Comparison of Diabetic Retinopathy Lesions Identified Using Ultrawide Field Imaging and Optical Coherence Tomography Angiography

Abstract

Introduction

Optical coherence tomography (OCT) angiography (OCTA) has the potential to influence the diagnosis and management of diabetic eye disease. This study aims to determine the correlation between diabetic retinopathy (DR) findings on ultrawide field (UWF) color photography (UWF-CP), UWF fluorescein angiography (UWF-FA), and OCTA.

Methods

This is a cross-sectional, prospective study. One hundred and fourteen eyes from 57 patients with diabetes underwent mydriatic UWF-CP, UWF-FA, and OCTA. DR severity was assessed. Ischemic areas were identified on UWF-FA using ImageJ and the nonperfusion index (NPI) was calculated. Diabetic macular edema (DME) was assessed using OCT. Superficial capillary plexus vessel density (VD), vessel perfusion (VP), and foveal avascular zone (FAZ) area were automatically measured on OCTA. Pearson correlation coefficient between the imaging modalities was determined.

Results

Forty-five eyes were excluded due to non-DR findings or prior laser photocoagulation; 69 eyes were analyzed. DR severity was associated with larger NPI (r = 0.55944, p < 0.0001) even after distinguishing between cones (Cone Nonperfusion Index [CPI]: r = 0.55617, p < 0.0001) and rods (Rod Nonperfusion Index [RPI]: r = 0.55285, p < 0.0001). In eyes with nonproliferative DR (NPDR), NPI is correlated with DME (r = 0.51156, p = 0.0017) and central subfield thickness (CST) (r = 0.67496, p < 0.0001). UWF-FA macular nonperfusion correlated with NPI (r = 0.42899, p = 0.0101), CPI (r = 0.50028, p = 0.0022), and RPI (r = 0.49027, p = 0.0028). Central VD and VP correlated with the DME presence (r = 0.52456, p < 0.0001; r = 0.51952, p < 0.0001) and CST (r = 0.50133, p < 0.0001; r = 0.48731, p < 0.0001). Central VD and VP were correlated with macular nonperfusion (r = 0.44503, p = 0.0065; r = 0.44239, p = 0.0069) in eyes with NPDR. Larger FAZ was correlated with decreased central VD (r = −0.60089, p = 0.0001) and decreased central VP (r = −0.59224, p = 0.0001).

Conclusion

UWF-CP, UWF-FA, and OCTA findings provide relevant clinical information on diabetic eyes. Nonperfusion on UWF-FA is correlated with DR severity and DME. OCTA metrics of the superficial capillary plexus correlate with the incidence of DME and macular ischemia.

Introduction

An estimated 783 million people globally will have diabetes mellitus (DM) by the year 2045 [1]. With this increase in the prevalence of DM, its associated ocular complications are also on the rise. Diabetic retinopathy (DR) and diabetic macular edema (DME) are leading causes of visual loss in the working age population worldwide [2, 3]. Thus, routine retinal examinations among patients at risk prior to the onset of advanced disease is a vital part of diabetes care in order to deliver timely management and prevent visual disability [4, 5].

DR and DME are traditionally evaluated using findings on fundus photography and fluorescein angiography (FA). More specifically, DR is assessed using the Early Treatment Diabetic Retinopathy Study (ETDRS) standard 7-field photos while DME is evaluated using stereoscopic photos of the central 30 degrees of the fundus (ETDRS standard field 2) [4, 6, 7]. In the past decade, however, optical coherence tomography (OCT) has become the new reference standard in the evaluation of DME due to its widespread availability and precise assessment of the retinal layer structure [8]. Likewise, ultrawide field (UWF) imaging has been increasingly utilized in the evaluation of patients with DR [4, 9–12]. UWF imaging has the ability to obtain reliable high-resolution images that can detect microvascular abnormalities on the fundus even on the far periphery [10, 13–15]. Images obtained using UWF imaging were found to have moderate to substantial agreement with images taken using the ETDRS standard 7-fields in determining the level of DR [10]. Together, UWF imaging and OCT can identify most diabetic retinal lesions including hemorrhages and microaneurysms, capillary closures, exudates, and macular thickening, but the finer details of the macula are still lacking. The identification of nonperfusion on FA is only an alternative measure of retinal tissue hypoxia and evaluation of the differences in macular and peripheral retinal perfusion is also overlooked [15, 16]. Techniques that consistently assess retinal perfusion can possibly provide a more accurate insight on the degree and scope of retinal ischemia in DR [15].

