Assessment of Macular Microvasculature in Healthy Eyes of Infants and Children Using OCT Angiography

Abstract

Purpose:

To assess macular vasculature in healthy infants and children using OCT angiography (OCTA).

Design:

Prospective cross-sectional study.

Participants:

One hundred thirty-five normal maculae of 89 healthy infants and children (mean age, 8.5±5.3 years; range, 9 weeks–17 years) treated at the Duke University Eye Center.

Methods:

We imaged 135 maculae of 89 pediatric patients using the standard Spectralis tabletop and investigational Spectralis with Flex module devices, both equipped with investigational OCTA software (Heidelberg Engineering, Heidelberg, Germany). OCT angiography images of the superficial vascular complex (SVC) and deep vascular complex (DVC) were analyzed for foveal avascular zone (FAZ) area and superficial and deep vessel density. We assessed effects of age, gender, race, axial length (AL), and central subfield thickness on FAZ and vessel density. Patients with both eyes imaged were assessed for agreement between the FAZ and vessel densities of the left and right eyes.

Main Outcome Measures:

The FAZ area, as well as vessel area density (VAD) and vessel length density (VLD) in the SVC and DVC.

Results:

The FAZ varied significantly with race; white patients showed a significantly smaller FAZ than black patients (mean difference, 0.11 mm²; P = 0.004). The FAZ did not vary with age, gender, or AL (P > 0.05). In the SVC, VAD and VLD varied significantly with age (P < 0.001) and AL (R² = 0.46; P < 0.001) but not gender (P >0.05). The SVC VLD was significantly different between races and ethnicities (P = 0.037), but VAD was not (P < 0.05). In the DVC, VAD and VLD also varied significantly with age (P < 0.001) and AL (R² = 0.46; P < 0.001) but not gender or race (P > 0.05). There was excellent agreement between the right and left eyes for FAZ (intraclass correlation [ICC], 0.97), SVC VLD (ICC, 1.00), and DVC VLD (ICC, 1.00).

Conclusions:

Quantitative studies of pediatric perifoveal vasculature should consider age, race, and AL. In eyes with unilateral disease, the perifoveal vasculature in the unaffected eye may be used as a control comparison because there is excellent agreement between eyes.

OCT angiography (OCTA) has greatly advanced the study of diseases affecting retinal vasculature^1,2 and allows for high-resolution, depth-resolved visualization of areas of retinal ischemia and neovascularization in diseases such as diabetic retinopathy,^1,3–7 neovascular age-related macular degeneration,¹ and familial exudative vitreoretinopathy.⁸ Assessment of retinal vascular features such as the foveal avascular zone (FAZ) area are clearly and reproducibly⁹ quantifiable on OCTA images. The noninvasive, dyeless nature of OCTA may be ideal to image infants and children; it allows frequent monitoring and follow-up and provides information regarding responses to treatment.¹⁰ Application of OCTA to infants and children to examine pediatric retinal microvasculature may provide insights on healthy microvascular development and deviations that could indicate disease.

Most studies using OCTA to examine and quantify normal, healthy retinal microvasculature have been performed in adults; a few studies include children, but none have included infants.^9,11–20 These studies in adults show that factors such as age^15,17,21 and gender^17,21 are significant factors when assessing the FAZ and vessel densities in the superficial and deep retinal vascular layers. OCT angiography imaging of adults and older children with a history of preterm birth show persistent foveal vasculature and small or absent FAZ.^22,23 Whether these factors manifest as physiologic differences in the perifoveal vascular development in infants and children is currently unknown. Although spectral-domain OCT imaging has been used to study developing infant structural retinal layers,^24–26 knowledge of infant retinal vascular development has been limited to a few human histologic studies²⁷ and animal models.²⁸ A previous histologic study by Hendrickson et al²⁵ demonstrated that foveal morphologic features continue to develop until 3.8 years of age and foveal maturity is reached at 13 years of age as soon as cone packing is complete. This infant foveal development was visualized on structural OCT by Vajzovic et al,^24,26 Maldonado et al,²⁹ and Lee et al.³⁰ In recent reports, our group demonstrated the feasibility of acquiring depth-resolved OCTA images of retinal microvasculature in healthy, full-term infants³¹ as well as infants with retinopathy of prematurity,³² retinoblastoma,³³ and familial exudative vitreoretinopathy.^8,34 To distinguish between pathologic and physiologic retinal vascular development in pediatric patients on OCTA, normative data of healthy eyes of full-term pediatric participants across the range of ages from birth to adulthood without intraocular disease are needed.

