The characteristics of choriocapillaris flow void in the unilateral polypoidal choroidal vasculopathy fellow eyes

To evaluate the morphological characteristics of flow void (FV) in the fellow eyes of the unilateral polypoidal choroidal vasculopathy (PCV). Fifty PCV fellow eyes (PCVF) and 31 age-matched normal ocular circulation controls were recruited in this retrospective study. The number of FV was analyzed according to the size in a centered 5 × 5 mm swept source optical coherence tomography angiography scans. We used indocyanine green angiography images to determine whether choroidal vascular hyperpermeability (CVH) has occurred. For the PCVF, the prevalence rate of CVH was 70% (35 of 50) The number of FVs was significantly lower in 400–25,000 μm2 (P = 0.005), 400–500 μm2 (P = 0.001), 525–625 μm2 (P = 0.001) and 650–750 μm2 (P = 0.018). compared to the controls. And showed no difference in size from 775 to 1125 μm2 between the two groups. The area under the receiver operating characteristic curve of PCVF with CVH and controls was 0.94 (95% CI 0.88–1.00) (P < 0.001). We found that the number of small FVs was significantly lower in the PCV fellow eyes than that in the eyes with control group.

decrease. The other was that ischemia due to congestion causing capillary dropout and FV increased. We speculated that the FV decrease that is due to congestion may appear in the small size of FV in early stage. Then it may increase in size later. To clarify this issue, we need to analyze FV sorted by size. To the best of our knowledge, an FV analysis by size in the eyes of subjects with PCV and fellow eyes has not yet been reported.
In the current study, we measured the number of FVs in groups stratified by the FV sizes to clarify the changes of FV in the fellow eyes of patients with unilateral PCV.

