Large Hypertransmission Defects Exhibit Choriocapillaris Flow Speed Impairment in Nonexudative Age-Related Macular Degeneration

Abstract

Purpose

To investigate choriocapillaris (CC) blood flow speed in regions associated with hypertransmission defects (hyperTDs) in nonexudative age-related macular degeneration (AMD) using variable interscan time analysis (VISTA) OCT angiography (OCTA).

Design

Retrospective cross-sectional analysis of a prospectively collected cohort.

Subjects

Thirty-one eyes from 29 subjects with nonexudative AMD.

Methods

Patients with age-related macular degeneration were imaged using a 600 kHz A-scan rate prototype swept-source OCT with a 5 × 5 mm field of view and 5 B-scan repeats (1.25 ms interscan time). Hypertransmission defects were traced on choroidal en face projections and categorized by their greatest linear dimension (GLD): large (≥250 μm), medium (63–250 μm), and small (<63 μm). Choriocapillaris blood flow speed was quantified using VISTA, which measures OCTA signal saturation dynamics across multiple interscan times. Variable interscan time analysis flow speed (VFS) was evaluated at the macula and within hyperTDs. Choriocapillaris flow speed impairment (ΔVFS) for each hyperTD was calculated as the difference between its VFS and the macular average. To assess spatial extent, ΔVFS was assessed beyond lesion boundaries. Traditional metrics of OCTA signal and flow deficits (FDs) were also evaluated.

Main Outcome Measures

Choriocapillaris blood flow speed impairment (ΔVFS) within and around hyperTDs.

Results

The macular average CC VFS was 1.47 ± 0.34 ms^–1, with no significant difference between eyes with (n = 14) and without (n = 17) hyperTDs. A total of 88 hyperTDs were analyzed: 19 large, 28 medium, and 41 small. Large hyperTDs showed significant CC flow impairment (ΔVFS = –0.37 ± 0.18 ms^–1, P_adjusted < 0.0001), with impairment extending 100 μm beyond lesion boundaries (P_adjusted = 0.0061). Medium-sized hyperTDs demonstrated moderate impairment (ΔVFS = –0.30 ± 0.47 ms^–1, P_adjusted = 0.031), while small hyperTDs did not. In linear mixed-effects modeling, large and medium hyperTDs were associated with significant reductions in flow speed (–0.40 ms^–1, P = 0.014; –0.31 ms^–1, P = 0.030, respectively), corresponding to approximately 25% decreases from macular average. OCT angiography signal and FD metrics also detected size-dependent flow impairment.

Conclusions

Large hyperTDs in nonexudative AMD exhibit reduced CC flow speed extending beyond the lesion boundary. Longitudinal studies will investigate whether CC flow predicts onset and progression of hyperTDs.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Geographic atrophy (GA) represents the late stage of nonexudative age-related macular degeneration (AMD) and involves progressive and irreversible vision loss. With the advent of OCT, precursor markers of GA have been extensively studied.^1,2 These include nascent GA, incomplete retinal pigment epithelial and outer retinal atrophy, complete retinal pigment epithelial and outer retinal atrophy, and hypertransmission defects (hyperTDs), all of which have been shown to predict GA progression.3, 4, 5 Among these, hyperTDs stand out due to their ease of identification.

Hypertransmission defects are hyperreflective features on en face OCT projection images of the choroid.⁶ They arise from increased light penetration into the choroid due to retinal pigment epithelium (RPE) disruptions.^7,8 Unlike other GA precursor markers that require review of multiple OCT B-scans, hyperTDs can be identified using a single en face OCT image. This advantage, coupled with their high interreader agreement, even for small lesions, makes hyperTDs a practical marker for GA risk.^9,10

Despite extensive studies of hyperTD size, persistence, and prevalence, the changes in the choriocapillaris (CC) associated with hyperTDs remain poorly understood. Investigating these vascular changes is essential, because GA progression involves not only the loss of the RPE and photoreceptors, but also loss of the CC.11, 12, 13 Understanding CC alterations in the context of hyperTDs could provide insights into AMD progression and potential therapeutic targets.