In this study, we acquired OCTA images of healthy maculae of pediatric patients born full term, from neonates to children 17 years of age, to develop normative data of healthy eyes and test several hypotheses. We hypothesized that: (1) the FAZ and perifoveal superficial and deep vessel densities would vary with age,^19,20,25 (2) the FAZ and perifoveal vasculature would vary based on gender,^17,21,35 (3) the FAZ and perifoveal vasculature would vary depending on race or ethnicity because of previously reported racial and ethnic differences in foveal morphologic features,^36,37 and (4) the FAZ and perifoveal vascular patterns would show interocular symmetry.^35,36

Methods

Study Population

This prospective, cross-sectional study was approved by the Duke University Health System Institutional Review Board and adhered to the Health Insurance Portability and Accountability Act and all tenets of the Declaration of Helsinki. Parents or legal guardians gave written informed consent for study participation. From September 2017 through April 2018, we imaged normal maculae of pediatric patients who were receiving care at the Duke University Eye Center with the standard Spectralis HRA+OCT tabletop and the investigational Spectralis with Flex module (Heidelberg Engineering, Heidelberg, Germany),³¹ both integrated with investigational OCTA software that has since received 510(k) clearance by the Food and Drug Administration. Patients with clinic appointments were imaged while sitting upright at the tabletop unit, whereas patients undergoing an examination under anesthesia or surgical procedures were imaged with the portable Flex unit while supine under general anesthesia in the operating room. For patients undergoing an examination under anesthesia, the axial length (AL) of the eye was also measured using a Master-Vu A-scan (Sonomed Escalon, Wayne, PA). To reduce the additional time under anesthesia for research imaging in the operating room, only 1 eye was imaged. For patients imaged in the clinic, one or both eyes were imaged depending on patient cooperation.

We included in the study normal eyes of healthy infants and children 0 to 18 years of age with a history of term birth (≥37 weeks’ gestational age), defined as those without any evidence of intraocular disease per ophthalmic examination by a pediatric optometrist or ophthalmologist at the Duke Eye Center. Patients undergoing examination under anesthesia or surgical procedures had diagnoses such as nasolacrimal duct obstruction, strabismus (the nonamblyopic eye was imaged³⁸), chalazion, trauma in the fellow eye (the unaffected healthy eye was imaged), nonpathologic myopia, astigmatism, combined hamartoma in the fellow eye, and a unilateral congenital cataract in the fellow eye without evidence or suspicion of congenital infection or other pathologic features.

Patients were excluded from the study if they were born prematurely²³ (<37 weeks’ gestational age) or had any systemic medical or neurologic condition that may impact foveal development or retinal vasculature, such as sickle-cell disease, diabetes,^6,39 Down syndrome, albinism, or seizures. Eyes were excluded if they had amblyopia,^38,40 had sustained prior trauma or had undergone intraocular surgeries, or had other ocular conditions or diseases including pathologic myopia, Stickler syndrome,⁴¹ and uveitis.⁴² Eyes were also excluded if the fellow eye had Coats disease,⁴³ Sturge-Weber syndrome, or unilateral congenital glaucoma⁴⁴ because of the potential for vascular impact in the seemingly unaffected eye. Eyes with high refractive errors that prevented the acquisition of focused OCTA images were also excluded.

Imaging and Analysis

All OCTA images of the macula were acquired using the 10°×10°scan angle that included 512 A-scans × 512 B-scans. Motion and fixation artifacts were minimized on the resulting OCTA images with the Spectralis’ TruTrack Active Eye Tracking feature. Segmentation of retinal layers was performed automatically by Spectralis software to generate en face OCTA images of vascular layers, including the superficial vascular complex (SVC) and deep vascular complex (DVC). The OCTA image of the SVC was generated by flow signal between the internal limiting membrane to 17 mm above the lower boundary of the inner plexiform layer. The DVC boundaries were from 17 mm above the lower boundary of the inner plexiform layer to the lower boundary of the outer plexiform layer. The Spectralis’ projection artifact removal algorithm was applied to the DVC OCTA images. An experienced reader (STH) reviewed each image for correct segmentation of the retinal layers and manually corrected any as necessary.