Methods
Patients. This was a retrospective comparative study that was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Fukushima Medical University (No. 2020-091), the Institutional Review Board of Fukushima Medical University approved the waiver for individual informed consent for this analysis.
We reviewed the medical records of 862 Japanese patients who were diagnosed as having unilateral PCV between June 2017 and August 2020 at Fukushima Medical University Hospital, Fukushima, Japan.
The inclusion criterion was a diagnosis of unilateral PCV evaluated by three ophthalmologists based on the presence of branching vascular networks, polypoidal lesions, and dilated aneurysmal lesions on ICGA (TRC 50DX, Topcon, Tokyo, Japan) according to the published criteria 1,2 . Thirty-one fellow eyes of patients diagnosed as having unilateral retinal vein occlusion (RVO) were selected as an age-matched control group. All RVO patients underwent fluorescein angiography (FA) for clinical evaluation, and were confirmed to have no abnormality. there was no evidence of retinal pathological change in any of the participants in either group.
The exclusion criteria for all eyes were as follows: (1) eyes with any combined retinal disease or intraocular disease, such as a macular hole, central serous chorioretinopathy, uveitis, or glaucoma; (2) patients with any systemic disease that affects CC circulation, such as hypertension, diabetes, gout, hypercholesterolemia, or neoplasm, or are allergic to the ICGA dye; (3) the refractive error is greater than − 6 diopters or the axial length is over 27 mm; (4) any observed drusen, retinal pigment epithelium (RPE) abnormality, or segmentation errors on the B-scan of the SS-OCT images.
In which 773 PCV patients met the exclusion criteria 1-3 and has been excluded. We then reviewed the SS-OCT images of the remaining 89 patients, 39 of whom met exclusion criterion 4, and were excluded.
All patients who did not meet the exclusion criteria were enrolled. Fifty fellow eyes of 50 patients were categorized in the PCV fellow eyes (PCVF) group. According to the results of the choroidal vascular hyperpermeability (CVH) status, the PCVF group was further divided into two groups; PCV fellow eyes with CVH (CVH[+]) and PCV fellow eyes without CVH (CVH[−]). Thirty-one patients who were diagnosed as having unilateral retinal vein occlusion (RVO) were selected as controls (Fig. 1 www.nature.com/scientificreports/ The CVH was defined as hyper fluorescence seen in the mid-phase of the ICGA images (10 min after the dye was injected), and the correspondent area of the OCTA images (Fig. 2).
The CVH was evaluated by two masked ophthalmologists. The 5 min images were as the reference and scored the ICGA images around 10 min. The scores were defined as follows. All participants of the control group were scored 0. A score of 1 was given when there was no distinguishable CVH in the area (CVH[−]); and a score of 2 was given when there was obvious and strong CVH in the area (CVH[+]). When the results were inconsistent, another experienced ophthalmologist made the judgment.
The subfoveal choroidal thickness (SFCT) was measured by masked ophthalmologists as the vertical distance between the RPE and the choroidoscleral border at the center of the fovea on the B-scan of the SS-OCT, using ImageJ software, version 1.52p (National Institutes of Health, Bethesda, Maryland, USA. https:// imagej. nih. gov/ ij/ index. html 10 . Imaging processing. To analyze the FV, we conducted the binarization of the CC slab and grouped the FV according to our previous research (Fig. 3) 11 .
We imported the CC slab bounded from 31 to 40 μm beneath the RPE reference 12 and the correspondent en-face structural image to MATLAB R2019a. (The MathWorks Inc. Natick, Massachusetts, USA). A compensation algorithm was used to eliminate the influence of the shadowing effect on the angiography images 13 . We then imported the compensated images into ImageJ and processed the "Phansalkar local threshold" (radius = 5) for binarization 14 . After binarization, we processed "Analyze Particles" and "watershed irregular" 15 to remove the signal noise.
Since the center of the macula of some subjects was displaced from the center of the image (− 210 to 455 μm), we performed manual correction to ensure that all images contained the same 5 × 5 mm 2 area centered at the fovea. We applied "Analyze Particles" to summarize the FV in each interval. Armonk, New York, USA). A P value of less than 0.05 was defined as statistically significant. The Kolmogorov-Smirnov test (K-S test) was performed to detect the normality of distribution. Because the K-S test of the FV number showed significance in most intervals, all variables are expressed as median and quartile deviation.
The independent samples median test, adjusted by Bonferroni correction, was used to calculate the differences between the groups. Spearman's rank correlation coefficient was performed to evaluate the correlation between FV and CVH. The Youden index was used to determine the best cut-off point for the receiver operating characteristic (ROC) curve between the CVH[+] group and the controls.
The size of FV analysis. We compared the number and distribution of FVs between the PCVF group and controls in the following intervals to confirm the difference between the fellow eyes of PCV and normal circulation eyes. The primary outcome was the number of FVs in different intervals between PCVF group and controls. The number of FVs was divided into the following seven groups. FVall: the FV sizes ranged from 400 to 25,000 μm 2 (16-1000 pixels). FV500: the FV sizes ranged from 400 to 500 μm 2 (16-20 pixels). FV625: the FV sizes ranged from 525 to 625 μm 2 (21-25 pixels). FV750: the FV sizes ranged from 650 to 750 μm 2 (26-30 pixels). FV875: the FV sizes ranged from 775 to 875 μm 2 (31-35 pixels). FV1000: the FV sizes ranged from 900 to 1000 μm 2 (36-40 pixels). FV1125: the FV sizes ranged from 1025 to 1125 μm 2 (41-45 pixels).
We calculated the total area of the FV sizes ranged from 400 to 25,000 μm 2 for comparison (FVarea). ROC analysis. We used binary logistic regression analysis for the CVH[+] and control groups depending on the results of the FV and CVH analyses.

The relationship between CVH and FV.
FV smaller than 750 μm 2 was grouped every 25 μm 2 (1 pixel) and subjected to a binary logistic regression analysis after standardization.
We performed ROC analysis to determine a cut-off value for maximizing sensitivity and specificity to discriminate between the two groups. The number of FVs. The relationship between FV size and FV number is presented in Fig. 4 and Table 2.
There were no differences in FVarea between the two groups (P = 0.586).
In a comparison between the PCVF and control groups, the number of FVs in the PCVF group at all intervals was lower than that in the control group. There was statistical significance between the PCVF and the control group in the following intervals; FVall (P = 0.005), FV500 (P = 0.001), FV625 (P = 0.001) and FV750 (P = 0.018). No difference was observed between the two groups in FV875 (P = 0.984), FV1000 (P = 0.930) or FV1125 (P = 0.984).  Table 4).