OCT angiography (OCTA) has been widely utilized to investigate the CC in AMD.^14,15 However, most OCTA studies to date evaluated flow deficits (FDs), a binary marker indicating CC regions with OCTA signal below a threshold.¹⁶ While analysis using FDs provides valuable insight into CC flow impairment, its binary nature fails to capture the spectrum of CC dysfunction or quantify blood flow speed. Because changes in blood flow speed may precede complete vascular dropout, measuring the flow speed or its surrogate marker could enable earlier detection and more nuanced understanding of CC impairment.

To address the lack of flow speed information from OCTA, we developed variable interscan time analysis (VISTA) OCTA, a computational method that quantifies blood flow speed by analyzing how the OCTA signal saturates with increasing time between repeated B-scans (the interscan time).¹⁷ Unlike traditional OCTA, which primarily detects the presence or absence of flow, VISTA quantifies blood flow speed by using high-speed, swept-source OCT (SS-OCT) (≥400 kHz) to rapidly acquire multiple repeated B-scans with a short interscan time (∼1 ms). Previous VISTA studies have demonstrated the ability to detect cardiac pulsatility in individual capillaries, high repeatability in normal retinal capillaries, and flow speed variations in both the deep capillary plexus and microaneurysms in diabetic retinopathy.17, 18, 19 In the CC, VISTA has shown faster flow speed at the center of CC lobules compared with the periphery, a hemodynamic pattern consistent with the unique structure of the CC.^20,21

In this study, we apply VISTA OCTA to evaluate CC blood flow speed in regions associated with hyperTDs. Specifically, we quantify CC blood flow speed in areas under hyperTDs, as well as around the lesion margins. We categorize hyperTDs based on size, including greatest linear dimension (GLD) ≥250 μm, a category termed large hyperTDs in prior studies, which are associated with progression to GA.^6,8 In parallel, we also analyze traditional metrics of OCTA signal and FDs.

Methods

Subjects

Subjects were enrolled in a prospective, observational OCT imaging study at the New England Eye Center (Boston, MA). The study was approved by the Tufts Medical Center Institutional Review Board and the Massachusetts Institute of Technology Committee on the Use of Humans as Experimental Subjects. All participants provided written informed consent prior to imaging, and the study adhered to the tenets of the Declaration of Helsinki.

This manuscript reports a retrospective analysis of a subset of this prospectively collected cohort. Eyes imaged between August 2022 and August 2024 using a high-speed prototype SS-OCT system were included. The inclusion criteria were a diagnosis of nonexudative AMD. The exclusion criteria included exudative AMD, GA, or other retinal pathologies such as diabetic retinopathy. Nonexudative macular neovascularization was not excluded.

OCT Imaging

Imaging was performed using a prototype SS-OCT system with a microelectromechanical system vertical-cavity surface-emitting laser (Praevium Research, Inc) operating at 600 kHz A-scan rate and 1050-nm center wavelength. The field of view was 5 mm × 5 mm and each B-scan consisted of 570 A-scans. Five repeated B-scans were acquired at each of 570 retinal positions, with an interscan time of 1.25 milliseconds, which is 2 to 5 times shorter than commercial OCTA. The total acquisition time for an OCTA volume was 3.57 seconds. Scans exhibiting severe motion artifacts, defocus, or vignetting, as evaluated from retinal vasculature in en face OCTA, were excluded. Because the prototype system lacked an eye-tracking module (commonly found in commercial OCT devices), it was more susceptible to such artifacts.

HyperTD Identification

Hypertransmission defects were identified using previously described en face methods.^9,10 A choroidal en face OCT image was generated by projecting 65 to 400 μm below Bruch’s membrane. A trained reader (M.U.J.) used ImageJ (Fiji, National Institutes of Health) to identify and trace hyperTDs, confirming choroidal hypertransmission in B-scans. The GLD of each hyperTD was computationally measured from the reader tracings.