For each OCTA image, we recorded the following data output provided by the Spectralis software: focus setting, OCTA quality, and 1-mm central subfield thickness (CSFT). The focus setting corresponds most closely to refractive error in spherical equivalent.⁴⁵ The OCTA quality, calculated by the Spectralis software, was the average quality of all the B-scans in the entire volume scan. The CSFT was obtained using Spectralis’ retinal thickness map.

Each OCTA image was also graded for quality (by STH) as great, good, fair, or poor. Images were of great quality if the reader felt confident that she could trace all of the vessels throughout the image; of good quality if the FAZ boundaries were clear, but there was artifact (i.e., because of motion or dry eyes) affecting approximately less than 20% of the image; of fair quality if the FAZ boundaries were clear, but approximately more than 20% of the image was affected by artifact such that the image should be excluded from vessel density analysis; and as poor quality if the image was of such low quality that it should be excluded from both FAZ and vessel density analysis. For any eyes with repeated images obtained at the same imaging session, we chose the best image for analysis based on the quality assigned by the reader (rather than the Spectralis output for OCTA quality, because these were averaged across the entire scan and thus may not reflect patches of signal dropout in an otherwise high-quality image); if the images were of the same quality, then the image with the higher number for the average OCTA quality was used. For analysis involving 1 eye per participant where 2 eyes were imaged, the eye with the higher reader quality grade was used, or if equivalent, then the eye was selected randomly.

To measure the FAZ area in each eye, we used a customized algorithm in MATLAB (MathWorks, Natick, MA; see “Interreader Reproducibility” below) to segment the central avascular area enclosed by the innermost capillary plexus^16,18,39 on the en face OCTA image of the DVC¹⁶ (Fig 1). Briefly, OCTA images were enhanced first using the Hessian multiscale filtering⁴⁶ so that vascular and avascular regions could be distinguished more effectively. Then, a point inside the FAZ area was autolocated to use as a seed point for the FAZ area segmentation using a level sets method developed by Li et al.⁴⁷ All automated FAZ tracings were reviewed and corrected manually as indicated. Calculation of the FAZ area in square millimeters used the scaled x- and y-dimensions provided by the Spectralis software to convert from pixels to square millimeters.

Figure 1.

MATLAB (MathWorks, Natick, MA)-processed OCT angiography images of the superficial vascular complex (SVC) and deep vascular complex (DVC) of a 14-month-old infant undergoing an examination under anesthesia for persistent hyperplastic primary vitreous in the fellow eye. The original images were binarized by comparing the intensity of each pixel against a local threshold that is the average intensity of pixels inside the 7×7-pixel box centered at the pixel of interest. Then the vessel area density (percentage) was measured by calculating the percentage of the overall 10°×10° scan that was occupied by vessels, excluding the foveal avascular zone (FAZ) area. The images were also skeletonized to measure vessel length density per millimeter. The original DVC image was filtered to facilitate automatic measurement of the FAZ.

To assess the vessel density in the SVC and DVC, we measured both the vessel area density (VAD) and vessel length density (VLD) in MATLAB. First, OCTA images were binarized by comparing the intensity of each pixel against a local threshold, defined as the average intensity of pixels inside the 7×7-pixel box centered at the pixel of interest. An advantage of a local threshold in comparison with a global threshold is that it is better in preserving small vessels during the binarization process. The VAD was then measured by calculating the percentage of the overall 10°×10°scan that was occupied by vessels, excluding the FAZ area as measured above (Fig 1). The VLD was then calculated by skeletonizing the OCTA image to obtain the vessel length per millimeter (mm⁻¹; Fig 1).

Interreader Reproducibility

We assessed the reproducibility of manual segmentation of the FAZ area between 2 readers (L.V. and S.T.H.) on a subset of the OCTA images in the study (n = 30). For each image, each reader independently traced the boundary of the FAZ on the en face OCTA image of the DVC, where the FAZ boundaries were most defined, using the freehand tool in FIJI (available at: https://imagej.net/Fiji).48 Agreement between the manual segmentation of the FAZ area between the readers was assessed using the intraclass correlation coefficient (ICC). Then, we validated the customized MATLAB algorithm for automatic FAZ segmentation using the same subset of images (n = 30). We assessed the reproducibility of FAZ area segmented by the gold standard of manual segmentation by one of the readers (STH) and the automatic segmentation using the customized MATLAB algorithm. Agreement between the manual and automatic segmentation was assessed using the ICC.