OCT angiography (OCTA) is a novel ocular imaging modality that uses a scanning laser diode to produce depth-resolved 3-dimensional images of the retinal capillaries. It detects the motion of red blood cells within sequential cross-sectional OCT scans performed repeatedly at the same location [17]. Unlike FA that necessitates intravenous injection of dye and may cause anaphylaxis, OCTA is noninvasive [11, 18]. It can evaluate pathologic macular changes more rapidly and safely. OCTA produces a 2-dimensional representation of the macular microvasculature in the specific layers, namely, superficial capillary plexus (SCP), deep capillary plexus (DCP), and the avascular retina. Furthermore, quantitative parameters such as vessel density (VD) and vessel perfusion (VP) are measured automatically [11, 19]. OCTA has the potential to provide clear details of the microvasculature of the retina in eyes with DR and DME without some of the drawbacks of traditional FA [18]. Some authors have suggested that OCTA is superior to FA in evaluating nonperfusion [20].

The application of OCTA as an adjunct imaging modality to color photos, fluorescein angiograms, and structural OCT scans in the retinal evaluation of patients with diabetes have potential to influence the way we diagnose and manage DR and DME. This study aimed to compare diabetic retinal lesions, especially on the macula, identified on UWF color photography (UWF-CP), UWF-FA, and OCTA.

Methods

Population and Sample

Adult patients with DM were recruited to participate in the study. Inclusion criteria were (1) known or diagnosed case of type 1 or type 2 diabetes; (2) 18 years of age or older at the time of informed consent; and (3) sufficiently clear ocular media and adequate pupil dilation to permit good quality photographic imaging. Exclusion criteria were (1) history or current evidence of hypersensitivity or clinically relevant sensitivity to fluorescein; (2) active periocular, ocular, or intraocular infection at the time of examination; (3) history of recurrent or currently active ocular inflammation; (4) history or clinical evidence of age-related macular degeneration or any retinal vascular disease other than DR; (5) any current evidence of dense cataracts or vitreous hemorrhage that precludes adequate view of the fundus, or retinal detachment at the time of examination; (6) previous macular laser treatment; and (7) any concurrent ocular condition that precludes satisfactory completion of study imaging examinations.

One hundred and fourteen eyes from 57 patients were enrolled in the study. The study design was compliant with the ethical standards stated in the 1964 Declaration of Helsinki and other relevant regulations. The study protocol was approved by the Institutional Review Board (IRB) of The Medical City, Metro Manila, Philippines. Informed consent was obtained from all study participants.

Image Acquisition and Analysis

This was a single-site, cross-sectional, prospective study comparing diabetic retinal findings identified on UWF-CP, UWF-FA, and OCTA. Images acquired during the same visit were prospectively collected from adult patients with diabetes at The Medical City Eye and Vision Institute (TMC-EVI), Ortigas Avenue, Pasig City, Metro Manila, Philippines.

All patients underwent imaging with UWF-CP and UWF-FA using Optos California (Optos plc, Dunfermline, Scotland, UK) and OCTA with structural OCT using the Zeiss Cirrus 5000 OCT AngioPlex (Carl Zeiss Meditec Inc., Dublin, CA, USA). The images were acquired through dilated pupils by a trained ophthalmic photographer. Mydriasis was attained using 1 drop of tropicamide + phenylephrine hydrochloride (0.5%/0.5%) eye drops. All collected images were anonymized and stored in a secure location.

All images were evaluated at a reading center under controlled standardized conditions. The reading center’s high resolution, high-definition LCD monitors were used which were regularly calibrated to a color temperature of 6,500 K and gamma setting of 2.2 (Spyder4PRO; Datacolor, Lawrenceville, NJ, USA). All UWF images were projected stereographically using proprietary software available from the manufacturer (Optos plc, Dunfermline, Scotland, UK). This stereographic projection was based on ray tracing of each pixel through a combined optical model of the Optos 200Tx and a Navarro UWF model eye. With this, both the surface area (mm²) and the density of cone and rod photoreceptors (cells/mm²) can be defined for every pixel [21]. Furthermore, by selecting a region of the fundus, the approximate number of rods and cones can be calculated in addition to an accurate retinal area measurement [22].