Regions Excluded from Analysis

Choriocapillaris beneath calcified drusen, hyperreflective foci, and large retinal vessels were excluded due to OCT/OCTA signal attenuation.22, 23, 24 For calcified drusen and hyperreflective foci, hyporeflective features on the choroidal en face image were used, with hyporeflectivity in the choroid posterior to these lesions confirmed on B-scans. Large retinal vessels (>50 μm in diameter) were identified from inner retina en face OCTA via a vesselness filter, a common technique for detecting vascular structures.²⁵

CC Hemodynamics Using VISTA

Variable interscan time analysis quantifies blood flow speed using OCTA saturation dynamics, as detailed in previous publications.^17,18 This method is related to techniques used in neuroscience research to assess blood flow speed.²⁶ A brief overview is provided in Figure 1.

Figure 1.

Quantification of CC blood flow speed using VISTA OCTA. A, Multiple repeated B-scans enable computation of OCTA signals using varying interscan times. For example, pairwise combinations of 5 B-scan repeats with an interscan time of Δt can be analyzed to generate OCTA at Δt, 2Δt, 3Δt, and 4Δt. B, Variable interscan time analysis compiles OCTA measurements to robustly evaluate OCTA signal saturation. Yellow boxes, shown on the CC en face OCTA image, indicate the spatial scale used for compilation (61 μm × 61 μm). C, Variable interscan time analysis measures how rapidly the OCTA signal saturates as a surrogate for blood flow speed. OCT angiography values from 2 regions are plotted. Gray dots represent individual compiled OCTA measurements. The mean OCTA at each interscan time (black open circles) is fitted to an exponential saturation model to derive the VFS. At Location X1 (from panel B), the mean OCTA signal saturates more gradually than at Location X2, resulting in a slower VFS (0.89 ms^–1 vs 2.10 ms^–1, respectively). D, The VISTA OCTA image encodes blood flow speed by mapping VFS values to a color scale. Scale bars: 500 μm. The CC shown is from the fovea of a healthy 46-year-old female subject. CC = choriocapillaris; OCTA = OCT angiography; VFS = variable interscan time analysis flow speed; VISTA = variable interscan time analysis.

OCT angiography detects blood cell motion by measuring differences between repeated OCT B-scans. The OCTA signal increases with time between scans (interscan time Δt) and then saturates. Faster blood flow speed causes the OCTA signal to saturate more rapidly with increasing interscan time.²⁷ Variable interscan time analysis measures the rate of this saturation, named VISTA flow speed (VFS), as a surrogate marker of blood flow speed.

To sample multiple interscan times, VISTA computes OCTA at Δt, 2Δt, … and (N – 1) Δt by selecting pairwise combinations from N repeated B-scans (Fig 1A). For each spatial location, OCTA measurements are then compiled within a small region (61 μm × 61 μm) to improve signal robustness, assuming that CC blood flow speed varies smoothly across space (Fig 1B). The mean OCTA signal at each interscan time is then fitted to a saturation model (Fig 1C):

$$\text{OCTA}(\Delta t)=\text{Saturated}\text{OCTA}\times \left(1–\exp (–\text{VFS}\times \Delta t)\right)\text{.}$$

Variable interscan time analysis flow speed can be mapped to color to encode blood flow speed in a single en face image (Fig 1D).

CC Hemodynamics in and around hyperTDs

Figure 2 demonstrates how VISTA can study CC hemodynamics associated with hyperTDs. The CC slab was defined starting 5.3 μm below Bruch’s membrane’s posterior surface as the anterior border with 11 μm thickness, consistent with previous guidelines.¹⁶

Figure 2.

