT); and flow deficits and areas of non-perfusion within the choriocapillaris. While measurements of VD and FAZ parameters are built into some commercially available OCTA devices, they are inconsistently used in the literature, with some groups opting to use custom software packages in their analysis [61][62][63][64]. These biomarkers show immense promise for detecting DR, monitoring disease severity, and tracking response to treatment. Both thoroughly reported and emerging OCTA findings in the settings of DM and DR are discussed below.

Perfusion and Vessel Density
Perfusion (PD) or vessel density (VD) measurements are some of the most widely reported OCTA parameters in the literature. VD relies upon a binarization algorithm, whereby the total area of pixels corresponding to blood vessels is calculated. This value is then divided by the total area of the scanned area of interest. Measures of skeletonized VD or vessel length density (VLD) binarize the image and perform an additional skeletonizing process that reduces the width of each blood vessel to 1 pixel. The skeletonized VD then quantifies the proportion of cumulative vessel length within the area of interest. This process has several advantages and limitations compared to the VD measurement. Skeletonized VD equally weighs the contribution of large vessels and smaller capillaries to density measurements and may, thus, better capture changes that are occurring in the retinal microvasculature compared to a VD measurement [65]. Given that DR manifests primarily as a disease of the microvasculature, this could be especially advantageous for detecting changes occurring at the capillary level [66]. Without additional information, VD measurements may prove difficult for interpreting changes in the vasculature correctly. Compensatory vasodilation can be observed with capillary drop-out, failing to noticeably alter VD. However, the skeletonizing process loses information about the size of blood vessels, so it does not capture the full picture either. An underreported parameter in the literature that may solve this dilemma is vessel diameter index (VDI), which is calculated by dividing the area occupied by blood vessels in the binarized image by the length of blood vessels in the skeletonized image. This parameter is, thus, a measure of the vessel caliber and can be easily localized to pinpoint areas of vascular abnormalities [65]. Vessel caliber could be used to track a number of physiologic processes, such as the release of vasoactive substances, metabolic and pressure autoregulation, and hypoxia [67][68][69]. Furthermore, this measurement may be uniquely primed to detect regional disparities between the central macula and peripheral retina when it comes to arteriolar diameter responses to ischemia [70]. With this in mind, the trio of OCTA-based VD, skeletonized VD, and VDI measurements collectively provide robust insights into the vasculature changes occurring in DR.
In all three capillary plexuses (SCP, MCP, and DCP), decreasing VD and VLD have both been found to be associated with increasing severity of DR ( Figure 3) [71][72][73][74][75][76][77][78]. Furthermore, diminishing VD has been linked to poorer visual acuity in patients with DR, heightening the functional significance these parameters may bear out [79,80]. Studies of OCTA metrics utilize a variety of analysis techniques and devices (including SD-OCTA, SS-OCTA, and Widefield-OCTA); segmentation boundaries of the capillary plexuses; and patient populations. Initial studies focused on the SCP and DCP as sites of interest, as manual segmentation techniques were not well-described at the time, nor was projection-artifact software fully optimized and implemented. Following SS-OCTA imaging of a 3x3mm area centered on the fovea, Tang et al. used an automated customized program to analyze the VD and VDI at the SCP of 434 eyes from 286 patients with diabetes [71]. They found decreasing VD and increasing VDI were associated with worsening severity of DR, suggesting capillary drop-out accompanied by vasodilation [71]. Similarly, Nesper et al. and Li et al. found decreasing VD at the parafovea (defined as the area between an inner 1 mm ring and outer 2.5 mm or 3 mm ring centered on the fovea) of the SCP and DCP as DR staging advanced with the use of a SD-OCTA device [73,74]. Nesper et al. also identified that VD at the DCP correlated most strongly with DR severity [74]. In the segmentation scheme used by both groups, the DCP supplies the outer plexiform layer, a site of communication between horizontal cells and the inner segments of photoreceptors [24,73,74]. In studies using rat and macaque models, the outer plexiform layer was identified as having some of the highest oxygen demands in the retina, rendering this area especially susceptible to the effects of ischemic damage [81,82]. In humans, areas of photoreceptor dysfunction and disorganization closely align with areas of non-perfusion within the DCP, leading to diminished retinal sensitivity [83,84]. Hence, changes in perfusion at the DCP, as detected by OCTA, may show the first signs and predict the development of DR.