Statistical Analysis

Statistical analysis was performed using SAS software version 9.4 (SAS Institute, Cary, NC) and JMP software version 13.0 (SAS Institute). Analysis to assess for significance of age, gender, race and ethnicity, AL, CSFT, focus setting, and OCTA quality on the OCTA measurements for FAZ area and vessel density (SVC VAD, SVC VLD, DVC VAD, and DVC VLD) used images of 1 eye of each patient. Multivariate analysis was used to assess for race, age, and interaction in models for FAZ area and vessel density. The relationships between continuous predictor variables and the continuous outcomes were assessed using linear regression. For vessel density, both linear and quadratic fits were applied in the regression model. In a model for FAZ area, race (black and white only, because smaller sample sizes and other races and ethnicities were excluded from this subanalysis between groups) and age were included, and an interaction term for race and age was tested for significance. The relationships between categorical variables and the outcomes were assessed using the 2-sample t test and Wilcoxon rank-sum test, respectively. For the subset of patients who had both eyes imaged during the same visit, we compared right and left eyes of each patient. Agreement between the eyes for FAZ area, SVC VAD, SVC VLD, DVC VAD, and DVC VLD was assessed using a paired t test and ICC.

Results

We successfully acquired OCTA images of 135 healthy eyes of 89 infants and children (mean age, 8.5±5.3 years; range, 9 weeks–17 years; females, 54%; white, 43%; black, 36%; Hispanic, 10%; Asian, 2%; other race or ethnicity, 4%; Table 1). No manual correction of the automated retinal layer segmentations was needed for any of the images. Of the 89 OCTA images (1 eye per patient), all underwent FAZ analysis, and 8 were excluded from vessel density analysis because of artifacts affecting image quality. Interreader reproducibility of manual FAZ segmentation was excellent (ICC, 1.00; 95% confidence interval, 0.99–1.00). When compared with manual FAZ segmentation, the customized MATLAB algorithm for automatic segmentation showed excellent agreement (ICC, 0.99; 95% confidence interval, 0.98–1.00).

Table 1.

Foveal Avascular Zone Area of Pediatric Patients

			Foveal Avascular Zone Area (mm2)
Race or Ethnicity	No. (%)	Age (yrs), Mean ± Standard Deviation	Mean ± Standard Deviation	Median	95% Confidence Interval	Range
White	42 (46)	7.5±5.4	0.289±0.108	0.280	0.255–0.322	0.05–0.63
Black	32 (43)	10.6±4.6	0.398±0.217	0.335	0.320–0.477	0.12–1.17
Hispanic	9 (10)	7.0±6.7	0.381 ±0.150	0.420	0.266–0.496	0.07–0.59
Asian	2 (2)	8.0±5.7	0.410±0.212	0.410	—	0.26–0.56
Other	4 (4)	5.9±5.7	0.433±0.099	0.430	0.275–0.590	0.33–0.54
Overall	89 (100)	8.5±5.3	0.347±0.168	0.320	0.311–0.382	0.05–1.17

Mean FAZ area of 1 eye from each of the 89 patients was 0.35±0.17 mm²; Table S1 (available at www.aaojournal.org) shows the FAZ area mean, median, and range split for various age categories (0–2 years, 3–6 years, 7–9 years, 10–12 years, 13–15 years, and 16–18 years). The FAZ area varied significantly among the different races and ethnicities (P = 0.008, Wilcoxon rank-sum test; Table 1). Patients who identified as white showed a significantly smaller FAZ than those who identified as black (mean difference, 0.11 mm²;P = 0.004, Wilcoxon rank-sum test). Central subfield thickness was significantly predictive of FAZ area (slope, −0.004; P < 0.001; R² = 0.49, linear regression); the thinner the CSFT, the larger the FAZ. Factors that were not significant predictors of FAZ area included age, gender, and AL (P > 0.05). Furthermore, assessment of whether the relationship between age and FAZ area was significantly different for black and white patients showed no significance (P = 0.52 for interaction). In the final regression model of FAZ area with age and race, race was significant (P = 0.02), but age was not (P = 0.06).