Variable interscan time analysis OCTA visualizes and quantifies CC blood flow speed in the macula and individual hyperTDs. Top row: structural OCT images of the choroid. The left panel displays a choroidal en face projection of the whole macula, with regions-of-interest containing hyperTDs. The middle and right panels show zoom-in views of 2 hyperTDs, manually traced (yellow outline). The B-scans along the dashed line confirm hypertransmission into the choroid. Bottom row: VISTA OCTA images of the CC. The left panel shows the whole macula region, with CC blood flow speed color-coded (blue = slow, red = fast). The middle and right panels zoom into the hyperTD regions (white outlines), revealing localized reductions in CC flow speed. Conventional CC OCTA images are shown for comparison. The average macular VFS is 1.19 ms^–1; the 2 hyperTDs exhibit lower VFS values of 0.81 ms^–1 and 0.86 ms^–1. CC = choriocapillaris; hyperTD = hypertransmission defect; OCTA = OCT angiography; VFS = variable interscan time analysis flow speed; VISTA = variable interscan time analysis.

The top row shows 2 example hyperTDs with their manually traced masks, along with B-scans with clear hypertransmission into the choroid. At the bottom row, false-color VISTA OCTA images encode blood flow speed in the CC; fast flow is shown in red, and slow flow in blue.²⁸ Compared with conventional CC OCTA, which offers limited hemodynamic information, the VISTA images clearly reveal slower flow (blue hue) under both hyperTDs.

Beyond qualitative visualization, VISTA also provides quantification of blood flow speed. Below the VISTA OCTA images, the measured VFS is shown for both the macula and the individual hyperTD regions. In this example, the 2 hyperTDs exhibit VFS of 0.81 ms^–1 and 0.86 ms^–1, while the macular average is 1.19 ms^–1. These reductions are consistent with the blue hue seen in the VISTA images. To quantify this local flow speed impairment while reducing effects from intersubject variation in macular average VFS, we compared the CC blood flow speed in a hyperTD to the macular average using a difference, ΔVFS = VFS_HyperTD - VFS_Macula. To evaluate the spatial extent of CC flow alterations, ΔVFS was also analyzed in 3 zones around hyperTDs: 0 to 100 μm, 100 to 200 μm, and 200 to 300 μm. Because OCT and OCTA are spatially coregistered, the traced hyperTD masks can be used to study blood flow speed. For lesions and their perilesional rings located at the scan borders, only the portions within the field of view were considered. The macular VFS was calculated within a 5-mm diameter circle centered on the fovea. We note that hyperTD areas were excluded from macular average VFS calculations to minimize potential lesion-related bias.

CC Flow Impairment Using Traditional OCTA and FDs

While VISTA offers quantification of blood flow speed using OCTA signals at multiple interscan time points, its implementation can be computationally intensive and is currently limited to high-speed OCT systems.²⁹ In contrast, conventional OCTA metrics obtained at a fixed interscan time are readily available on commercial devices.

To assess whether these conventional metrics can detect CC impairment associated with hyperTDs, we analyzed 2 commonly used markers from prior studies: (1) normalized OCTA signal and (2) FD percentage, calculated as the area of FDs relative to the total area, using an intercapillary distance threshold (24 μm) suggested by Zhang et al.³⁰ For both metrics, local perfusion impairment was assessed by comparing values within each hyperTD to the macular average to determine the normalized OCTA signal difference (ΔOCTA) and FD difference (ΔFD), where ΔOCTA = OCTA_HyperTD - OCTA_Macula and ΔFD = FD_HyperTD - FD_Macula, consistent with the ΔVFS approach. The normalized OCTA signal at interscan time Δt was calculated using the following formula¹⁴:

$$\text{OCTA}(\Delta \mathrm{t})={\left[\text{OCT}(\mathrm{t}+\Delta \mathrm{t})–\text{OCT}\left(\mathrm{t}\right)\right]}^2/\left[\text{OCT}{(\mathrm{t}+\Delta \mathrm{t})}^2+\text{OCT}{\left(\mathrm{t}\right)}^2\right]$$

Here, OCT(t) and OCT(t + Δt) refer to the OCT signal amplitude measured at different time points and the squared difference in OCT signal at different times is normalized by the OCT signal square. OCT signal is proportional to the square root of light intensity. We use split-spectrum amplitude decorrelation angiography to reduce noise from axial eye motion.³¹ Detailed processing parameters are described in our previous publication.¹⁷ Because the OCTA metric is normalized by the underlying OCT signal intensity, it inherently compensates for local variations in reflectivity, which can be caused by focal lesions such as drusen, assuming that the base OCT signal is substantially above the noise level. We note that multiple OCTA algorithms have been developed and used in different commercial instruments and by different research groups.³² For example, optical microangiography applies postacquisition compensation for signal attenuation caused by drusen.^33,34