A masked trained and certified image grader experienced in evaluating UWF images (R.P.S.) assessed the UWF-CP, UWF-FA, and OCTA images for the presence of hemorrhages and/or microaneurysms, capillary nonperfusion, and the presence or absence of macular edema. Independently, a senior retina specialist trained and certified in diabetic retinal image grading (P.S.S.) provided secondary grading. In case of disagreement, a final grade was generated using consensus agreement.

Using UWF-CP, the level of DR was graded using previously validated protocols taking into consideration both the number and extent of distribution of retinal lesions [10, 23]. Ischemic areas were identified on UWF-FA using ImageJ (Fiji) software (National Institutes of Health, Bethesda, MD, USA) [24] using a method described in detail in previous studies [15, 23, 25, 26]. In brief, the total gradable retina of each UWF-FA image was demarcated by drawing a free-hand line using ImageJ software and was saved as an image mask. The areas of retinal nonperfusion on the macula and on the periphery were identified and demarcated using the same method (Fig. 1). The two image masks then underwent segmentation using exact pixel coordinates in each mask. The total gradable area (TA) and nonperfused area (NPA) were calculated by adding the size of all pixels that make up the mask while also correcting for peripheral image distortion. The sums were then converted to square millimeters. The nonperfusion index (NPI) in square millimeters was calculated as NPA divided by the TA. Based on human retinal cone and rod distribution, the amount and extent of involvement of rods and cones were similarly calculated.

Fig. 1.

Demarcation of total gradable area (TA) and nonperfusion area (NPA) using ImageJ (Fiji) software. a Using free-hand lines, the TA was encircled green, while areas of nonperfusion were demarcated with yellow. b A magnified view of nonperfused areas in the far retinal periphery.

The presence of center-involving DME (ciDME) was based on the presence of retinal thickening above the 95th percentile in the central subfield caused by diffuse thickening, foveal cystic spaces, or subretinal fluid on macular OCT. Central subfield thickness (CST) was assessed based on the measurement (in µm) and normative distribution of OCT software. OCTA findings were assessed using AngioPlex software and Forum viewer (Carl Zeiss Meditec Inc.). OCTA scans were assessed for image quality and acceptability of automated segmentation. Only scans with signal strength of ≥6 (scale of 0–10) were selected based on the manufacturer’s recommendation. Four eyes with unsatisfactory segmentation underwent manual adjustment of the scans using tools available in the software. The OCTA software scans utilizing a 6 mm × 6 mm area gave automated measurements of the VD and VP of the SCP. VD and VP were subdivided into central (fovea), inner (parafovea), outer (perifovea), and full (whole macular area). OCTA software also provided automated measurements of the foveal avascular zone (FAZ) area and circularity (Fig. 2).

Fig. 2.

a OCTA 6 mm × 6 mm scan showing the VD of the different regions of the macula and the FAZ (demarcated with yellow). b Accompanying en face structural OCT.

Statistical Analysis

Correlation between the three imaging modalities was determined by calculation of Pearson correlation coefficient (r). Coefficient interval of 0–0.199 was assessed as very weak correlation, 0.20–0.399 as weak correlation, 0.40–0.599 as moderate correlation, 0.60–0.799 as strong correlation, and 0.80–1.00 as very strong correlation. The distributions of continuous variables were compared using Wilcoxon rank-sum test. A p value <0.05 was considered significant. Statistical interpretation was performed using SAS software version 9.4 (SAS Inc., Cary, NC, USA).

Results

Over a 7-month study period, 114 eyes from 57 patients were imaged and evaluated. Forty-five eyes were excluded due to non-DR findings or prior laser panretinal photocoagulation, leaving 69 eyes (41 right eyes and 28 left eyes) from 43 patients for detailed analysis. Of the eyes that were excluded in the analysis, 17 were excluded due to incidental findings such as 9 eyes with non-diabetic retinal lesions (5 central retinal vein occlusion; 2 epiretinal membranes; and 2 age-related macular degeneration); 4 eyes with vitreous hemorrhage precluding adequate view of the macula; and 4 eyes with poor quality scans on OCTA. Of the 97 eyes with diabetic eye disease, 28 eyes had evidence of prior panretinal photocoagulation laser and were also excluded from analysis as accurate nonperfusion mapping could not be performed in these eyes. Of the 69 eyes included in the detailed analysis, 12 eyes had no DR, 36 eyes had nonproliferative DR (NPDR [mild to moderate – 12; severe to very severe – 24]), and 21 eyes had proliferative DR (PDR [non-high-risk – 11; high-risk – 10]). Baseline characteristics of patients’ eyes are summarized in Table 1.