OCTA metrics utilize a variety of analysis techniques and devices (including SD-OCTA, SS-OCTA, and Widefield-OCTA); segmentation boundaries of the capillary plexuses; and patient populations. Initial studies focused on the SCP and DCP as sites of interest, as manual segmentation techniques were not well-described at the time, nor was projectionartifact software fully optimized and implemented. Following SS-OCTA imaging of a 3x3mm area centered on the fovea, Tang et al. used an automated customized program to analyze the VD and VDI at the SCP of 434 eyes from 286 patients with diabetes [71]. They found decreasing VD and increasing VDI were associated with worsening severity of DR, suggesting capillary drop-out accompanied by vasodilation [71]. Similarly, Nesper et al. and Li et al. found decreasing VD at the parafovea (defined as the area between an inner 1 mm ring and outer 2.5 mm or 3 mm ring centered on the fovea) of the SCP and DCP as DR staging advanced with the use of a SD-OCTA device [73,74]. Nesper et al. also identified that VD at the DCP correlated most strongly with DR severity [74]. In the segmentation scheme used by both groups, the DCP supplies the outer plexiform layer, a site of communication between horizontal cells and the inner segments of photoreceptors [24,73,74]. In studies using rat and macaque models, the outer plexiform layer was identified as having some of the highest oxygen demands in the retina, rendering this area especially susceptible to the effects of ischemic damage [81,82]. In humans, areas of photoreceptor dysfunction and disorganization closely align with areas of non-perfusion within the DCP, leading to diminished retinal sensitivity [83,84]. Hence, changes in perfusion at the DCP, as detected by OCTA, may show the first signs and predict the development of DR. Studies employing widefield SS-OCTA demonstrated that individuals with DR have lower VLD in the SCP and DCP compared to non-DM controls [72,75]. Mastropasqua et al. analyzed a wide field of view (15 × 15 mm scan area) for changes in VLD at an inner 5.5 mm macular region and in 4.5 mm circles peripheral to the macula [72]. Patients with advanced DR, classified as a moderate/severe NPDR group and a PDR group, had significantly lower VLD in the 5.5 mm macular area of the SCP and DCP when compared to respective non-DM and DM groups without retinopathy [72]. Significantly lower VLD persisted in the peripheral area temporal to the macula in groups classified as having advanced DR compared to non-DM and DM subjects without retinopathy [72]. Similarly, Hirano et al. used a widefield SS-OCTA system to measure VLD in a 12 × 12 mm scan area, finding decreasing VLD in the SCP and DCP as DR severity worsened across subject groupings [75]. Hence, widefield OCTA is a valuable tool for evaluating changes in the peripheral retina.
Innovation of projection artifact removal software and the introduction of novel segmentation schemes have enabled the visualization and the subsequent quantification of the MCP ( Figure 1B) [35,[85][86][87]. Previously, most automatic segmentation by commercial OCTA devices incorporated the MCP into the SCP. Analysis of the MCP as a distinct vascular layer opened a window of opportunity to understand how DR may have spatially Studies employing widefield SS-OCTA demonstrated that individuals with DR have lower VLD in the SCP and DCP compared to non-DM controls [72,75]. Mastropasqua et al. analyzed a wide field of view (15 × 15 mm scan area) for changes in VLD at an inner 5.5 mm macular region and in 4.5 mm circles peripheral to the macula [72]. Patients with advanced DR, classified as a moderate/severe NPDR group and a PDR group, had significantly lower VLD in the 5.5 mm macular area of the SCP and DCP when compared to respective non-DM and DM groups without retinopathy [72]. Significantly lower VLD persisted in the peripheral area temporal to the macula in groups classified as having advanced DR compared to non-DM and DM subjects without retinopathy [72]. Similarly, Hirano et al. used a widefield SS-OCTA system to measure VLD in a 12 × 12 mm scan area, finding decreasing VLD in the SCP and DCP as DR severity worsened across subject groupings [75]. Hence, widefield OCTA is a valuable tool for evaluating changes in the peripheral retina.
