Longitudinal Changes of Parafoveal Vessel Density in Diabetic Patients without Clinical Retinopathy Using Optical Coherence Tomography Angiography

Abstract Purpose The purpose of this study was to identify the rate of parafoveal vessel density (VD) changes associated with the progression from non-diabetic retinopathy (NDR) to early stages of DR over a year. Methods This longitudinal cohort study enrolled diabetic patients from the Guangzhou community in China. The patients with NDR at baseline were included and underwent comprehensive examinations at baseline and after 1 year. A commercial OCTA device (Triton Plus, Topcon, Tokyo, Japan) was employed to quantify the parafoveal VD in the superficial and deep capillary plexuses. The rates of change in parafoveal VD over time in the incident DR and NDR groups were compared after a year. Results A total of 448 NDR patients were included in the study. Among them, 382 (83.2%) were stable and 66 (14.4%) developed incident DR during the 1-year follow-up. The average parafoveal VD in the superficial capillary plexus (SCP) reduced significantly more quickly in the incident DR group than in the NDR group (-1.95 ± 0.45%/year vs. −0.45 ± 0.19/year, p = 0.002). The VD reduction rate for the deep capillary plexus (DCP) was not significantly different for the groups (p = 0.156). Conclusions The incident DR group experienced a significantly faster reduction in parafoveal VD in the SCP compared with the stable group. Our findings further provide supporting evidence that parafoveal VD in the SCP may be used as an early indicator of the pre-clinical stages of DR.


Introduction
Diabetic retinopathy (DR), the most significant microvascular complication of diabetes mellitus (DM), is the leading cause of visual impairment among working-age adults globally. 1Currently, the diagnosis of DR still depends on clinical symptoms such as microaneurysms, retinal hemorrhages, lipid exudates, and intraretinal microvascular abnormalities on fundus examination. 2However, these morphological alterations might be the advanced stage of vascular lesions in nature.Retinal nonperfusion and ischemia are likely to occur before microvascular structural changes.Previous studies, based on both patients and animal models, have indicated that retinal microvascular abnormalities and autoregulatory deficits may be involved in the early stages of DR. [3][4][5] Histopathological studies have also found that retinal microvascular damage precedes detectable abnormalities such as microaneurysms in DR. 6 Furthermore, the impairment of microvasculature and resulting macular ischemia are recognized as crucial pathological factors contributing to visual impairment in DR. 7 Therefore, it is very important to determine whether the longitudinal change in parafoveal retinal microvasculature is associated with the presence of DR.It may help us to understand DR pathogenesis and predict the risk of DR development.
Optical coherence tomography angiography (OCTA) has provided new insights into retinal microvasculature in several ocular diseases, including DR. 8 Vessel density (VD) is a potential biomarker for macular ischemia in DR and other retinal vascular diseases. 9urrently, the results of published OCTA investigations on the association between parafoveal microvascular alterations and the presence of DR are controversial.1][12] However, Nesper et al. 13 reported that parafoveal VD did not reach significant differences between healthy controls and NDR for retinal superior and deep layers.Additionally, in the longitudinal study, the evidence of changes in parafoveal VD from NDR to early stages of DR is lacking.As current longitudinal studies mostly focus on the association between retinal VD changes and disease progression in the eyes of DR. [14][15][16] Whether the change in retinal VD over time is associated with DR presence remains unclear.
Therefore, we aimed to elucidate the relationship between the rate of VD change and DR presence.This study evaluated parafoveal VD in the NDR cohort from the community at baseline and after 1 year, compared the rates of parafoveal VD changes between the eyes of incident DR and stable group (NDR group) based on OCTA.

Participants
This prospective longitudinal study was performed at Zhongshan Ophthalmic Center, Sun Yat-sen University, China.The research protocol adhered to the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board Ethics Committee of the Zhongshan Ophthalmic Center (2017KYPJ094).Informed consent was obtained from all the participants.
The participants in this study ranged in age from 30 to 80 years and had type 2 diabetes mellitus without a history of ocular therapy.
Patients with cancer, hypertension, kidney disease, hemorrhagic stroke, clinically significant heart disease including serious arrhythmias, and any other systemic illness other than diabetes were excluded from the study.They were also excluded if they had DR or any other ocular diseases (inflammatory ocular diseases, ocular surface diseases, glaucoma, vitreous macular disease), a history of ocular trauma or surgery, a history of pan-retinal photocoagulation or focal laser treatment, opacities of the refractive medium, amblyopia, poor fixation, a best-corrected visual acuity (BCVA) of less than 20/200, spherical equivalent of < À6.0 D, astigmatism of > þ3.0 D, or axial length (AL) > ¼26.0 mm.Participants with diabetic macular edema or epiretinal membranes on an optical coherence tomography (OCT) scan were also excluded because of probable artifacts that could affect the measurement of vascular flow by OCT angiography.