Table 1.

Baseline characteristics of eyes included in the study

N = 97 eyes	Mean±SD or N (%)
Age, years	60.1±9.5
Diabetes duration, years	10.3±7.7
Diabetes type 2	94 (96.9)
Diabetic retinopathy severity
No DR	12 (12.4)
Mild to moderate NPDR	12 (12.4)
Severe to very severe NPDR	24 (24.7)
PDR	11 (11.3)
PDR HRC	10 (10.3)
Post-PRP	28 (28.9)

DR, diabetic retinopathy; NPDR, nonproliferative DR; PDR, proliferative DR; HRC, high-risk characteristics; PRP, panretinal photocoagulation; SD, standard deviation.

DR Severity

Increasing DR severity on UWF color photos was associated with a greater NPI on UWF-FA (r = 0.55944, p < 0.0001). This remains significant even after distinguishing between cones (Cone Nonperfusion Index [CPI]: r = 0.55617, p < 0.0001) and rods (Rod Nonperfusion Index [RPI]: r = 0.55285, p < 0.0001). Figure 3 presents the mean NPI, CPI, and RPI of patients among the different severity levels of DR.

Fig. 3.

The correlation of diabetic retinopathy (DR) severity with nonperfusion index (NPI), cones nonperfusion index (CPI), and rods nonperfusion index (RPI).

Increasing DR severity was moderately correlated with the presence of ciDME (r = 0.52306, p = 0.0013) and weakly correlated with increased CST on OCT (r = 0.37859, p = 0.0013). The correlation was stronger when only including eyes with NPDR in the analysis for both ciDME (r = 0.5342, p = 0.0008) and increased CST (r = 0.50656, p = 0.0016).

Nonperfusion and Macular Edema

The presence of macular nonperfusion on UWF-FA was only weakly correlated with the presence of ciDME (r = 0.27789, p = 0.0208) and increased CST (r = 0.26575, p = 0.0273) on OCT. The correlation was stronger when including the 36 eyes with NPDR only: presence of ciDME (r = 0.41154, p = 0.0126) and increased CST (r = 0.5762, p = 0.0002). Furthermore, macular nonperfusion on UWF-FA was moderately correlated with a higher NPI (r = 0.42899, p = 0.0101), CPI (r = 0.50028, p = 0.0022), and RPI (r = 0.49027, p = 0.0028) in eyes with NPDR.

NPI was only weakly correlated with ciDME on OCT (r = 0.39718, p 0.0014) and with increased CST (r = 0.24189, p = 0.0582); however, the correlation is stronger in eyes with NPDR (ciDME [r = 0.51156, p = 0.0017]; increased CST [r = 0.67496, p < 0.0001]). CPI was moderately correlated with ciDME (r = 0.46521, p = 0.0001) and weakly correlated with increased CST (r = 0.31795, p = 0.0118) which became stronger when only including eyes with NPDR in the analysis (ciDME [r = 0.56269, p = 0.0004]; CST [r = 0.75833, p < 0.0001]). Likewise, RPI was moderately correlated with ciDME (r = 0.45586, p = 0.0002) and weakly correlated for CST (r = 0.31798, p = 0.0118). When considering eyes with NPDR only in the analysis, the correlation became stronger for both ciDME (r = 0.55739, p = 0.0005) and CST (r = 0.74847, p < 0.0001).

VD, VP, and Nonperfusion

Central VD on OCTA was weakly correlated with the presence of macular nonperfusion (r = 0.28167, p = 0.019) on UWF-FA. This correlation was stronger (r = 0.44503, p = 0.0065) when considering the eyes with NPDR only in the analysis. Similarly, central VP was weakly correlated with the presence of macular nonperfusion (r = 0.29033, p = 0.0155) which was stronger (r = 0.44239, p = 0.0069) when including only NPDR in the analysis.

Central VD of the SCP did not correlate with NPI (r = 0.00571, p = 0.9649), CPI (r = 0.04371, p = 0.7359), and RPI (r = 0.03211, p = 0.8043). Full macular VD also did not correlate with NPI (r = −0.11455, p = 0.3753), CPI (r = −0.11834, p = 0.3596), and RPI (r = −0.12479, p = 0.3338). Likewise, central VP did not correlate with NPI (r = 0.00588, p = 0.9638), CPI (r = 0.03927, p = 0.7619), and RPI (r = 0.02817, p = 0.8279). Full macular VP did not correlate with NPI (r = −0.0821, p = 0.5259), CPI (r = −0.08149, p = 0.5289), and RPI (r = −0.09051, p = 0.4842). The figures remained not significant when considering only NPDR eyes. Inner and outer VD as well as inner and outer VP was not found to be correlated to NPI, CPI, and RPI.