Innovation of projection artifact removal software and the introduction of novel segmentation schemes have enabled the visualization and the subsequent quantification of the MCP ( Figure 1B) [35,[85][86][87]. Previously, most automatic segmentation by commercial OCTA devices incorporated the MCP into the SCP. Analysis of the MCP as a distinct vascular layer opened a window of opportunity to understand how DR may have spatially unique impacts. In general, studies have found that trends within the MCP mirrored those occurring at the SCP and DCP; as DR severity increased, VD and VLD both decreased [76][77][78]. Interestingly though, DR severity had different effects on the vascular layers. Onishi et al. studied the adjusted flow index (AFI), an approximation of blow flow, at the three capillary plexuses of the retina [76]. They noted that while AFI failed to decline substantially at the SCP as DR severity progressed amongst the study groups, it did decrease steeply at the MCP and DCP; significant differences between the slope of the SCP AFI regression line and the slopes of both the MCP and DCP AFI regression lines were confirmed with linear mixed-effects regression models [76]. Ashraf et al. used OCTA to analyze the VD and VLD of all three capillary plexuses, observing that early DR manifested in changes primarily affecting the MCP and DCP, while advanced DR impacted the SCP more dramatically [77]. Collectively, these studies support segmentation schemes that analyze the SCP, MCP, and DCP separately across the spectrum of DR severity.
Some OCTA studies found diminished VD in the eyes of DM patients without DR compared to the eyes of healthy controls in the SCP, MCP, or DCP [76,[88][89][90]. Others, however, have not found significant differences in VD between non-DM and DM subjects without DR at the SCP or DCP [74,91,92]. These varying results likely stem from differences in one or more of study demographics, correction for co-variates, segmentation boundaries, and OCTA devices across studies. A unique study observed an increase in perfusion density at the full-retinal slab in the eyes of DM subjects without DR compared to healthy controls after the removal of non-capillaries from image analysis [93]. Rosen et al. subtracted larger, non-capillaries from angiograms and analyzed the area perfused by capillaries (PCD) at seven 200 µm wide annuli starting at the margin of the FAZ; in each of these seven annuli, PCD was significantly higher in the DM group without DR than in the non-DM group [93]. This finding may be explained by compensatory vasodilation occurring in capillaries, a hypothesis consistent with other studies that found an increase in vessel caliber with worsening DR [71,94]. Whether OCTA measures of perfusion can distinguish between non-DM and DM groups remains unclear. Further prospective and longitudinal studies may help to understand microvascular changes occurring in those that develop DM.

FAZ Area and FAZ-Related Measurements
The foveal avascular zone (FAZ) is the central area of the macula devoid of retinal capillaries. Early studies that sought to quantify the area (mm 2 ) of the FAZ relied upon fundus fluorescein angiography. The use of fluorescein angiography revealed significantly larger FAZ circumference and diameter in DM subjects compared to controls, as well as increasing FAZ size and irregularity with worsening retinopathy [95,96]. The area and perimeter of the FAZ are readily measured using OCTA, oftentimes through built-in software. In general, there is strong agreement between OCTA and fluorescein angiography when it comes to the assessment of the FAZ area in patients with DR [38,97]. FAZ area may also be an important biomarker for visual functioning and prognosis in the setting of disease. In patients with DR, a positive correlation between best-corrected visual acuity (LogMAR) and FAZ area has been found by several studies [80,98,99]. In other words, enlarged FAZ areas are associated with poorer vision.