Categories of HyperTD

Hypertransmission defects were categorized in 3 groups based on: GLD ≥250 μm, GLD between 63 and 250 μm, and GLD <63 μm. The 250-μm criterion is adopted as a size threshold for persistent or large hyperTDs.^6,8 The 63-μm threshold was chosen according to the criteria for “small” drusen.³⁵ While this threshold was chosen to stratify hyperTDs, we note that 63 μm is an arbitrary reference for hyperTDs, because drusen and hyperTDs are distinct entities.

Statistical Analyses

Group Comparison

Choriocapillaris blood flow speeds in eyes with and without hyperTDs were compared using the Mann–Whitney U test.

Linear Mixed-Effects Model for Covariate Analysis

To evaluate the association between CC flow speed impairment (ΔVFS) and lesion- or subject-level variables, we used a linear mixed-effects model implemented in MATLAB (2020a, MathWorks, fitglme) with the following formula:

$$\Delta \text{VFS}\sim 1+\text{VFS}\_\left(\text{macula}\right)+\text{Age}+\text{GLD}\text{Classification}+(1|\text{Patient}\_\text{ID})$$

VFS_macula, age, and GLD classification were treated as fixed effects. Patient_ID, the anonymized subject number, was included as a random intercept to account for multiple lesions from the same subject. The random-effects covariance structure assumed independent and identically distributed random intercepts with normally distributed residuals. The model was fitted using the Laplace approximation to maximum likelihood. A P < 0.05 was considered statistically significant.

Random-Intercept Model with Multiple Comparisons

To test whether local changes in VFS or OCTA metrics (ΔVFS, ΔOCTA, and ΔFD) were significantly different from zero under various conditions (3 lesion-size groups × 4 distance zones or interscan times), we used a simpler random-intercept model:

$$\Delta \text{Metric}\sim 1+(1|\text{Patient}\_\text{ID})$$

This model accounts for multiple lesions from the same subject, but does not include additional fixed covariates. In total, 36 such comparisons were performed (12 for each metric). To control for multiple comparisons, P values were adjusted using the Benjamini–Hochberg false discovery rate correction (m = 36). Both raw and adjusted P values are reported. A P_adjusted < 0.05 was considered statistically significant.

Results

Participants

A total of 31 eyes with nonexudative AMD from 29 subjects were analyzed (age, mean ± standard deviation [SD], 75.6 ± 8.1 years; 14 [48%] female subjects). The inclusion and exclusion process is illustrated in Figure S1 (available at www.ophthalmologyscience.org). Table 1 summarizes the prevalence of hyperTDs in eyes. Thirteen eyes (42%) had ≥1 hyperTD with GLD >63 μm. Five eyes (16%) had ≥1 large hyperTD (GLD ≥250 μm).

Table 1.

Subject Demographics

Subjects (N)	Age (Yrs)		Female (N/%)	Eyes (N)	Eyes with HyperTD (N/%)
	Mean	SD			Any HyperTD	GLD >63 μm	GLD ≥250 μm
29	75.6	8.1	14 (48%)	31	14 (45%)	13 (42%)	5 (16%)

GLD = greatest linear dimension; hyperTD = hypertransmission defect; SD = standard deviation.

HyperTD GLDs

A total of 88 hyperTDs were identified. The number of hyperTDs in each category was: 41 (GLD <63 μm), 28 (GLD between 63 and 250 μm), and 19 (GLD ≥250 μm) as shown in Table 2. The mean ± SD of GLD was 137 ± 130 μm. For each size category, the mean GLD was 40.2 ± 12.5 μm (GLD <63 μm), 135 ± 58.6 μm (GLD between 63 and 250 μm), and 350 ± 87.3 μm (GLD ≥250 μm). Figure S2 (available at www.ophthalmologyscience.org) shows the frequency distribution of the hyperTD GLD.