Vessel density in the SVC, measured as both VAD (percentage) and VLD (vessel length per millimeter), was predicted by age (Fig 2), AL, and race or ethnicity (Table 2). The relationship between age and vessel density was fit better to a quadratic, rather than linear, model; VAD and VLD increased with age until 10 to 15 years of age, after which vessel density started to decrease. Further subgroup linear regression analysis of the SVC VAD and VLD for 0 to 7 years of age and 8 to 18 years of age showed that age was a significant predictor in children 0 to 7 years of age for SVC VAD and VLD (P < 0.001; R² = 0.45 and R² = 0.50, respectively); although age was also a significant predictor of SVC VLD in children 8 to 18 years of age, it did not explain most of the variation in this linear model (P = 0.017; R² = 0.12; Fig S1, available at www.aaojournal.org). In the patients from whom we acquired an AL measurement (age range, 9 weeks–5 years; n = 25 eyes; AL mean, 21.1±1.24 mm; AL median, 21.2 mm; AL range, 18.7–23.2 mm), AL was a significant predictor of VAD and VLD (n = 23; P < 0.001; VAD R² = 0.45; VLD R² = 0.43). The SVC VLD was significantly different between races (P = 0.036), but VAD was not. The SVC vessel densities were not significantly different between males and females (P > 0.05).

Figure 2.

Graphs showing vessel density in the superficial vascular complex (SVC) and deep vascular complex (DVC) reported in terms of both vessel area density (VAD [percentage]) and vessel length density (VLD [per millimeter]) in linear (red line) and quadratic (green line) regression models with age. For the SVC, both linear and quadratic models were significantly predictive of VAD and VLD (P < 0.001 for all). The relationship between age and vessel density (VAD and VLD) was better fit when using the quadratic (adjusted R² = 0.57 and adjusted R² = 0.54, respectively) rather than linear model (R² = 0.38 and R² = 0.27, respectively). For the DVC, the linear and quadratic regression models were a poor fit for both VAD (linear R² = 0.10, quadratic adjusted R² = 0.27) and VLD (linear R² = 0.03, quadratic adjusted R² = 0.27), indicating that although age significantly predicted the DVC VAD and VLD (P < 0.001, quadratic fit), most of the variation in the data was not explained by age.

Table 2.

Vessel Area Density and Vessel Length Density of the Superficial and Deep Vascular Complexes in Pediatric Patients

	Superficial Vascular Complex		Deep Vascular Complex
Race or Ethnicity	Vessel Area Density (%)	Vessel Length Density (mm−1)	Vessel Area Density (%)	Vessel Length Density (mm−1)
White
No.	38	38	38	38
Mean±SD	34.4±2.7	27.5±3.4	39.3±1.7	33.1±2.2
Median	34.6	27.0	39.8	33.0
95% CI	33.5–35.3	26.4–28.6	38.7–39.9	32.4–33.8
Minimum–maximum	29.5–39.3	22.2–34.8	34.2–43.4	28.9–38.1
Black
No.	28	28	28	28
Mean±SD	36.1±2.1	29.6±3.1	39.4±1.1	33.2±2.0
Median	36.4	30.0	39.8	33.5
95% CI	35.3–36.9	28.4–30.8	39.0–39.9	32.5–34.0
Minimum–maximum	31.5–40.1	23.0–38.1	37.4–41.1	29.4–37.8
Hispanic
No.	9	9	9	9
Mean±SD	34.8±2.8	28.2±3.4	39.3 ±2.1	33.7±1.8
Median	35.1	28.2	39.4	33.6
95% CI	32.7–37.0	25.6–30.8	37.7–40.9	32.3–35.0
Minimum–maximum	30.7–38.6	24.5–33.6	35.5–42.0	31.2–36.6
Asian
No.	2	2	2	2
Mean±SD	35.1±1.8	27.6±2.3	40.3±0.2	34.3±1.5
Median	35.1	27.6	40.3	34.3
95% CI	—	—	—	—
Minimum–maximum	33.8–36.3	26.0–29.2	40.1–40.4	33.2–35.3
Other
No.	4	4	4	4
Mean±SD	33.8±1.0	25.5±1.4	38.9±0.8	31.0±2.0
Median	33.4	25.5	38.9	31.1
95% CI	32.2–35.4	23.2–27.8	37.67–40.1	27.8–34.2
Minimum–maximum	33.1–35.3	23.8–27.3	38.1–39.6	28.8–33.1
Overall
No.	81	81	81	81
Mean±SD	35.0±2.5	28.2±3.3	39.4±1.5	33.1±2.1
Median	35.3	28.2	39.7	33.2
95% CI	34.5–35.6	27.5–28.9	39.0–39.7	32.7–33.6
Minimum–maximum	29.5–40.1	22.2–38.1	34.2–43.3	28.8–38.1

CI = confidence interval; SD = standard deviation.