As corroborated extensively in the literature, FAZ size increases with the progression of DR (Figure 4) [71,74,78,[100][101][102][103]. Several of these studies evaluated the FAZ of the SCP and DCP separately. This approach may be erroneous, as Campbell et al. demonstrated that the three capillary plexuses coalesce to define the perimeter of the FAZ [24]. Measuring the FAZ from a full-thickness slab with all retinal microvasculature visualized could better approximate its area. This is not the only issue that arises when using FAZ-based metrics. Within normal eyes, the size of the FAZ is highly variable, influenced by patient age, sex, race, central foveal thickness, and axial length [104][105][106][107][108][109]. Foremost among these factors is axial length, a measure of the length of the eyeball from the corneal surface to the retinal pigment epithelium. Longer axial lengths are associated with myopia and can affect the size of the FAZ. Those with longer axial lengths (e.g., high myopes) will have a seemingly smaller FAZ, while those with shorter axial lengths will have a larger appearing FAZ. Both Linderman et al. and Sampson et al. have demonstrated a need to correct for magnification effects stemming from axial length [104,110]. After correction, Sampson et al. found up to a 51% increase in the FAZ area for those with myopia and up to a 20% decrease in the FAZ area for those with hyperopia [110]. Studies that don't implement correction protocols could be confounded by control and experimental groups with different axial length distributions.
affect the size of the FAZ. Those with longer axial lengths (e.g., high myopes) will have a seemingly smaller FAZ, while those with shorter axial lengths will have a larger appearing FAZ. Both Linderman et al. and Sampson et al. have demonstrated a need to correct for magnification effects stemming from axial length [104,110]. After correction, Sampson et al. found up to a 51% increase in the FAZ area for those with myopia and up to a 20% decrease in the FAZ area for those with hyperopia [110]. Studies that don't implement correction protocols could be confounded by control and experimental groups with different axial length distributions. While axial length can be readily measured using optical biometry, this is not a builtin feature for commercial OCTA devices. Alternative parameters to FAZ area do exist, however, and include FAZ circularity and acircularity indices. These values are unitless quotients, meaning that they are not subject to the magnification effects that can stem from axial length, and they do not require correction. Acircularity index is the ratio of the perimeter of the FAZ to the perimeter of a circle with equal area; an acircularity index of 1 represents a perfect circle, while more irregular FAZ perimeters have a higher acircularity index. Circularity index is a similar measurement of the ratio of the FAZ area to the area of a circle with equal perimeter; a circularity index of 1 corresponds to a perfect circle, and more irregularity is found with decreasing circularity index. A positive correlation between FAZ acircularity index and DR severity has been found by multiple studies ( Figure  4) [93,[111][112][113], and a negative correlation between FAZ circularity and DR severity has also been identified [71,[113][114][115]. Krawitz et al. measured the FAZ acircularity index and FAZ axis ratio in type 2 DM subjects without retinopathy, with NPDR, and with PDR [111]. They defined axis ratio as the ratio between the major and minor axis of an ellipse fit with second-order moments; an elongated FAZ outline had greater eccentricity and, hence, a greater axis ratio [111]. After classifying the NPDR group into mild, moderate, and severe stages, Krawitz et al. used a generalized linear model to show a significantly increasing FAZ acircularity index and axis ratio with increasing disease severity from non-DM to PDR groups [111]. The researchers did not find significant differences in acircularity index or axis ratio between non-DM and DM subjects without DR, which is in agreement with other studies of its kind [93,111,113]. When axial length correction is not possible or feasible, FAZ acircularity or circularity parameters serve as comparable substitutes for assessing DR severity through OCTA. While axial length can be readily measured using optical biometry, this is not a built-in feature for commercial OCTA devices. Alternative parameters to FAZ area do exist, however, and include FAZ circularity and acircularity indices. These values are unitless quotients, meaning that they are not subject to the magnification effects that can stem from axial length, and they do not require correction. Acircularity index is the ratio of the perimeter of the FAZ to the perimeter of a circle with equal area; an acircularity index of 1 represents a perfect circle, while more irregular FAZ perimeters have a higher acircularity index. Circularity index is a similar measurement of the ratio of the FAZ area to the area of a circle with equal perimeter; a circularity index of 1 corresponds to a perfect circle, and more irregularity is found with decreasing circularity index. A positive correlation between FAZ acircularity index and DR severity has been found by multiple studies (Figure 4) [93,[111][112][113], and a negative correlation between FAZ circularity and DR severity has also been identified [71,[113][114][115]. Krawitz et al. measured the FAZ acircularity index and FAZ axis ratio in type 2 DM subjects without retinopathy, with NPDR, and with PDR [111]. They defined axis ratio as the ratio between the major and minor axis of an ellipse fit with second-order moments; an elongated FAZ outline had greater eccentricity and, hence, a greater axis ratio [111]. After classifying the NPDR group into mild, moderate, and severe stages, Krawitz et al. used a generalized linear model to show a significantly increasing FAZ acircularity index and axis ratio with increasing disease severity from non-DM to PDR groups [111]. The researchers did not find significant differences in acircularity index or axis ratio between non-DM and DM subjects without DR, which is in agreement with other studies of its kind [93,111,113]. When axial length correction is not possible or feasible, FAZ acircularity or circularity parameters serve as comparable substitutes for assessing DR severity through OCTA.