The ROC curve of the CVH[+] and control groups.
Based on the Youden index, the cut-off points of the FVs for predicting fellow eyes in PCV patients with CVH was 0.56, with a sensitivity of 91.4%, specificity of 86.0%, and an area under the curve (AUC) of 0.94 (95% CI 0.88-1.00) (P < 0.001) (Fig. 5).

Discussion
In our retrospective study, the number of FVs smaller than 750 μm 2 substantially decreased in the PCVF group compared to the control group; whereas, the total FV area did not differ between the two groups. Most of the PCVF showed CVH (70%; 35 of 50). The status of CVH negatively correlated to the number of FVs. This is the first study to determine the difference between the fellow eyes of PCV and control eyes using a non-invasive, commercially available SS-OCTA instrument.
Previous studies of ICGA have shown that the pathological changes of PCV and central serous chorioretinopathy have a similar choroidal vascular circulatory disturbance 16 . For the very early stage of the PCV which the choroid shows no pathological changes, recent researches 4,5,17 cannot tell the difference. However, we found that FV had decreased in the fellow eyes of PCV compared to the controls.
There were different protocols of CC slab obtained by OCTA 12,18-21 . We decided on a protocol that has been implemented in other studies as the standard of our research 12 . We then used our previously published FV analysis method which involved grouping the FV by size and performing statistical analyses on each group 11 . Our grouping analysis strategy can effectively detect the difference of FV in specific intervals which was not identified in previous studies.
Due to the power-law distribution (Fig. 6), there was a relatively larger number of smaller-sized FV, whereas larger FVs were of a relatively smaller number 22 . If a pathological change occurs in the smaller-sized FV, averagesized FVs may not be affected; on the other hand, if a pathological change occurs in the larger FVs, it was too scarce to affect the total number of FVs. Evaluating the number of grouped FV sizes has another advantage, in that drusen are relatively large in size and frequently appear in PCV 23 . The drusen area can decrease the penetration of OCTA laser to underestimating the FV density 20 . Our grouping strategy can reduce the influence of drusen by calculating the number of FVs rather than the FV area. www.nature.com/scientificreports/ The upper limit of the interval in our study was 1125 μm 2 . Due to the power-law distribution 24 , an FV larger than 1125 μm 2 contained less than 5% of the total, which were excluded from the statistics. An FV smaller than 750 μm 2 in PCVF was lower in number compared to control group because a small FV is more sensitive to radius changes. For example, when the radius is reduced by 1 μm in two FVs with original areas of 20 μm 2 and      www.nature.com/scientificreports/ 1125 μm 2 , the areas are reduced by 40% and 6%, respectively. Therefore, FVs of all sizes will be affected when hyperpermeability occurs; however smaller FVs are more substantially affected. The decreasing size of the FV represents an increase in the blood flow signal area, which might be due to the hyperpermeability of CC, an increase of vessel diameter, or an increase in blood flow.
Recently, OCTA has been effectively detect typical pathological manifestations of PCV 25 . Luo et al. reported similar results to ours in that the proportion of the choriocapillaris flow deficit area of PCV fellow eyes has a decreasing trend, but there was no statistical difference compared with an age-matched control group 5 . However, in the present research, the number of FVs in small sizes was lower compared to that of the controls, while the total area had no difference.
We believe that the insignificant difference in FV in Luo's research was due to the proportion area analysis strategy or the impact of underlying diseases, which may have resulted in an overall decrease in trend but was not statistically significant. Choroidal circulation can change due to many systemic and ocular factors, such as age [26][27][28] , choroidal thickness 29 , intraocular pressure 30,31 , and systemic diseases [32][33][34][35][36] . These factors may also change the morphology of the CC, resulting in FV changes. The previous research also revealed that the size of FV in diabetic retinopathy eyes is larger than that in control elderly eyes, and this trend positively correlated with the severity of diabetic retinopathy [32][33][34] . Hypertension can also increase the size and number of FVs [35][36][37] . Eyes with advanced Age-Related Eye Disease Study Category stage (AREDS stage) will cause an increase in the size and number of FVs 38,39 . We excluded subjects who had suspected drusen or RPE abnormality on the B-scan, or diseases that had an impact on the CC structure. It can be considered that the different number of FV was not due to from age or any other systematic diseases.
