Effect of Macular Vascular Density on Central Visual Function and Macular Structure in Glaucoma Patients

In patients with glaucomatous parafoveal scotoma, evidence of compromised vascular circulation was commonly seen. The purpose of this study is to evaluate the relationship between macular vascular density (VD) and central visual function and structure in glaucoma patients. We enrolled 46 eyes of normal tension glaucoma (NTG) patients with parafoveal scotoma. All subjects underwent measurement of segmented macular thickness in each layer and optical coherence tomography angiography (OCTA) to assess VD of macula. Correlation coefficients of VD with structural parameters were identified and multivariate regression analyses were performed to verify factors affecting the MD of SITA 10-2. Superficial VD in NFL, GCL and IPL showed significant correlation with thickness of those layers, but deep VD in INL did not show meaningful correlation with any structural parameters. However, deep VD showed significant correlations with central visual field parameters such as MD of SITA 10-2. By multivariate regression analysis, the significant factors affecting central visual function were deep VD. Different multivariate regression models including segmented macular thicknesses were compared and R2 value was best for the model with deep VD, not containing superficial VD (R2 = 0.326, p = 0.001). Assigning subjects as worse or better visual functional group using regression line, deep VD of worse functional group was significantly lower than that of better group. In couclusion, decreased deep VD was an independent risk factor for central scotoma in addition to structural thinning. Taking both macular thickness and vascular circulation into acount, the deterioration of central visual function could be predicted more precisely.


Results
The baseline characteristics of the 46 subjects are summarized in Table 1. 21 subjects were male and 25 were female, and mean untreated intraocular pressure(IOP) was 16.00 mmHg. The mean axial length was 24.68 ± 1.44 mm and central corneal thickness was 541.16 ± 30.24 μm. The MD and PSD values from SITA 10-2 were worse than the values from SITA 24-2 (−6.37 dB vs. −4.11 dB, respectively, for MD; 7.28 dB vs. 5.78 dB, respectively, for PSD). The correlations between VF sensitivity (1/L) of central 12 points (inside bold lines in Fig. 1) in SITA 24-2, VF sensitivity (1/L) of SITA 10-2 and MD of SITA 10-2 were analyzed to evaluate the Best corrected visual acuity, LogMAR 0.03 ± 0.06 Table 1. Baseline characteristics of study subjects. Data are presented as means ± standard deviation. MD = Mean deviation; PSD = Pattern standard deviation. repeatability information of VF results, and they all had good correlations (r = 0.828, p < 0.001 for VF sensitivity of 12 points in SITA 24-2 and of SITA 10-2; r = 0.776, p < 0.001 for VF sensitivity of 12 points in SITA 24-2 and MD in SITA . Several studies have evaluated the quality and repeatability of the density of retinal vascular plexus by OCTA [14][15][16][17] . Intraclass correlation coefficients (ICCs) of our VD measurement were 0.799 for superficial VD and 0.846 for deep VD, both showed good to excellent repeatabilities (all p < 0.05).
To investigate the trends in structural and vascular circulatory changes according to central visual function, subjects were divided into two groups on the basis of MD in SITA 10-2 ( Table 2). In group of better MD in SITA 10-2, the functional parameters(MD and PSD) in SITA 24-2, average cpRNFL and macular ganglion cell-inner plexiform layer (GCIPL) thickness were better than those of worse MD group as expected (p < 0.001 for functional parameters; p = 0.002 for cpRNFL; 0.012 for GCIPL). Segmented thickness and volume of NFL, GCL and IPL were thinner and smaller in worse MD group (p = 0.038, 0.002 and 0.007 for thickness, respectively; p = 0.047, 0.001 and 0.007 for volume, respectively). The mean values of macular vascular densities were different between two groups, but only deep vascular layer showed statistical significance (28.85% vs. 27.73%, p = 0.212 for superficial VD; 32.11% vs. 31.03%, p = 0.037 for deep VD). Table 3 showed the correlations between vascular densities and various structural or functional parameters. Superficial VD showed significant correlation with GCIPL and segmented NFL, GCL, IPL thickness, but deep VD did not show meaningful correlation with any structural parameters. In Figs 2 and 3, superficial VD increased according to thickness of NFL, GCL and IPL, but this tendencies were attenuated in deep VD, and there was not significant relationship between deep VD and INL in which we measured the density of deep vascular layer.