Fractal Dimension Analysis
The complexity of vessel branching in the retina has been studied using fractal dimension (FD). Early studies of FD in DR primarily utilized color fundus photographs. These studies reached conflicting conclusions, with some finding no association between FD and DR risk and/or severity [116], others finding increased FD to be associated with DR risk and/or severity [117][118][119], and others yet finding decreased FD to be associated with DR risk and/or severity [120,121]. These mixed results warranted further study, and OCTA has been used to add some clarity to the literature, with studies yielding more consistent findings of FD across the spectrum of DR severity.
FD at the SCP and DCP has been well characterized as negatively correlated with worsening DR severity, implying that branching patterns become less complex due to vessel dropout associated with DR progression [71,75,94,122,123]. In the setting of diabetic macular edema, lower FD was also correlated with poorer visual acuity, suggesting a possible use as a biomarker for visual functioning in DR [124]. Zahid et al. generated a measurement for FD at the SCP and DCP using the box-counting method, finding that DM eyes with DR had significantly lower FD compared to control eyes [122]. Kim et al. uncovered similar findings at the SCP, noting that mild NPDR, severe NPDR (a combination of moderate and severe NPDR subjects), and PDR groups all had significantly lower FD relative to controls [94]. At the DCP, though, significant differences in FD emerged only when comparing the PDR group to the healthy group [94]. Zahid et al. grouped five eyes with mild NPDR and eight eyes with PDR together, which prevented the comparison of different NPDR stages [122]. Due to the modest number of eyes included in both studies, larger sample sizes may be needed to detect the point in time at which significant differences in FD arise throughout the pathophysiological process of DR. FD may also have value in monitoring response to treatment. Fayed et al. compared the FD of perifoveal vessels in quiescent PDR eyes to PDR eyes also receiving PRP (PDR+PRP) [123]. Focusing on the SCP, the PDR+PRP group had a higher FD value than the PDR group, suggesting that this morphological change may be at least somewhat reversible [123]. FD may be a useful adjunct to other OCTA parameters in assessing DR severity, progression, and treatment response. However, analysis of FD at each of the three capillary plexuses represents a logical next step, as to our knowledge, this is not documented in the literature.

Emerging Insight into Vessel Tortuosity
Relatively few studies have used OCTA to examine vessel tortuosity (VT), a measure of retinal geometry that compares the length of a vessel's flow path to a straight line between the starting and ending points of the vessel's path. Studies using fundus photographs found VT to be increased among DR subjects compared to healthy controls or to DM subjects without DR [125,126], and longer diabetes duration was also associated with greater tortuosity of retinal arteriolar vessels [127]. These studies imaged the large vessels proximal to the optic disc, which left an opportunity for OCTA imaging to investigate changes occurring in the capillaries of the macula. Using SD-OCTA, Lee et al. noted a trend of increasing VT for all vessels in the SCP and DCP with worsening NPDR; progression from moderate-severe NPDR to PDR was marked by a significant decrease in VT [128]. Fayed et al. also investigated VT, focusing only on large perifoveal vessels of the SCP, while excluding capillaries from their measurements of VT [123]. They also found significantly increasing VT with DR severity, but did not find a significant decrease in VT amongst PDR patients compared to NPDR patients [123]. The discrepancy in this last finding likely stems from the pattern of vasculature analysis used by the two groups. The method employed by Lee et al. included capillaries, while that employed by Fayed et al. did not [123,128]. Profound capillary dropout in PDR could artificially deflate measurements of VT and account for the contrasting conclusions reached by the groups. A study of the cerebral cortex, which found vessels closer to the capillary bed to be more tortuous, supports this interpretation [129]. OCTA is a valuable and effective tool for monitoring VT, yet future studies are required to fully characterize the use of VT as a biomarker for DR.
