Macular and peripapillary vessel density measurement in the evaluation of Non-proliferative Diabetic Retinopathy

Purpose: To compare vessel density in macular and peripapillary area between control subjects and patients with non-proliferative diabetic retinopathy (NPDR) using optical coherence tomography angiography (OCTA) and to evaluate the association between RNFL thickness and different stage of diabetic retinopathy. Methods: A total of 170 eyes (normal control, 43; mild NPDR, 43; moderate NPDR, 42; severe NPDR, 42) underwent OCTA imaging. Optical coherence tomography angiographic parameters were vessel densities of superficial capillary plexus (SCP), deep capillary plexus (DCP) in macular area and peripapillary area. Results: Vessel density of SCP and DCP in macular area, peripapillary area as well as RNFL thickness were 53.13%±3.03%, 52.07%±2.28%, 53.78%±3.66% and 128.86μm±11.32μm, respectively, in control subjects; 50.24%±3.81%, 47.40%±45.02%, 46.50% ±6.03% and 124μm±13.97μm, respectively, in mild NPDR; 46.80%±5.23%; 44.39%±3.99%; 44.64% ±4.23%; 121.02μm±20.86μm, respectively, in moderate NPDR; 42.82%±5.46%, 42.34%±5.14%, 43.16%±4.47%, 118.60μm±21.91μm in severe NPDR. The reduction of vessel density of SCP and DCP in macular area, peripapillary area as well as RNFL thickness were correlated with increasing severity of DR. Vessel density of SCP and DCP in macular area, peripapillary area and foveal density 300 (FD 300) in normal control were significantly higher than that of mild, moderate and severe NPDR groups. (all P<0.001). Vessel density of DCP shows better ability to identify the severity of DR (0.913; 95% CI=0.867-0.958; cut off value:0.75) than FD 300, vessel density of SCP in macular area and peripapillary area. Conclusion: Macular and peripapillary vessel density as well as RNFL thickness decreased as DR progresses. Vessel density in DCP could be an objective and sensitive indicator for monitoring progression of DR. OCTA might be clinically useful to evaluate microvascular and microstructural alterations in macula and optive nerve head (ONH), thus providing a new method to study the course of DR. age-related macular degeneration, retinal vascular occlusion, refractive error >3 diopters[D]),


Introduction
Diabetic retinopathy is the leading cause of blindness in economically active people worldwide and also known as a microvascular complication of diabetes mellitus. [1,2] Traditionally, fluorescein angiography (FA) has been used for the diagnosis and classification of DR even it has a number of potential side effects. [3,4] Optical coherence tomography angiography (OCTA) is an alternative non-invasive angiographic technique which can demonstrate a 3-dimensional non-invasive vascular mapping at the microcirculation level.
[5] With utilizing of OCTA, visualization of morphology of retinal and choroidal vessels and identification of ischemic area on a layer-by-layer basis all became possible [6]. Previous researches have used OCTA to indicate that vessel density in capillary plexus and choriocapillaries reduced as DR progressed[7-9].However, more and more research has indicated DR as a neurovascular disease as evidence showed prominent decreased RNFL thickness in early stage of DR [10,11]. Since optic nerve head (ONH) microcirculation as well as RNFL thickness can now be easily demonstrated as well as quantified at present, it would be meaningful to analyze the neurovascular alteration in optic disc as well as in macular area in diabetic patients.
Although OCTA has been used to conduct studies towards retina and optic disc [12][13][14], seldom did studies combine these two areas in one research. Actually, it should be cautioned when investigating DR patients with glaucoma, or DR patients with other fundus disease that might affect macular perfusion. Simply depending on vessel density of either one might cause bias. Hence, measuring vessel density of both optic disc and macular area might provide a more comprehensive assessment on DR patients even both microcirculation origin from different sources. In this study, we aim to investigate macular and peripapillary vessel density, and RNFL thickness between normal subjects and diabetic patients with different stages of non-proliferative diabetic retinopathy (NPDR) using OCTA and to analyze the relation among these indicators.