General information and laboratory parameters
Customized questionnaires were used to obtain baseline data on demographics and medical history (e.g.age, sex, duration of diabetes, and medical history).Their medical history, as well as physical and normal laboratory testing, were used to confirm their health condition.Physical examinations were conducted according to standard procedures, including measurements of height, weight, systolic blood pressure, and diastolic blood pressure.Body mass index was calculated as the weight (kg) divided by the square of the body height (m 2 ).Hemoglobin A1c (HbA1c), triglyceride, total cholesterol, low-density lipoprotein cholesterol, and highdensity lipoprotein cholesterol concentrations were determined in a laboratory certified by the Chinese government.All laboratory parameters were obtained in a Chinese government-approved laboratory.

Ocular examination and imaging protocol
Complete ocular examinations were performed for all participants, including measurement of BCVA and intraocular pressure, refractive error (autorefractometry) slit-lamp biomicroscopy, and fundus examinations.DR was clinically diagnosed based on a complete medical history and full ophthalmologic examination, including an external slit lamp fundus exam and ETDRS 35 degree 7-standard field color retinal photographs (Canon CX-1; Canon, Tokyo, Japan).Two experienced examiners (WW and YM) determined the eyes with mild NPDR or the severity of DR using the ETDRS severity scale. 17,18The DR severity at baseline and after 1-year follow-up was assessed separately in a masked fashion.

OCTA imaging acquisition
The OCTA images were acquired using a DRI OCT Triton machine (Topcon Corporation, Tokyo, Japan) with a wavelength of 1050 nm.Topcon DRI OCT operates at 100,000 Ascans per second with an axial resolution of approximately 8 mm and a transverse resolution of 20 lm in the tissue. 19upillary dilation was performed for each patient, and OCTA imaging was conducted using a 6 Â 6 volume scan pattern centered on the fovea.Furthermore, images of the superficial capillary plexus (SCP) and deep capillary plexus (DCP) were obtained for each eye by applying automatic layer segmentation of the built-in software (IMAGEnet6, v1.23.15008, basic license 10).The manual correction was performed by a trained technician (XG) in cases with segmentation errors, such as incorrect segmentation of the SCP and DCP layers.There was agreement by 2 investigators (WW and XG) that a segmentation error required manual correction and sufficiently corrected.Professor (WH) makes the final decision when answers conflict.
The SCP was segmented with the inner border at 2.6 mm below the internal limiting membrane and the outside boundary at 15.6 mm below the junction between the inner plexiform and the inner nuclear layers.The inner border of DCP was set at 15.6 mm below the inner plexiform and the inner nuclear layers, while the outer border was set at 70.2 mm below them. 20For better quality control, the following are the exclusion criteria of OCTA images 21 : (1) a low signal strength index (SSI50); (2) motion artifacts; (3) improper segmentation of tissue layers or slabs; (4) blurry imaging (fine capillary networks unable to differentiate against background signal); (5) signal loss; (6) inadequate centration; and (7) projection artifacts on DCP.The determination is made by 2 investigators (WW and XG), with the professor's (WH) decision in case of divergent opinions.

Measurement of VD parameters
The ImageJ software was used to analyze all of the OCTA images (National Institutes of Health, Bethesda, MD, USA).First, to minimize background noise, we utilized a nonlocal mean (NLM) denoising filter.Second, an adjustable threshold tool was used.In this study, the tool automatically selected lower and higher thresholds (130-255).Primary vessel-related pixels were acquired using a threshold tool.Third, we applied Niblack as a local thresholding algorithm to binarize the images, which were subsequently skeletonized.After removing the areas occupied by the main blood vessels from the skeletonized image, VD was calculated.The percentage area filled by perfused blood vessels in the scanned region is denoted as the parafoveal VD.For each SCP and DCP image, the total surface area occupied by the blood vessels was measured.The macular area was measured in a 3 mm inner diameter, 6 mm outside diameter annulus centered on the fovea, and the annular region was partitioned into 4 equal sectors (Figure 1).Shoji et al. 22 demonstrated that the Niblack algorithm for evaluating macular VD has high repeatability.For the DCP slab, we used a built-in projection artifact removal tool.