Several theories may explain changes occurring to VT and fit within the pathophysiological process of DR. In a rat model deprived of oxygen, vascular endothelial growth factor (VEGF) signaling increased arteriolar tortuosity [130]. Furthermore, in eyes with central retinal vein occlusion, concentrations of aqueous VEGF were significantly correlated with the observed degree of venous tortuosity [131]. A similar model may be followed in DR, where VEGF is released in response to ischemia, promoting tortuous alterations to the microvasculature. A computational model found that tortuosity also improves responses to ischemic changes by increasing tissue oxygenation, especially in concert with anastomoses [132]. Alternatively, impaired autoregulatory responses mediated by endothelial cell and pericyte dysfunction, which are hallmarks of DR, could lead to mechanically unstable vessel walls susceptible to tortuous changes [133,134].

OCTA Findings at the Choriocapillaris
OCTA has also been used as a tool to study the choroidal vasculature, which is the primary blood supply to the highly metabolically active photoreceptors [135]. Indocyanine green angiography (ICGA) is the current gold standard for visualizing the choriocapillaris [136]. However, OCTA holds an advantage over ICGA in that it offers non-invasive, depth-resolved, and non-obscured imaging of the choriocapillaris. Furthermore, recent advances with SS-OCTA have enabled better visualization of the choroid. Compared to SD-OCTA, which uses a light source of approximately 850 nm, SS-OCTA uses a longer wavelength near 1050 nm that offers better penetration into the choroid [137,138]. Hence, OCTA wields the power to image both the retinal and choroidal circulation in a completely non-invasive manner.
Several parameters have been described to measure choriocapillaris perfusion or the lack thereof, including perfusion or vessel density, percent area of non-perfusion (PAN), and flow deficit percentage. Commercial OCTA machines, which can automatically quantify vessel density at the choriocapillaris, may not be able to resolve the distinctive capillaries at this layer. While the average intercapillary area of the choriocapillaris is 5-20 µm, the lateral resolution of commercial SD-OCTA instruments is only 15-20 µm [139,140]. Hence, measurements of flow voids or flow deficits may better represent the vascular changes occurring at the choriocapillaris due to DM and DR. Studies analyzing perfusion density at the choriocapillaris have revealed differences between DM and DR patients and non-DM patients. Using SS-OCTA, Mastropasqua et al. measured the perfusion density of the choriocapillaris at a central 5.5 mm region [72]. They observed significantly lower perfusion density amongst the PDR group compared to non-DM controls and to DM patients without DR [72]. Conti et al. found decreased choriocapillaris flow at the area overlying the parafovea amongst PDR and NPDR patients compared to non-DM controls, but they did not find any changes in the flow at the choriocapillaris amongst patients with DM without DR [91]. On the contrary, Li et al. did find significantly lower flow at the fovea of the choriocapillaris in DM subjects without DR compared to non-DM controls [73]. However, given the aforementioned weaknesses of measuring flow or perfusion density, the findings from these studies are likely not as rigorous as those which analyzed flow deficits.