VD, VP, and Macular Edema

Central VD on OCTA was correlated with the presence of ciDME (r = 0.52456, p < 0.0001) and increased CST (r = 0.50133, p < 0.0001) on OCT. Central VP was also correlated with the presence of ciDME (r = 0.51952, p < 0.0001) and increased CST (r = 0.48731, p < 0.0001).

Foveal Avascular Zone

A larger FAZ area was correlated with decreased central VD (r = −0.60089, p = 0.0001) and with decreased central VP (r = −0.59224, p = 0.0001) when considering only eyes with NPDR in the analysis. A larger FAZ area was correlated with a decrease in FAZ circularity (r = −0.57963, p = 0.0002) in eyes with NPDR.

Visual Acuity

There was a statistically significant correlation between visual acuity and the presence of macular nonperfusion on UWF-FA. Those with worse best-corrected visual acuity (BCVA) had worse nonperfusion (r = 0.48842, p < 0.0001). BCVA is correlated with NPI (r = 0.35801, p = 0.0043), moderately correlated with CPI (r = 0.41942, p = 0.0007), and moderately correlated with RPI (r = 0.41693, p = 0.0007). These correlations were stronger for macular nonperfusion (r = 0.61424, p < 0.0001) when including eyes with NPDR only in the analysis. This trend was also observed with NPI (r = 0.58985, p = 0.0002), CPI (r = 0.66359, p < 0.0001), and RPI (r = 0.65323, p < 0.0001).

There was statistically significant correlation between worse BCVA and the presence of ciDME (r = 0.48738, p < 0.0001) and increased CST (r = 0.52175, p < 0.0001) on OCT. This correlation is stronger when only including eyes with NPDR in the analysis for both ciDME (r = 0.54955, p = 0.0005) and CST (r = 0.68188, p < 0.0001). Table 2 presents the comparison of mean values of UWF-FA and OCTA parameters between eyes with ciDME and without ciDME on OCT.

Table 2.

Comparison of mean scores of variables between eyes with ciDME versus eyes with non-ciDME on OCT

	ciDME present on OCT (N = 17)		ciDME absent on OCT (N = 52)		p value
	Mean	SD	Mean	SD
BCVA, logMAR	0.5764706	0.4070157	0.2038462	0.2449182	0.0002*
CST on OCT, μm	460.2941176	180.3592820	243.8461538	39.4622603	<0.0001*
VD central (fovea)	11.5529412	4.1894388	6.5192308	3.3557115	0.0001*
VD inner (parafovea)	16.6235294	1.2671471	15.5730769	2.7061379	0.2660
VD outer (perifovea)	15.9647059	1.7388722	15.6711538	2.2887908	0.6875
VD full (entire macula)	15.9941176	1.5421051	15.3846154	2.2174340	0.3946
VP central (fovea)	0.2637647	0.0993010	0.1467308	0.0788297	0.0002*
VP inner (parafovea)	0.4105882	0.0344530	0.3746538	0.0693742	0.0644
VP outer (perifovea)	0.4115294	0.0438565	0.3891538	0.0615535	0.1880
VP full (entire macula)	0.4021765	0.0443019	0.3791154	0.0589419	0.1540
FAZ area, mm2	0.4958824	0.7397302	0.3323077	0.1343191	0.7338
FAZ circularity	0.5752941	0.1273341	0.6446154	0.1192536	0.0343*
NPI on UWF-FA	0.1195798	0.1028341	0.0475145	0.0626897	0.0029*
CPI on UWF-FA	0.1317215	0.1143574	0.0427976	0.0572092	0.0011*
RPI on UWF-FA	0.1403803	0.1227645	0.0464768	0.0628157	0.0010*

ciDME, center involving diabetic macular edema; OCT, optical coherence tomography; SD, standard deviation; CST, central subfield thickness; BCVA, best-corrected visual acuity; VD, vessel density; VP, vessel perfusion; FAZ, foveal avascular zone; NPI, nonperfusion index; CPI, cones nonperfusion index; RPI, rods nonperfusion index; UWF-FA, ultrawide field fluorescein angiography.