Statistical analyses
All statistical analyses were performed using STATA software (version 16.0; STATA/MP 14.0; STATA Corp., USA).The Kolmogorov-Smirnov test was used for normality tests, and when normality was confirmed, the Student's t-test was used for data analysis.Categorical variables were examined using Fisher's exact test.Mixed-effect models were used to calculate the mean parafoveal VD change per year and the mean parafoveal VD percentage change per year, and the model was adjusted for age and gender.Univariate linear regression was employed to evaluate the correlations between the parafoveal VD and demographic and ocular parameters.Variables with p < 0.1 in the univariate analysis were included in the multivariable models.Statistical significance was set at p < 0.05.

Results
A flowchart of the inclusion procedure is presented in Figure 2. A total of 625 patients with DM were screened, and 511 were eligible for inclusion.Fifty-two patients were lost to follow-up, and 11 patients were excluded because of poor-quality OCTA.A total of 448 eyes were eligible for analysis; 382 (83.2%) had no signs of DR, and 66 (14.4%) had incident DR after 1year.The clinical and demographic characteristics of patients are shown in Table 1.The average age of the patients was 65.38 ± 7.97 years, and 179 were men (39.96%).The average body mass index was 24.46 ± 3.22 kg/m 2 , and the mean HbA1c was 6.93 ± 1.15%.Of all patients with DM, 382 (83.2%) had no signs of DR, and 66 (14.4%) had incident DR after 1 year.The incident DR group was significantly younger than the NDR group (63.08 ± 7.32 years and 65.77 ± 8.02 years, respectively; p ¼ 0.011).Compared with the NDR group, the incident DR group had lower mean systolic blood pressure (127.26 ± 17.08 mmHg vs. 132.48± 17.94 mmHg, respectively; p ¼ 0.028), higher mean HbA1c (7.27 ± 1.29% vs. 6.87 ± 1.11%, respectively; p ¼ 0.009), higher mean total cholesterol (5.13 ± 1.24 mmol/L vs. 4.83 ± 1.07 mmol/L, respectively; p ¼ 0.040), and higher mean low-density lipoprotein cholesterol (3.10 ± 1.13% vs. 2.81 ± 0.90%, respectively; p ¼ 0.021)

Rate of change in parafoveal VD
The baseline average parafoveal VD in different sections of the incident DR and NDR groups is presented in Table 2.At baseline, there was no significant difference between the two groups in each sector of the parafoveal VD (all p > 0.05).
Table 3 presents the rate of parafoveal VD change in the incident DR and NDR groups after adjusting for age and sex over the 1-year follow-up period in SCP and DCP.In the NDR group, parafoveal VD decreased on average 0.45 ± 0.19%/year in SCP and 0.38 ± 0.14%/year in DCP.In the incident DR group, the parafoveal VD decreased by an average of 1.95 ± 0.45%/year in the SCP and 0.89 ± 0.34%/year in the DCP group.Both the incident DR and NDR groups showed a significant decrease in parafoveal VD compared to baseline when measured as a percentage change per year (SCP: p ¼ 0.016 and p < 0.001; DCP:

Risk factors associated with the rate of change in parafoveal VD
Multivariate linear regression analyses were used to adjust for the confounding factors associated with the rate of parafoveal VD change in SCP and DCP respectively (Table 4).The confounding factors were included in the univariate analysis of SCP (Supplementary Table 1) and DCP (Supplementary Table 2).Variables with p values of <0.1 in the univariate analysis were included in the multivariate analysis.In the multivariate model, after adjusting for  In the DCP group, only the correlation between the change in parafoveal VD and baseline DCP parafoveal VD was statistically significant (b ¼ À0.764; 95% CI À0.856 to À0.671; p < 0.001).There was no significant difference in the change in parafoveal VD in the DCP between the incident DR and NDR groups after adjusting for confounding factors (p ¼ 0.104).