Previous histopathological studies have reported that the CC of the eyes with advanced PCV can have plumped endothelial cells, primitive or narrow lumen, or dilated vessels [40][41][42] . Our results showed that the reduction of the CVH status negatively correlated with number of FVs. However, the number of FVs of CVH[−] was between the number of those of CVH[+] and the control group, and showed no significant difference (Table 4) CVH is more commonly seen in eyes with pachychoroid 43 , but it also occurred in few PCV and fellow eyes [44][45][46][47][48][49] . A previous study of CVH using monkeys and rabbits had proved that the dye of ICGA was exuded and the CVH was formed due to vascular endothelial damage 50 , suggesting that CVH exhibits vessel abnormalities at the innermost choroid. Based on the mechanism of OCTA, the blood flow signals came from the movement of red blood cells 6 , and the decrease of the area without blood flow signals (FV) means that the area occupied by red blood cells increased. However, red blood cells should not exist outside the vessels. In other words, the diameter of the CC that parallel to the RPE was increased (Fig. 7).
One possibility is that due to the vasodilation of the choroidal vessels, the CC was squeezed toward the RPE, causing mechanical deformation of the CC (the diameter of the CC perpendicular to the RPE decreased and the diameter parallel to the RPE increased). As the CC lumen is squeezed, local blood flow can be slowed down and occurred turbulence. Slow blood flow and turbulence cause thrombosis and may be further related to the pathogenesis of PCV (Fig. 7G). Another possibility is that the diameters perpendicular to the RPE and parallel to the RPE were both dilated, induced CC endothelial damage, which causes the leakage of ICGA dye, and exhibited CC hyperpermeability. At the same time, due to endothelial damage, the concentration of vascular endothelial growth factors increased. Which in turn further promoted vasodilation and eventually formed pachychoroid disease (Fig. 7H). Figure 6. Demonstration of the FV distribution in a fellow eye of PCV with CVH and a control eye. The plot shows the typical power-law distribution of FV. The small sizes had relatively large numbers. The line shows the difference between the two eyes. Our strategy focused on a specific range, which avoids statistical biases due to the large disparity in numbers. PCVF PCV fellow eyes, Control age-matched control group, F-C difference between the PCVF and control groups. www.nature.com/scientificreports/ These two assumptions have the same characteristics on the CC slab of SS-OCTA: the diameter parallel to the CC slab increased, and the FV became smaller (Fig. 7C,D).
Our result shows that the hyperpermeability of PCVF occurs earlier than detectable RPE abnormalities on B-scans of SS-OCT.
Based on the Youden index, our model has high sensitivity (91.4%) and specificity (86.0%). There is currently no method to detect specific signs before the first eye of PCV onset. Our model shows a reliable prediction (AUC = 0.94) of PCV using non-invasive, dye-free and commercially available OCTA instruments. Therefore, this model, with the number of FVs as a predictor, is helpful. The change of FVs can give a clue for the early screening of PCV in the future.
The limitations of present study are its retrospective nature and small sample size. Although we strictly controlled the underlying conditions of both groups, fellow eyes of RVO may have had some unknown conditions that could have limited the generalizability of our findings in clinical practice. We used the watershed method to separate the large FVs, which may lead to underestimation of the number of large FVs and overestimation of the number of the small FVs. Moreover, CVH can also occur in other diseases, including central serous chorioretinopathy. Further research is necessary to specifically distinguish PCV with CVH from other diseases.
In conclusion, the number of FV was significantly lower in fellow eyes of PCV than in eyes with normal ocular blood circulation. In most of the PCV fellow eyes of this research, CVH occurred, which should be the earliest detectable sign of PCV when investigations yield normal results..

Data availability
The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.