*Statistically significant.

Distribution of Rod and Cone Involvement in Areas of Nonperfusion

The reported ratio of rods to cones in the normal human retina is 20 to 1 or greater [27]. In this cohort, the ratio of rods and cones in the overall retinal area imaged was 21.52 to 1 (rods:cone), and in the areas of nonperfusion, this was increased to 22.93 to 1 (p < 0.0001). Furthermore, there is a trend of increasing rod involvement with increasing DR severity (no DR: 22.50:1, mild to moderate NPDR: 22.98:1, severe and very severe NPDR 23.01:1, PDR 23.37:1, p = 0.0016). There were no statistical differences in the rod to cone ratio in eyes with no ciDME and with ciDME on OCT (no ciDME: 23:02:1 vs. ciDME: 22.89:1, p = 0.76). No other significant correlations were identified between the rod to cone ratio and BCVA or the various OCTA parameters.

Discussion

This prospectively collected, cross-sectional study found good and biologically plausible correlation of diabetic retinal findings between UWF-CP, UWF-FA, and OCTA. Briefly, DR severity on UWF-CP is correlated with increased NPI on UWF-FA and with the presence of ciDME. In addition, macular nonperfusion on UWF-FA and the SCP’s OCTA metrics correlated with the presence of DME on OCT.

It has been documented that the development of retinal lesions among those with diabetes is linked to the presence of nonperfusion, suggesting that DR is a disease of the eye’s microvasculature [15, 16, 28]. In our study, the more severe the DR level as assessed by using a standardized protocol on UWF color photos and UWF-FA, the higher the NPI. Other studies suggesting that retinal ischemia is strongly associated with DR severity have been published previously [15, 29]. Ehlers and colleagues reported that quantitative angiographic metrics on UWF-FA, including panretinal ischemic index, were the strongest correlated features with increasing severity of DR [29]. In our study, we identified increasing rod involvement with increasing DR severity, suggesting a potential marker of increased risk for worsening that will need to be prospectively investigated with a larger and more diverse cohort of eyes.

DME is present in as high as 30% of adults with longstanding disease and can occur at any stage of DR [30]. Our study has shown that DME is correlated with increasing DR severity level. This agrees with results of a previous study where eyes with mild NPDR have a DME prevalence of only 3%. This increases to 38% in moderate to severe NPDR and to 71% in PDR [31]. In our study, increased NPI was associated with the presence of ciDME. A study by Patel and colleagues [30] found that eyes with larger areas of ischemia and more severe DR have higher incidence of refractory DME. These observations are supported by the correlation between retinal nonperfusion and macular edema, suggesting an ischemic process in the development of DME [32]. Wessel and colleagues [33] reported that eyes with retinal nonperfusion were 3.75 times more likely to develop DME. Areas of retinal ischemia stimulate the release of vascular endothelial growth factor (VEGF) in the eye that is involved in the pathogenesis of macular edema. VEGF is a potent vasodilator that weakens the endothelium of macular capillaries leading to leakage of lipid and diffuse edema [34]. The release of ocular VEGF is induced by the ischemic retina; hence, eyes with larger areas of nonperfusion can be expected to produce more VEGF response and will likely develop macular edema. As the identification of the ischemic areas of the retina and the calculation of NPI become more readily available or automated in the future, the detection of retinal nonperfusion could play a crucial role in the surveillance and monitoring of DME among individuals with diabetes. Future efforts should focus on incorporating the automated measurement of NPI into our routine monitoring of patients with DME. However, the current study did not determine the threshold of retinal ischemia at which point macular edema is likely to occur.