The noticeable results in Table 3 were the relationships between VD and VF test. Deep VD showed significant correlation with unlogged center sensitivity (mean value of sensitivity within central 10° as shown in Fig. 1) of SITA 24-2 as well as MD of SITA 10-2 (r = 0.385 and 0.390; p = 0.008 and 0.007, respectively). However, superficial VD showed different patterns -correlated with MD of SITA 24-2 and PSD of SITA 10-2 (r = 0.297 and −0.311; p = 0.045 and 0.035, respectively). Figure 4 showed the different mean VD values after dividing subjects into two groups based on the logarithmic regression line showing the relationship between structural parameters and functional parameters (MD of SITA 10-2). The patients with relatively worse function than expected (MD values below the regression line) had lower VD (except the case of superficial VD in IPL-MD graph). When patients were divided based on the regression line showing the relationship between MD and NFL, VD of patients with better function was 28.42% in superficial layer and 32.12% in deep layer, versus patients with worse function was 28.12% in superficial layer    Table 4, multivariate regression analysis identified GCL average thickness as the significant factor affecting MD from SITA 10-2 (p = 0.009 in model 1), and deep VD as marginally significant factor (p = 0.090 in model 1). After eliminating factors which were not statistically significant step-by-step, deep VD as well as GCL thickness turned out to be statistically meaningful (p = 0.044 for deep VD; p = 001 for GCL thickness in model 2). Similarly, in Table 5, deep VD and GCL thickness were marginally significant factors affecting center sensitivity from SITA 24-2 (p = 0.066 and 0.077). After eliminating factors which were not statistically significant step-by-step, deep VD was still statistically meaningful factor (p = 0.031 in model 2).
The results of various regression analysis models including macular vascular density and inner segmented macular thickness as variables were compared in Table 6. The difference of four analysis models were existence of macular VD in variables. The adjusted R 2 was best in model 3 and model 4 follows, which included segmented macular thicknesses and deep VD as variables. Model 2 which included only superficial VD and structural parameters showed a lower R 2 value than model 3 and 4 which included deep VD (adjusted R 2 = 0.281 for model 2, 0.326 for model 3 and 0.309 for model 4, all p < 0.05). R 2 differences between models were calculated; consistently with the above results, the difference between model 1 and 3 was greatest (R square change = 0.041, p = 0.107).

Discussion
The most important results of various studies of patients with NTG indicate that those with a central VF defect show problems with the vascular component [7][8][9][10][11][12]     These findings could support and visualize microvascular incompetence in glaucoma patients with central scotoma. Our study also focused on changes in the vascular status of the macula in patients who had early central scotoma with a glaucomatous optic disc, not only for superficial layer but deep layer. As shown in Table 3, superficial macular VD was correlated with structural and functional measurements, and deep macular VD did not show significant correlations with any structural parameter. However, deep macular VD was related to the parameters representing central visual function such as MD of SITA 10-2 and central sensitivity from SITA 24-2. By multivariate regression analysis, deep macular VD had a role on determining functional parameters along with structural parameters (Tables 4, 5). The evaluation of different models assessing the effect of variables on MD from SITA 10-2 revealed that the power of explanation represented by adjusted R 2 values was stronger with deep VD plus structural parameters than with superficial VD or with only structural indices. These outcomes imply an independent effect for deep macular VD on changes in the visual fields of our study participants. Our results showed denser retinal capillary of deep layer in subjects who were functionally better than those of functionally worse subjects (Fig. 4). In scatter plots of those comparisons, one outlier was found and that point might be the result of floor effect of macular structure. However, even when we removed that subjects, grouping according to logarithmic regression line were constant in NFL thickness-MD graph. In GCL-MD and IPL-MD graphs, the trends of comparison were maintained and statistical significance also remained constant (p = 0.084 for GCL, 0.032 for IPL). Several other studies were performed similarly to evaluate the relationship between visual function and VD. Takusagawa et al. suggested that only superficial VD was correlated with VF sensitivity 23 . Another study proposed that superficial VD was related with only mGCIPL and deep VD did not show significant relationship with any factor 24 . These findings were in conflict with our results. The main difference between those results and ours are the different study subjects. NTG subjects with central scotoma were only included in our study which were known to have a prominent relationship with vascular incompetence. Superficial retinal vascular layer may have effect on visual field deficits, but the influence could be confused because of the location of the superficial vascular network. It is difficult to determine whether decreased superficial VD of our subjects was independent factor of central scotoma or secondary epiphenomenone due to thinning of the NFL or GCL. However, the deep retinal vessels did not show meaningful relationships with structural parameters, even with INL, the layer where deep VD was measured. So the meaningful correlations between deep VD and central VF parameters might be    25 . More histological changes were found in deep capillary layer of diabetic patients than in the superficial capillary layer 26,27 . A study of diabetic mice found decreases in the retinal capillaries only located in the deep retinal layer 28 . Glaucoma patients with central scotoma also have features of impaired systemic vascular circulation, accordingly, the effect of decreased circulation may be more prominent in deep retinal capillary layer.