For the DVC vessel density, age and AL were significant predictors. Age significantly predicted the DVC VAD and VLD (P < 0.001, quadratic fit); however, the linear and quadratic regression models were a poor fit for both VAD (linear R² = 0.10, quadratic adjusted R² = 0.25) and VLD (linear R² = 0.03, quadratic adjusted R² = 0.28), indicating that although age was significant, most of the variation in the data was not explained by age (Fig 2). Further subgroup linear regression analysis of the DVC VAD and VLD for those 0 to 7 years of age and 8 to 18 years of age showed that age was a significant predictor in children 0 to 7 years of age for DVC VAD and VLD (P < 0.001; R² = 0.41 and R² = 0.43, respectively); although age was also a significant predictor of DVC VLD in children 8 to 18 years of age, it did not explain most of the variation in this linear model (P = 0.023; R² = 0.11; Fig S1). Axial length (in patients 9weeks–5 years of age) significantly predicted vessel density and was a moderate fit for VAD (n = 23; P = 0.002; R² = 0.38, linear regression) and VLD (n = 23; P < 0.001; R² = 0.47, linear regression). The DVC VAD and VLD were not significantly different between races or genders.

The mean focus setting used was 0.584±2.85 (n = 89; median, 1.20; range, −6.54 to 6.96). Assessments of whether the focus setting and OCTA quality significantly affected the FAZ area and vessel density (VAD and VLD) measurements were unreliable. The regression model relationships between the variables were poor. Thus, most of the variation was not the result of the focus setting and OCTA quality.

In the subset of children with both right and left eyes imaged (n = 40 patients), there was no difference between the 2 eyes for the FAZ area and vessel length densities (P = 0.30, paired t test). The mean FAZ area difference between the 2 eyes was 0.01±0.01 mm². The mean differences for the vessel densities were as follows: SVC VAD, 0.02±0.30%; SVC VLD, 0.11±0.36 mm⁻¹; DVC VAD, 0.05±0.27%; and DVC VLD, 0.07±0.20 mm⁻¹. The right and left eyes showed remarkable similarity for the FAZ area (ICC, 0.97; 95% confidence interval, 0.94–0.98). Agreement in vessel densities was moderate to good when comparing the VLD of the right and left eyes for the SVC and DVC (ICC, 0.72 and 0.82, respectively) and moderate for VAD comparisons in the SVC and DVC (ICC, 0.69 and 0.38, respectively).

Discussion

We successfully imaged and analyzed depth-resolved perifoveal microvasculature in healthy full-term infants and children and assessed the relationships between the FAZ area and vessel density with the following patient characteristics: age, gender, race or ethnicity, AL, and central subfield thickness. This study utilizes OCTA to quantify and analyze the normal perifoveal vasculature in a large cohort of healthy infants and young children (PubMed search on June 13, 2019, for optical coherence tomography angiography AND [pediatric OR infant OR children]). In our cohort of pediatric patients, we found that FAZ area differed between white and black children, whereas vessel density, particularly in the SVC, varied with patient age.

The racial differences in FAZ area between white and black pediatric patients are consistent with the racial and ethnic differences in foveal pit morphologic features; the larger FAZs seen in black patients could be explained by the deeper and broader foveal pit morphologic features.³⁶ Furthermore, we observed that foveal thickness showed a significant inverse relationship with FAZ area: the thinner the CSFT, the larger the FAZ, a finding also noted previously by Falavarjani et al.²⁰ A previous study by Borrelli et al¹⁸ in 5- to 17-year-olds that included only 1 black patient did not find an association between race and FAZ area. In terms of vessel density, different races and ethnicities did not show significant differences in the DVC, but the results for the SVC were mixed: the VLD was significantly different, whereas the VAD difference was not significant (P = 0.07). Such differences caused by different methods of vessel density analysis are important to note when determining the significance of various patient characteristics on perifoveal vasculature.