Our results suggest that central (foveal) VD and VP are correlated with the presence of ciDME. A study by Jorgensen and colleagues [16] reported that DME was associated with increased oxygen saturation in retinal venules and decreased oxygen extraction. Using a retinal oximeter to determine retinal vascular perfusion, they hypothesized that DR resulting in macular edema is accompanied by an increase in retinal venule oxygen saturation and a decrease in oxygen extraction. This means that because of the retinal tissues’ incapacity to properly extract oxygen from retinal arterioles, the oxygen saturation in the retinal venules will exhibit the high oxygen content that remains after passing through the capillaries. They further proposed that diabetic maculopathy may not be a separate complication of DR but rather represents a mere progression of retinal lesions that spread to involve the foveal area. In another study, it was proposed that the impaired autoregulation of arteriolar diameter in DR precedes the progression of retinopathy to macular edema [35]. These changes in the microvascular circulation of the retina consequently causes reduced oxygen extraction in the diabetic eye such that while the retinal tissues in general are hypoxic, VD and VP may still remain at normal levels in patients with DME. In contrast, another study by Sun and colleagues [36] reported that eyes with lower VD of the SCP are more likely to develop DME. The study suggested that the breakdown of the inner blood-retinal barrier causes leakage from the SCP and consequent imbalance in fluid entry versus fluid removal. The exact mechanism on how lower VD is correlated with DME has not been fully elucidated and the authors recommended future studies with larger sample sizes to fully uncover the complex pathophysiology underlying this disease process [36]. Another study suggested that the accumulation of fluid intraretinally in eyes with DME masks the flow signal emitted by perfused vessels underneath and may corrupt the measurement of OCTA metrics [28]. It is possible that the macula’s unique inherent anatomic characteristics also play a role in the perfusion status of an edematous macula. Specifically, establishing if cell-specific or layer-specific changes in the macula affect VD and VP in macular edema is subject to further investigation. It is important to note that the measurement of central VD among eyes with DME may have limitations due to segmentation errors caused by changes in the layer architecture and artifacts due to the presence of intraretinal cystic changes [17, 37]. Reducing these artifacts through further improvements in automated segmentation in future software versions would be useful in ensuring a more accurate OCTA evaluation of DME. In our study, four eyes with poor automated segmentation underwent manual review and adjustment of the scans using software tools from the device supplier in order to minimize inaccuracies in OCTA metrics.

Interestingly, our study failed to demonstrate a correlation between NPI on UWF-FA and macular VD and macular VP on OCTA. This did not change when distinguishing between CPI and RPI. This is consistent with ETDRS data that reported the predictive power of retinal nonperfusion was strengthened when excluding the area within 1,500 μm of the center of the macula, suggesting that nonperfusion peripheral to the retinal center plays a larger role in DR progression rather than with macular edema or macular ischemia [38]. In simple terms, the degree of peripheral ischemia seen on UWF-FA was not associated with macular ischemia on OCTA. Differences in observations on UWF-FA and OCTA seem conflicting but may just represent the underlying differences in the technology used by the two modalities in obtaining images [18]. Because we were limited to only obtaining automated measurements of the SCP in this study, it is possible that differences in SCP and DCP characteristics also played a role in this finding. Further studies are necessary to investigate the anatomic and functional differences between the SCP and DCP of the macula and their influence on the expression of diabetic macular disease.

In eyes with NPDR, the presence of macular nonperfusion on UWF-FA is correlated with central VD and central VP. This finding is unsurprising as eyes with larger FAZ on FA can be expected to have lower VD on OCTA. A previous study evaluating macular perfusion in retinal vein occlusions reported that macular ischemia on FA corresponds well with ischemia of the SCP on OCTA [39]. This may also hold true for DR owing to similarities in disease pathophysiology but further investigation into this topic is warranted.

Expectedly, poor BCVA was found to be associated with macular nonperfusion and increased NPI on UWF-FA, with the presence of macular edema on OCT, and with enlarged FAZ on OCTA. This is well-documented in numerous studies in the past [8, 11, 40–42]. Nonperfusion and capillary dropout at the macula causes decreased oxygen supply to the photoreceptors, while edema causes disruption of the normal architecture of the macula which both consequently lead to visual impairment.

Diabetic macular ischemia (DMI) is defined as enlargement of the FAZ or disruption of the foveal capillary net on FA [36]. However, FA is an invasive test, requires dye injection and can be time-consuming. Alternatively, the fovea can be quantified on OCTA without being complicated by dye leakage [11]. Our study reports that FAZ area measured via OCTA is inversely correlated with FAZ circularity in eyes with NPDR. The enlargement of the FAZ is likely due to capillary dropouts along the borders of the FAZ. These observations are supported by previous studies reporting that the FAZ is pathologically enlarged with loss of symmetry and circularity in eyes with DR [11, 43, 44]. OCTA can reliably determine FAZ irregularities and areas of foveal capillary dropout equally in both eyes [18] and because OCTA is noninvasive, DMI can now be assessed more readily. While it takes longer than structural OCT alone, the acquisition time of OCTA can be as short as a few seconds and is much faster than FA which takes around 10–15 min after dye injection (excluding pre-injection preparation procedures) [17, 45–48]. Advances in scanning speeds, automated fovea identification, blink detection correction, and eye tracking technology of currently available OCTA devices also means that acquisition of scans is possible even in eyes with poor fixation and reduced visual acuity [17, 49].