Discussion
In this longitudinal study, OCTA was used to measure baseline parafoveal VD in DM patients without DR and compare the rate of parafoveal VD change between the early incident DR and NDR groups over a 1-year follow-up period.We found a significant decrease in parafoveal VD in both the NDR and early incident DR groups compared to baseline.Notably, there was a greater decrease in parafoveal VD in the SCP over time in the early incident DR group compared to the NDR group.When we adjusted for confounding factors, the results remained statistically significant.However, we observed that the rate of change in the parafoveal VD in the DCP did not significantly differ in the NDR and early incident DR groups.To the best of our knowledge, this is the first community-based study involving a large sample to evaluate the longitudinal changes in VD in the NDR cohort.
Few longitudinal studies have investigated the rate of VD changes in patients with NDR, with controversial results.Kim et al. 23 found that parafoveal VD was significantly lost in 1 year compared to baseline in patients with no DR or mild non-proliferative DR (around À1.876%/year).However, this study did not evaluate the rate of VD change between groups with different progressions.Marques et al. 24 showed that parafoveal capillary dropout increased in 3 years in eyes with minimal, mild, and moderate DR using the Cirrus HD-OCT 5000 device.They found that the rates of VD decrease in the SCP were 2.02%/year in the DR progression group and 0.63%/year in the stable group.Aschauer et al. 25 recently published valuable longitudinal observations in a cohort of 95 patients with type 2 diabetes with no/minimal DR, and 13 eyes developed DR within 2 years.The rates of parafoveal VD loss in the SCP were 2.425%/year for the incident DR group and 0.98%/year for the stable group.The VD loss values in these studies were different.This was probably because of the relatively small sample size and insufficient adjustment for confounders.Several factors (HbA1c, body mass index, and total cholesterol) could not be adjusted, which has been suggested to be associated with the risk of DR. 26,27 This study involved a large sample and adequately adjusted for factors that may affect VD change.The result showed annual rates of reduction in VD of 1.95 ± 0.45% and 0.45 ± 0.19% in SCP in the incident DR and stable NDR groups.
The current study provides a new perspective on DR development.Our results showed that the decline in VD was significant in the NDR cohort over time, and was more rapid in the incident DR group than in the NDR group.These findings suggested that microvascular changes occur before the visible microaneurysm of DR, and eyes with NDR with a faster rate of VD decline are more likely to develop DR.We speculated that the rate of VD change may play an important role in DR development, and early microvascular alterations are the initial stages of DR that precede clinically detectable changes in fundus color photographs. 28,29DR pathogenesis involves hyperglycemia, oxidative stress, and inflammation that damage microvasculature, disrupt the blood-retinal barrier, and cause retinal hypoxia. 30High glucose concentrations and hypoxia can lead to structural dilation of retinal vessels, resulting in weakened walls and microaneurysms.][33] Advanced imaging techniques, such as OCTA, can identify these early retinal vascular alterations before they become apparent in fundus color photographs.
Meanwhile, our study found that VD was increased in some early-stage DR.We hypothesized that this might be due to tissue hypoxia induced vessel dilation. 34Previous studies showed that tissue hypoxia triggers retinal neurovascular autoregulatory dilation, increasing retinal blood flow and capillary congestion. 35Furthermore, hyperglycemia and inflammation may also contribute to early vasodilation. 36,37s a result, it is possible that the retinal VD may increase in some patients with early-stage DR.As the disease worsens, there is a loss of retinal capillaries and a decrease in vessel density in DR.Additionally, hypoxic and high glucose conditions may impair vasoconstriction of small arteriolar in the inner plexiform layer, leading to dilation of capillaries in the SCP. 38The variability of both blood glucose levels and vascular constriction may increase the variability of VD measurements in SCP.In contrast, the DCP sourced from choroidal blood vessels may exhibit more stable characteristics. 39ur study further showed that the incident DR group showed a greater VD decrease, especially in all of the inner sectors of SCP.This finding suggested that microvascular damage in the inner sector may be more significant compared to the outer sector.DR is characterized by microvascular damage that usually initiates in the macula, specifically in the perifoveal region.The foveal area of the retina, with its high metabolic demand, is particularly vulnerable to tissue hypoxia. 40Consequently, the inner sector adjacent to the fovea may experience earlier and more prominent microvascular injury.
Moreover, the rate of parafoveal VD reduction in the SCP was significantly higher in the incident DR group than in the NDR group, whereas the rate of VD reduction in the DCP was not significantly different between the two groups.2][43][44] Marques et al. 43 and Vujosevic et al. 41 indicated that decreased perifoveal VD was significant in deep layers between eyes with DR with and without progression.The findings have been inconsistent, partly because of the different sample sizes and confounding factors adjusted for.Consistent with our study, Aschauer et al. 25 reported a significant decrease in superficial but not deep VD in the eyes of DR patients.Durbin et al. 45 found that SCP demonstrates a good capability at differentiating healthy eyes from eyes with DR.The mechanism of decreased parafoveal VD in the SCP is not yet clear.Previous studies have shown that retinal layer thinning is observed during the early stage of DR and is associated with changes in VD. 25 We speculated that the superficial layer of the macula, which contains ganglion cells, may be impaired earlier during the initial stage of DR.Kim et al. 23 reported that progressive loss of macular ganglion cell/inner plexiform layer thickness was strongly correlated with a decreased in vessel density in the SCP.Gardner et al. 46 suggested that superficial VD and retinal nerve fiber layer thickness seem to be particularly vulnerable in the early stages of DR, which may be related to incipient disintegration of the neurovascular unit.Our findings support previous results suggesting that the superficial retina is more vulnerable to microvascular changes over time than the DCP in the development of DR.
The major strength of this study is that it builds on prospective data collected from a relatively large communitybased sample.All OCTA measurements were performed using a 6 Ã 6-mm scan pattern.VD in the SCP and DCP was measured using an objective quantification method (OCTA).This study has several limitations.First, the follow-up period was insufficient.Second, the blood-retina barrier dysfunction is an important phenotype of DR that cannot be evaluated using OCTA technology. 47Third, parafoveal SCP and DCP with a 6 Â 6 mm field of view were evaluated, and the peripheral retina was not obtained.Fourth, only eyes with good-quality images were included, which may have resulted in selection bias and affected the generalizability of our results.Fifth, manufacturer-recommended default settings were used for the SCP and DCP, and the inherent errors in segmentation were unknown.
In conclusion, eyes with DM patients showed evidence of significant capillary dropout over one year.The incident DR group presented a significantly more rapid decrease in parafoveal VD in the SCP than in the stable NDR group.Our longitudinal study further confirmed that the rate of retinal VD change plays an important role in DR development.Parafoveal VD in the SCP may serve as an early indicator of the preclinical stages of DR and needs further investigation.