Subjects
We recruited patients with type 2 diabetes mellitus (T2DM) and healthy control subjects who presented between January 2017 to January 2018 from Guangdong General Hospital. Written informed consent was obtained from all participants. A single eye was randomly selected if both were eligible. Exclusion criteria for all participants were as follows: any other ocular disease that may affect ocular circulation (e.g. glaucoma, age-related macular degeneration, retinal vascular occlusion, refractive error >3 diopters[D]), intraocular surgery, hypertension exceeding 150/100mmHg, intraocular pressure (IOP) >21mmHg. Individuals with a family history of glaucoma were also excluded in the present study. All participants were tested for best IOP, corrected visual acuity (BCVA), and refractive error (autorefractometry). Slit-lamp and fundus examinations using direct and/or indirect ophthalmoscope were performed. ETDRS 35 degree 7-standard fields color retinal photographs (Topcon TRC; Topcon, Tokyo, Japan) was required for each participant. DR was graded according to the International Diabetic Retinopathy Severity Scale.
Fundus photographs evaluation and stage classification were done by two experienced graders (DC and ZNH).

Optical coherence tomography angiogram acquisition and analysis
Optical coherence tomography angiography was performed using AngioVue Optical coherence tomography angiography system (RTVue-XR Avanti; Optovue, Fremont, CA, USA, version 2017.1.0.151). Each subject underwent two imaging sections that include a 6 × 6-mm region centered in the fovea and a 4.5 × 4.5-mm scan centered in the optic disc. Split-spectrum amplitude decorrelation angiography (SSADA) software algorithm was used for computing a flow map of each scan. Motion Correction Technology of Optovue software was used to compensate for motion artifacts.
Imagine with inadequate scan quality ( 6/10) were excluded from the analysis. Retinal layers were segmented between the inner limiting membrane and the retinal pigment epithelium on the basis of the optical coherence tomography structural image. Segmentation of the angiogram would be checked to avoid analysis error. An en face optical coherence tomography angiogram was produced by maximum decorrelation projection of the segmented retina.
The software automatically segmented the tissue into the macula (superficial, deep, outer retina, and choriocapillaries) and optic nerve head (optic nerve head, vitreous, radial peripapillary capillary, and choroid). Macular vessel density parameters were measured in the superficial and deep retinal optical coherence tomography angiogram of the macula. The peripapillary vessel density was measured for the radial peripapillary capillary (Fig1).
The boundaries were as follows: a slab extending from the internal limiting membrane to 10 μm above inner plexiform layer was generated for detecting the vessel density of superficial capillary plexus. A slab extending from 10 μm above inner plexiform layer to 10 μm below the outer plexiform layer for the measurement of vessel density of deep capillary plexus and a slab extending from inner limiting membrane to retinal nerve fiber layer for the measurement of peripapillary vessel density.
Foveal avascular zone area (mm2) evaluated in the superficial plexus was identified as non-flow area and measured by the built-in software that delineated the segment of the foveal avascular zone automatically. Foveal thickness was calculated from the mean central point of fovea from inner limiting membrane to retinal pigment epithelium. Foveal density 300 (FD 300) is identified as the vessel density of the annular area surrounded by 300μm outwards with the FAZ inner boundary.
Two readers (DC and ZNH) evaluated the angiographic features on optical coherence tomography angiogram. Both readers were masked to the DR status. Image quality was considered by including images having Scan Quality of at least 6/10. In patients with poor images, we repeated the scans until an image with at least fair quality could be obtained.

Statistical Analysis
Statistical analysis was performed with SPSS 19.0 software (SPSS. Inc, Chicago, IL, USA). Qualitative variables are presented as number and percentage. Quantitative variables are presented as means and standard deviations. One-way analysis of variance test was used to compare the mean values of vessel density of superficial, deep and peripapillary area between groups. Receiver operator characteristic (ROC) curve was used to assess the relationship between the severity of DR and OCTA parameters and detect the sensitivity and specificity of the ROC curve. Criterion significance was assessed at the P < 0.05 level.
This study has been performed in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committee of Guangdong General Hospital (registration number: gdrec2016232A).