The study results were limited to the machines used to evaluate the participants’ eyes. In particular, the OCTA machine and software used in this study can segment the retina into superficial, deep, and avascular layers but can only provide automated measurements of the VD, VP, and FAZ for the SCP. Hence, this study was not able to determine the correlation of the DCP OCTA metrics with the DR findings on UWF-CP and UWF-FA. Furthermore, the current OCTA technology is limited in its small field of view. The 6 mm × 6 mm area images only the macula, thus OCTA parameters in the retinal periphery were not detected and analyzed. The use of the Optos camera to image the retina also has its inherent limitations. One of which is the amount of peripheral distortion that can overestimate lesions found at the temporal and nasal retinal periphery that has been documented in prior studies [50, 51]. Another limitation of the study was the substantial reduction of the sample size when eyes with co-existing conditions and prior history of laser panretinal photocoagulation were excluded in the final analysis. One advantage of the study was the prospective recruitment of study participants among patients referred to the eye instrument center for DR screening. Another advantage was the use of automated measurements of the OCTA parameters. To the best of our knowledge, our study is the first to distinguish nonperfusion between rods and cones on UWF-FA and compare these with DR severity and OCTA metrics.

Conclusion

In summary, our study suggests that UWF-CP, UWF-FA, and OCTA parameters are complementary and offer relevant information on diabetic eyes. The correlations among the findings on UWF-CP, UWF-FA, and OCTA provide clues on the pathogenesis and progression pattern of DR and DME from which insights on clinical decision-making can be drawn. NPI calculation on UWF-FA permits a more detailed assessment of the perfusion status of the diabetic eye, especially on the periphery, and correlates well with DR severity and DME. In the future, NPI analysis may help clinicians classify DR severity or DME more accurately and predict disease progression. Segmenting the fundus into the various fields and calculating the NPI in each field and its correlations is the subject of a future report. Likewise, as OCTA becomes more widely available, it can potentially become another component in the surveillance and management of patients with diabetic eye disease, particularly in eyes with DME or DMI. In the meantime, further investigation is warranted to provide a more definitive answer as to whether OCTA for DR should be incorporated in routine clinical practice.

Acknowledgments

The authors would like to thank Dr. Victor Caparas, Dr. Glenn Alog, Dr. Margarita Javier, Mrs. Myra Mejia, Mr. Noel Cruz, and Mr. Cloyd Pitoc for their assistance in the conduct of the study.

Statement of Ethics

The study protocol was reviewed and approved by the Institutional Review Board (IRB) of The Medical City, Metro Manila, Philippines (protocol approval number GCSOVS2017036). Written informed consent was obtained from all study participants.

Conflict of Interest Statement

Recivall P. Salongcay, Lizzie Anne C. Aquino, and Claude Michael G. Salva: no conflicts of interest to declare. Tunde Peto: personal fees – Novartis, Bayer, Roche, Heidelberg Engineering, Optos, outside of the submitted work; financial support – Optomed, outside of the submitted work. Paolo S. Silva: financial support – Optomed, Hillrom, outside of the submitted work.

Funding Sources

This study was funded in part by the UK Medical Research Council (MRC) and the Philippine Council for Health Research and Development (PCHRD) through the Newton-Agham Grant (Project Reference: MR/R025630/1). The funding agencies have no role in the design or conduct of this research.

Author Contributions

The authors made substantial contributions to the following aspects of the study: Initial conception and research design – Recivall P. Salongcay and Paolo S. Silva. Data acquisition and research execution – Recivall P. Salongcay, Lizzie Anne C. Aquino, and Claude Michael G. Salva. Data analysis and interpretation and Manuscript preparation – Recivall P. Salongcay, Tunde Peto, and Paolo S. Silva. Funding acquisition – Tunde Peto and Paolo S. Silva.

Funding Statement

This study was funded in part by the UK Medical Research Council (MRC) and the Philippine Council for Health Research and Development (PCHRD) through the Newton-Agham Grant (Project Reference: MR/R025630/1). The funding agencies have no role in the design or conduct of this research.

Data Availability Statement

All data generated or analyzed during this study are included in this article. Further inquiries can be directed to the corresponding author.