Increased flow deficits in the presence of DM or DR have been reported by multiple SS-OCTA studies ( Figure 5) [92,140,141]. In two separate studies, Dai et al. calculated choriocapillaris flow from OCTA en-face images using a global thresholding algorithm that segmented flow deficits larger than 24 µm and consequently above spacing between individual capillaries [92,140,142]. Compared to non-DM patients, DM patients had a larger percentage of the image centered on the macula that corresponded to flow deficits [92]. In an extension of their earlier study, Dai et al. investigated flow deficits in patients with NPDR and PDR, finding that both groups had a significantly higher percentage area occupied by flow deficits compared to controls [140]. Gendelman et al. corroborated these results, reporting a positive association between flow deficit percentage and DR severity [141]. These studies are in agreement with histologic analysis, which has revealed choriocapillaris dropout and ischemia in DM eyes with and without DR [143][144][145]. A further OCTA study found an association between visual functioning in DR eyes and impaired choriocapillaris flow, as measured by flow deficits [146]. Given that the blood supply of photoreceptors is highly reliant upon the choriocapillaris circulation, visual changes in patients with DR may be linked with ischemia to these regions with high metabolic demands. SS-OCTA is a valuable tool to assess the retinal and choroidal vasculature at the same time. results, reporting a positive association between flow deficit percentage and DR severity [141]. These studies are in agreement with histologic analysis, which has revealed choriocapillaris dropout and ischemia in DM eyes with and without DR [143][144][145]. A further OCTA study found an association between visual functioning in DR eyes and impaired choriocapillaris flow, as measured by flow deficits [146]. Given that the blood supply of photoreceptors is highly reliant upon the choriocapillaris circulation, visual changes in patients with DR may be linked with ischemia to these regions with high metabolic demands. SS-OCTA is a valuable tool to assess the retinal and choroidal vasculature at the same time. Further quantitative studies have the opportunity to shed light on the impact of DR on the choriocapillaris.

Limitations and Shortcomings of OCT Angiography in Quantitative Analysis
The majority of quantitative OCTA data in the literature is cross sectional, limiting its predictive value for the progression of DR. However, a study by Sun et al. over a median follow-up time of 27 months identified that larger FAZ area, lower VD, and lower FD were all significantly associated with DR progression [147]. An additional study with a one-year follow-up by You et al. used OCTA to detect avascular areas within the parafovea, noting that eyes with progression in DR severity had significantly larger avascular areas in the SCP at baseline compared to eyes without DR progression [148]. Also over a one-year follow-up period, Greig et al. reported that a larger FAZ area and reduced VD in the temporal region at baseline can predict the risk of progression [149] Hence, a variety of quantitative OCTA metrics do hold immense promise as DR biomarkers. Longitudinal studies with longer follow-up periods and increased patient enrollment will provide further validation of these OCTA biomarkers in tracking DR.
Another drawback of comparing OCTA metrics from multiple studies is the lack of interchangeability across quantitation algorithms [150], OCTA devices [151], and scan sizes [152]. This precludes the formation of singular normative databases. Instead, a plethora of normative values is likely needed given the wide array of custom analysis techniques reported in the literature.
Numerous studies have made evident that OCTA measurements of the microvasculature may differ based on age [153][154][155], sex [156,157], HbA1c levels [102,158,159], and a variety of other systemic or metabolic vascular risk factors [160][161][162][163]. Additionally, previous work by Chun et al. illuminated differences in OCTA quantitative parameters between healthy, young Black and White patients [164]. Compared to White patients, Black patients were found to have lower vessel density at the parafovea of the DCP, a larger foveal avascular zone at the SCP and DCP, and lower blood flow at the choriocapillaris in the area underlying the macula [164]. These findings support the need to match study and

Limitations and Shortcomings of OCT Angiography in Quantitative Analysis
The majority of quantitative OCTA data in the literature is cross sectional, limiting its predictive value for the progression of DR. However, a study by Sun et al. over a median follow-up time of 27 months identified that larger FAZ area, lower VD, and lower FD were all significantly associated with DR progression [147]. An additional study with a one-year follow-up by You et al. used OCTA to detect avascular areas within the parafovea, noting that eyes with progression in DR severity had significantly larger avascular areas in the SCP at baseline compared to eyes without DR progression [148]. Also over a one-year follow-up period, Greig et al. reported that a larger FAZ area and reduced VD in the temporal region at baseline can predict the risk of progression [149] Hence, a variety of quantitative OCTA metrics do hold immense promise as DR biomarkers. Longitudinal studies with longer follow-up periods and increased patient enrollment will provide further validation of these OCTA biomarkers in tracking DR.