Demographics
A total of 170 eyes of 43 normal controls and 127 eyes of 127 diabetic patients with NPDR (mild NPDR, 43; moderate NPDR, 42; severe NPDR, 42) were assessed initially for image quality after they met inclusion and exclusion criteria described in methods. Baseline demographics of the entire cohort were comparable in all 4 groups (Table 1). Of the 170 patients, 92 (54.1%) were male and 78 (45.9%) were female. Gender and age showed no significant differences in between-groups comparison. Of the 170 selected eyes, 88 (51.8 %) were right eyes, and 82 (48.2 %) were left eyes. Patients were classified into normal group and NPDR (mild NPDR, moderate NPDR, severe NPDR) groups. Vessel density in different quadrants of peripapillary area and RNFL thickness were also shown in Table 2. Significant differences in between-groups comparison were found in the vessel density at all quadrants (p 0.05). In the present study, peripapillary vessel density was significantly correlated with vessel density of SCP and DCP (Fig.4). In terms of RNFL thickness, significant difference was also found in comparison among groups (p=0.03). Table 3 detailed the performance of FD 300, vessel density of SCP and DCP in macular area, and peripapillary area for identifying severity of DR based on AUROC curve. AUC in FD300, SCP and DCP of macular area and peripapillary area and were 83.5%, 90.5%, 91.3% and 93.0%, respectively. Vessel density of DCP shows better ability to identify the severity of DR (0.913; 95% CI=0.867-0.958; cut off value:0.75) than in FD 300, vessel density of SCP in macular area and peripapillary area (Fig. 5).

Discussion
In the present study, as DR progressed, the reduction in vessel density of macular area and peripapillary area as well as RNFL thickness were noted. As for peripapillary area, vessel density from different quadrants all decreased and was significantly related to the scale of DR. In terms of AUROC analysis, vessel density of DCP in macular area showed better ability to identify the severity of NPDR when compared to vessel density of SCP, RPC density and FD300.
The pathology of DR remained as the object of conjecture and hypotheses. Traditionally, DR has been considered as one of the microvascular diabetic complications and can generally be described as progressing through two stages, including NPDR and proliferative DR [15]. Hyperglycemia was considered to be an important factor in the etiology of DR and initiates downstream events such as basement membrane thickening, pericyte apoptosis and retinal capillary non-vessel density [16,17]. Indeed, DR is now more accurately defined as a neurovascular rather than a microvascular disease.
Further studies is definitely required for the clarification of which one trigger the occurrence of DR.
According to the previous research, the vascular networks become more complex with multiple overlapping capillary beds away from the foveal center. It has raised the concern that OCTA may not be able to display some of the finer capillary structure, which might influence the accuracy of vessel densities [31]. The alteration of foveal avascular zone, such as capillary drop out around FAZ area, has even been identified in patients with early stage of DR [32][33][34]. FD 300 is identified as the vessel density of the annular area surrounded by 300μm outwards with the FAZ inner boundary. It is reasonable to consider vasculature in FD300 was organized into less layers and FD300 as another sensitive indicators for the severity of DR. Thus, we intend to use FD 300 to predict the progress of NPDR in the present study. However, AUROC analysis showed that vessel density of DCP shows better ability to identify the severity of DR than other vessel density index. It is still of great interest to analyze FD 300 in different fundus diseases. Nevertheless, there are limitations of the current study. GCC measurement was not taken into account in the present research. Further study would be highly recommended to take this factor into consideration. Also, we acknowledge that this is merely an extrapolation of our results, a longitudinal study should be further conducted before such conclusion could be verified.
In conclusion, our results indicated that as DR progressed, the reduction of vessel density in macular and peripapillary area as well as RNFL thickness were prominent. Vessel density of DCP in macular area showed good ability to identify the severity of NPDR. With regard to the optic nerve, microvascular insufficiency and RNFL defects were noted distinctly. OCTA might be a good tool for the pathophysiological investigation for DR since OCTA can quantify vessel density and neurodegenerative changes in subjects with diabetes.  Vessel density in SCP, DCP and peripapillary area were reported as percentage.
NPDR: Non-proliferative diabetic retinopathy; PDR: proliferative diabetic retinopathy SCP: superficial capillary plexus; DCP: deep capillary plexus  Figure 1 Example images of vessel density and RNFL measurement in a control subject (a. radial peripapillary capillaries; b. superficial capillary plexus; c. deep capillary plexus; d. quantification of FD300.). Vessel densities and RNFL thickness were calculated automatically (red box).

Figure 2
A series of color fundus photos (top row), SSADA OCTA color-coded vascular perfusion maps (second to the last row) demonstrating the perfusion density changes seen as nonproliferative diabetic retinopathy progresses. From left to right, we start with a normal eye and progress to mild NPDR, moderate NPDR, and finally severe NPDR.

Figure 3
Box plots demonstrating the vessel density changes with normal control and different stage of DR.