Several studies also demonstrated that there were decreased VD in both nonarteritic anterior ischemic optic neuropathy (NAION) and glaucoma 29  neuropathy starts with the compression of lamina cribrosa but NAION is caused by disorders of blood circulation in retrolaminar portion of the opbic nerve. However, the vascular incompetence could affect visual functional change or progression in glaucoma. So, NAION and NTG have common features in that retinal VD decreased compared with normal subjects 31 .
The present study has several limitations. First, there were small number of subjects with early glaucoma damage in this study. The further study with large number of patients are needed. Second, we used three machines to obtain RNFL thickness, macular segmentation thicknesses, and VDs. There could be issues of comparison between different machines. The absolute values of parameters may not be interchangeable between different machines, however, there are reports showing that correlation between different machines are good [32][33][34] . Therefore, analyzing correlations between the values from different machines could be performed. Third, there could be projection artifacts in en face imaging process. We had considered this interference when sorting out adequate subjects, however, it could hinder the precise evaluation of retinal capillary circulation. Fourth, although we were able to inspect only 3 × 3 mm macular scan when we conducted this study, it would have been more accurate if we used 6 × 6 mm scan which was known to have a higher diagnostic value 23 .
In spite of these limitations, in this study, it is remarkable that macular vascular density of deep layer may be an independent factor that affects central VF defect. Through OCTA, the effect of vascular incompetence can be visualized in deep layer of retina when we evaluate NTG patients with central scotoma. By considering both vascular circulation and thickness of macula, deterioration of central visual function in glaucoma patients could be predicted more precisely.

Materials and Methods
Study design and population. This cross-sectional study was performed according to the tenets of the Declaration of Helsinki. It was approved by the Institutional Review and Ethics Boards of Seoul St. Mary's Hospital, South Korea. Written informed consent was obtained from all participants.
A total of 46 normal tension glaucoma (NTG) patients with central scotoma who attended Seoul St. Mary's Hospital between March 2016 and January 2017 were enrolled in the study. The history of diabetes, hypertension or cardiovascular disease were documented and symptoms of vascular incompetence such as migraine, Raynaud's phenomenon and hypotension were also recorded. Subjects with uncontrolled diabetes, uncontrolled hypertension or cardiovascular event with sequale were excluded. The comprehensive ophthalmic examinations were done for all subjects including visual acuity, Goldmann applanation tonometry, slit-lamp examination, gonioscopy, automated perimetry using both 24-2 and 10-2 SITA program (Humphrey Visual Field Analyzer; Carl Zeiss Meditec, Inc, Dublin, CA, USA). Circumpapillary retinal nerve fiber layer (cpRNFL) and mGCIPL thickness were obtained using Cirrus spectral-domain OCT (Carl Zeiss Meditec, Inc, Dublin, CA). Macular structural segmentation was performed using Heidelberg SD-OCT device (Heidelberg Engineering, Heidelberg, Germany). OCT angiography was recorded by DRI OCT Triton system (Topcon, Tokyo, Japan).
All subjects included in this study should meet following criteria: (1) Best corrected visual acuity was 20/40 or better, (2) spherical equivalent(SE) was within ±5.0 diopters, (3) open angle on gonioscopy, and (4) intraocular pressure of lower than 21 mmHg. Exclusion criteria were as followings: (1) Patients with neurologic disease which could cause VF loss or retinal disease, (2) intracranial lesion which could make VF problem such as pituitary adenoma, (3) history of periorbital trauma, (4) advanced glaucomatous VF defect (mean deviation < −12dB) that may have diffuse central and peripheral VF loss.

Definition of central visual field defect.
A visual field test result was regarded as reliable when fixation loss was <20%, false-positive rate was <15%, and false-negative rate was <15%.
Initially to define subjects with central visual field defect (CVFD), we first analyzed SITA 24-2 results. CVFD was defined as VF defects within central 10° on pattern deviation probability map with clusters of three points with a probability of less than 5%, or two or more test points with a probability of less than 1% or smaller. All subjects had VF defects located within the superior or inferior hemifield of the central 10° regardless of the presence of defects outside the central 10° (Fig. 1).