Figure 1 .
Figure 1.Optical coherence tomographic angiography provides measurements of retinal parafoveal vessel density in the inner (3 mm) and outer (6 mm) ETDRS circles.Parafoveal vessel density in superficial and deep capillary plexus were measured respectively.

Figure 2 .
Figure 2. Flowchart for the screening and inclusion of patients.

p
¼ 0.007 and p ¼ 0.008, respectively).When comparing the two groups，the average parafoveal VD in incident DR in SCP showed a larger rate of change than that in the NDR group, decreasing by an average of 1.95 ± 0.45%/year and 0.45 ± 0.19%/year, respectively (p ¼ 0.002) (Figures3(A) and 4(A)).No differences were found in any sector in DCP, except in the inner temporal sector (p ¼ 0.019), when comparing the rate of parafoveal VD change over time between the incident DR and NDR groups (all p > 0.05) (Figures 3(B) and 4(B)).

Figure 3 .
Figure 3. Boxplots show the average parafoveal retinal vessel density change over 1 year in the two groups.Average parafoveal retinal vessel density changes in the incident DR and NDR groups in the superficial capillary plexus (SCP) (A) and deep capillary plexus (DCP) (B).

Figure 4 .
Figure 4. Inverted histogram showing the rate of parafoveal retinal vessel density over time (%/year) in the SCP (A) and DCP (B) adjusted for age and sex in the incident DR and NDR groups.

Table 1 .
Baseline demographic and ocular characteristics of study participants.

Table 2 .
Distribution of baseline parafoveal vessel density in participants who newly developed DR during follow-up and those with stable NDR.

Table 3 .
Comparison of the rates of parafoveal vessel density changes over 1 year in eyes with or without DR by sector.
2 : baseline vs. 1 year follow up P Ã : NDR group vs. DR group Values are presented as the estimated mean change (%/year) and have been adjusted for age and sex.95% CI: 95% confidence interval; NDR: non-diabetic retinopathy; DR: diabetic retinopathy; SCP: superficial capillary plexus; DCP: deep capillary plexus.Bold indicates significant at p < 0.05.

Table 4 .
Multivariable linear regression analysis for the relationship between parafoveal VD change per year and diabetic retinopathy.