Our study contributes to the existing literature by providing information regarding OCTA measurements for FAZ and vessel density from infancy through the adolescent years. There seems to be a quadratic relationship between age and vessel density: VAD and VLD increased with age until approximately 10 to 15 years of age, after which the vessel density started to decrease. Interestingly, the peak of vascular density coincided with the age at which cone packing in the human fovea is complete, marking the attainment of foveal maturity.²⁵ Our study was limited to pediatric patients, and thus we were unable to determine whether vessel density continues to decrease throughout young adulthood or stabilizes. A previous study by Iafe et al¹⁵ that analyzed the VLD of patients 9 to 88 years of age showed decreasing mean VLD between the fifth and sixth decades of life, and the decrease became significant between 60 and 70 years of age. Another study by Leng et al¹⁹ that studied patients 8 to 87 years of age with a mean age of 42.8±19.9 years found decreasing vessel density with increasing age. Together, our study findings suggest that perifoveal vasculature continues to grow starting in infancy until approximately the time of foveal maturity, at which point the vascular density stabilizes and then decreases later in adulthood.

We did not observe an effect of age on FAZ area in infants and children, even after including race in the multivariate analysis. This finding contrasts with that of Borrelli et al,¹⁸ who showed that increased age correlated with larger FAZs. Also, although we did not find any differences between genders for FAZ area or vessel density in our study, other studies have reported larger FAZs in females¹¹ and increased superficial capillary plexus density in males.¹⁸ The different racial and ethnic composition of our study population when compared with that of others may explain some of the different results among studies; however, multivariate analysis of our data did not show a significant relationship of age and gender as a function of race. Other explanations to be considered include differences in study patient characteristics (whether patients in other studies were born full term²³), statistical analysis (whether only 1 eye from each patient was used for independent data), vascular layer methodologies used by different authors to generate the en face OCTA images,⁴⁹ and whether variations in AL were accounted for when calculating FAZ area on the OCTA images.^13,50 Furthermore, differences in sample size, especially with regard to the number of infants and young children, may play a role in statistical differences. Our study assessing 89 eyes of healthy patients ranging from 9 weeks to 17 years of age contributes new information to the existing knowledge of quantitative OCTA measurements. Many factors such as race and ethnicity, age, and AL need to be considered when one evaluates whether a patient’s FAZ and vessel density are within so-called normal healthy ranges.

The cross-sectional design of our study limits our ability to draw conclusions regarding the effect of age on perifoveal microvasculature. Further follow-up of patients is necessary to assess the longitudinal changes in FAZ and vessel density as patients age. The imaging locations and methods also present different challenges. In the operating room, to minimize the additional time under anesthesia to acquire research images, we imaged only one rather than both eyes, thus precluding the assessment of interocular symmetry in these infants. In the clinic, to minimize impediment on clinic flow, AL was not measured, thus limiting our AL measurements in children older than 6 years. Although the same Spectralis OCTA software and camera head were used to image patients in both the operating room and the clinic, the patients were supine and had been administered supplemental oxygen and anesthetic agents for imaging in the operating room, and patients were upright in the clinic; this may have affected quantitative measures such as of VAD.

Based on the interocular symmetry demonstrated in this study for FAZ and vessel density, OCTA can be particularly useful in assessing eyes with unilateral ocular disease. Because the agreement in interocular vessel densities was better when using VLD instead of VAD, VLD may be a preferable measure of vessel density to the VAD. Furthermore, as discussed by Iafe et al,¹⁵ measurement of VLD is more important to assess pathologic features, because the finer vessels are often affected first and are the site of oxygen exchange and tissue perfusion. Thus, it is possible that VLD, rather than VAD, may prove more sensitive in early or mild disease states.

In conclusion, we used OCTA to measure the FAZ and vessel densities in normal maculae of healthy pediatric patients across ages from birth to 17 years. These data may provide a basis to assess foveal development and pathophysiologic features. These baseline measurements of healthy eyes in patients as young as 9 weeks not only provide a reference with which to compare pathologic features in infants and children but also, in the future, may help the clinician to detect, diagnose, or monitor disease.

Abbreviations and Acronyms:

AL

axial length

CSFT

central subfield thickness

DVC

deep vascular complex

FAZ

foveal avascular zone

ICC

intraclass correlation coefficient

OCTA

OCT angiography

SVC

superficial vascular complex

VAD

vessel area density

VLD

vessel length density