Central retinal visual field function was evaluated through both SITA 24-2 and 10-2 results. Central retinal VF sensitivity was calculated by converting logarithmic dB scale to nonlogarithmic scale using formula [dB scale = 10 log(1/Lambert)] in central 12 points of SITA 24-2. Mean deviation (MD) and pattern standard deviation (PSD) were also evaluated in SITA 10-2.
Macular vascular density by OCTA. OCTA scans were acquired by the DRI OCT Triton system (Topcon, Tokyo, Japan). The DRI OCT Triton system uses a swept source laser with a wavelength of 1050 nm and scan speed of 100,000 A-scans per second. The OCTA is based on Topcon OCT angiography ratio analysis (OCTARA) algorithm and 3 × 3 mm volume of macular scan was obtained. An active eye tracker was used to reduce motion artifact during imaging. The automated layer segmentation was performed for superficial vascular plexus (2.6 µm below internal limiting membrane to 15.6 µm below the junction between inner plexiform and inner nuclear layers (IPL/INL)) and deep vascular plexus (15.6 µm below IPL/INL to 70.2 µm below IPL/INL). En face projections of volumetric scans allow for visualization of structural and vascular details within segmented retinal layer boundaries 14,35 . Highly myopic or hyperopic eyes were excluded to minimize magnification or minification effect of en face images. The images with image quality score over 70 were selected. Eyes with poor image qualities with following criteria were excluded: -(1) poor fixation resulting in double vessel pattern and motion artifacts, (2) blurred image that hinder the clarity of vessel contour, and (3) macular segmentation error. The quality of each image was independently evaluated by two glaucoma specialists (SJJ and HYP).
To calculate macular vascular density (VD), ImageJ software (National Institutes of Health, Bethesda, MD, USA) was used. As shown in other studies, a binary slab was created according to the ImageJ 'mean threshold' algorithm, which automatically computes the threshold value as the mean of the local grayscale distribution. Each binarized 8-bit image was converted into red-green-blue (RGB) color model and then split into the three channels (red, green, and blue). After assigning white pixels as vessel and black pixels as background, vascular density was defined as the ratio between vessel pixels and the total area (Fig. 5) 16,36,37 . Intraobserver repeatability of our VD measurement was also calculated from data of 20 normal eyes tested twice.
Measurement of segmented macular thickness. All patients underwent macular structural segmentation using Spectralis SD-OCT device (Heidelberg Engineering, Heidelberg, Germany). OCT scans were performed by the same experienced operator. The OCT scan images of included patients were absence of movement artifact and well centered.
Automated macular segmentation was performed by stored software -assigning retinal boundaries of the inner limiting membrane(ILM), the boundaries between the RNFL and the ganglion cell layer (GCL), the GCL and the IPL, the IPL and the INL, the INL and outer plexiform layer (OPL), the OPL and the outer nuclear layer (ONL). To minimize segmentation errors, segmented layer was manually verified and performed repeatedly.
The segmented retinal thickness map shows three concentric rings with diameters of 1, 3, and 6 mm. The intermediate and outer rings were divided into quadrants by two intersecting lines and the thickness of each zone was separately measured as follows: inner superior, inner nasal, inner inferior, inner temporal, outer superior, outer nasal, outer inferior, and outer temporal zone. The average thickness of each segmented layer was calculated as the mean value of 8 subfields excluding central foveola area with a 1-mm radius (Fig. 6) 38 . The volume of each layer within a 6-mm diameter was also measured automatically.
Statistical analysis. All statistical analyses were performed with SPSS version 24.0 (SPSS Inc., Chicago, IL). P < 0.05 was considered statistically significant. Descriptive results were calculated as the value of mean and standard deviation. The Student t-tests was used to evaluate structural and perfusional differences between groups divided according to the severity of central VF defect. The Shapiro-Wilk analysis was used for assessing normality and Pearson correlation analysis was used to evaluate the relationships between thickness of cpRNFL, thickness of GCIPL, thicknesses and volumes of segmented macular layers (NFL, GCL, IPL, INL, OPL and ONL), functional parameters of perimetry and macular vascular densities. Univariate and multivariate linear regression analyses were performed to identify significant factors that affected the functional values of VF tests. The logarithmic regression line was used to divide subjects into better or worse functional group than expected from the macular thickness. We performed VD comparisons between those groups using Student t-test. The statistical significance of differences in R square values of different regression analysis models was also assessed by multivariate analysis of variance (MANOVA).