Another drawback of comparing OCTA metrics from multiple studies is the lack of interchangeability across quantitation algorithms [150], OCTA devices [151], and scan sizes [152]. This precludes the formation of singular normative databases. Instead, a plethora of normative values is likely needed given the wide array of custom analysis techniques reported in the literature.
Numerous studies have made evident that OCTA measurements of the microvasculature may differ based on age [153][154][155], sex [156,157], HbA1c levels [102,158,159], and a variety of other systemic or metabolic vascular risk factors [160][161][162][163]. Additionally, previous work by Chun et al. illuminated differences in OCTA quantitative parameters between healthy, young Black and White patients [164]. Compared to White patients, Black patients were found to have lower vessel density at the parafovea of the DCP, a larger foveal avascular zone at the SCP and DCP, and lower blood flow at the choriocapillaris in the area underlying the macula [164]. These findings support the need to match study and control groups by race when analyzing OCTA quantitative parameters, while also creating diverse normative databases. Race, along with the aforementioned potential confounding variates, are oftentimes left unaddressed or unreported in OCTA studies of DR. It is important to account for any differences in these categories across the studied population, such that the relationship between DR and OCTA parameters is precisely interrogated.

Future Directions
One of the most obvious extensions of OCTA is in artificial intelligence and automated diagnosis of DR. Expanding the use of artificial intelligence and deep learning techniques in diabetic eye screenings would improve efficiency and drive down costs, which jointly have the potential to improve patient outcomes amid the diabetes epidemic [165]. This may be especially true with regards to rural areas, where eye care providers are found in lower numbers [166]. Using an ordinary least squares modeling method, Cano et al. correctly classified 94% of PDR from non-DM subjects with a 0.99 area under receiver operating characteristic curve (AUROC); the same model classified 91% of mild NPDR subjects from non-DM subjects with a 0.93 AUROC [167]. Alam et al. trained a support vector machine using six OCTA quantitative features and achieved a 93% accuracy in distinguishing non-DM and mild NPDR eyes [168]. A computer-aided diagnosis system from Sandhu et al. reached accuracies of 94% and 92% AUROC using features from the SCP and DCP to distinguish between mild NPDR and non-DM eyes [169]. In these automated diagnostic systems, it is equally important to distinguish between DR and other retinopathies. Alam et al. demonstrated 92% specificity in distinguishing DR from sickle cell retinopathy [170]. Machine learning has proven to be a valuable tool for classifying both early and late stages of DR. Applying these models to larger and more diverse test sets will clarify their potential to be implemented on large scales.
Numerous groups have begun to investigate handheld OCTA systems that offer greater portability and flexibility for imaging patients that would otherwise prove difficult [171][172][173][174]. These handheld probes offer greater ease of imaging to pediatric patient populations or to those who must remain supine. It is easy to see the reach these systems bring, as they are not faced with the same constraints as traditional OCTA devices, namely being mounted to a tabletop. In conjunction with automated diagnosis programs, handheld OCTA devices could very well be used to efficiently and accurately screen for DR in areas underserved by eye care providers. Rank et al. recently reported their progress towards OCT and OCTA on a photonic integrated chip, which would drastically reduce both the size and costs of these imaging devices [175]. Collectively, these innovations across a variety of different sectors will continue to drive the feasibility of OCTA as a screening tool for DR.

Conclusions
OCTA has come a long way within a short period, enhancing our understanding of the pathophysiology and progression of DR. Given the myriad of microvasculature changes that can be detected and quantified using OCTA, there is every reason to believe that OCTA can be effectively implemented as a valuable imaging tool for DR. Challenges ahead include establishing normative databases, given the complexity and breadth of image analysis techniques in existence, as well as longitudinal analysis of quantitative trends in patients with DR.

Conflicts of Interest:
The authors declare no conflict of interest.