Table 2.

Hypertransmission Defects and Their GLDs

HyperTD Category Based on GLD	Number of HyperTD (%)
All GLD	88 (100%)
GLD <63 μm	41 (47%)
GLD between 63 μm and 250 μm	28 (32%)
GLD ≥250 μm	19 (22%)

GLD = greatest linear dimension; hyperTD = hypertransmission defect.

CC VFS in the Macula and HyperTDs

The mean ± SD VISTA flow speed of macular CC (5-mm diameter circle centered on the fovea) was 1.47 ± 0.34 ms^–1. This was 1.48 ± 0.24 ms^–1 for eyes without any hyperTDs, and 1.47 ± 0.44 ms^–1 for those with ≥1 hyperTD. There was no significant difference between the 2 groups (Mann–Whitney U test, P = 0.83, Figure 3, left). Figure 3 (right) also shows a scatter plot of macular VFS versus the largest hyperTD GLD per eye.

Figure 3.

Choriocapillaris VFS at the macula. A, Macular CC VFS in eyes with and without hyperTDs. B, Scatter plot of macular VFS versus largest hyperTD size (GLD) per eye. CC = choriocapillaris; GLD = greatest linear dimension; hyperTD = hypertransmission defect; VFS = variable interscan time analysis flow speed.

The mean ± SD VISTA flow speed of hyperTDs was 1.21 ± 0.69 ms^–1. For each size category, the mean VFS was 1.40 ± 0.90 ms^–1, 1.12 ± 0.42 ms^–1, and 0.91 ± 0.21 ms^–1, for GLD <63 μm, GLD between 63 and 250 μm, and GLD ≥250 μm, respectively. Additional data of hyperTD VFS versus GLD and VFS distribution are presented in Figure S3 (available at www.ophthalmologyscience.org).

Local Blood Flow Speed Impairment in HyperTDs

The local blood flow speed impairment in hyperTDs was quantified by ΔVFS = VFS_HyperTD - VFS_Macula, where a negative value indicates reduced CC flow speed within the hyperTD relative to the macula. Figure 4 shows scatter plots of ΔVFS versus GLD in each size category. All hyperTDs with GLD ≥250 μm have ΔVFS <0, showing local flow speed impairment in the lesions.

Figure 4.

Local blood flow speed impairment of hyperTDs vs their GLDs. Flow impairment is quantified by ΔVFS = VFS_HyperTD– VFS_Macula, where a negative ΔVFS (below the dotted line at 0) indicates that the hyperTD has slower CC blood flow speed than the macular average. The plots show hyperTDs grouped by GLD: A (GLD <63 μm), B (GLD 63–250 μm), and C (GLD ≥250 μm). CC = choriocapillaris; GLD = greatest linear dimension; hyperTD = hypertransmission defect; VFS = variable interscan time analysis flow speed.

The mean ± SD of ΔVFS was 0.01 ± 0.72 ms^–1, –0.30 ± 0.47 ms^–1, and –0.37 ± 0.18 ms^–1 for GLD <63 μm, GLD between 63 and 250 μm, and GLD ≥250 μm, respectively. The flow impairment was not significant for GLD <63 μm (P = 0.94, P_adjusted = 0.94), but was significant for GLD between 63 and 250 μm (P = 0.015, P_adjusted = 0.031) and GLD ≥250 μm (P < 1e-4, P_adjusted < 1e-4; adjusted with Benjamini–Hochberg correction with m = 36).

Table 3 summarizes fixed effects on ΔVFS and their statistics, from a linear mixed-effects model analysis that includes all 88 hyperTDs. Hypertransmission defects with GLD between 63 and 250 μm were associated with a significant flow speed impairment compared with those with GLD <63 μm (fixed-effect estimate: –0.31 ms^–1, P = 0.030). Hypertransmission defects with GLD ≥250 μm were associated with greater flow speed impairment (fixed-effect estimate: –0.40 ms^–1, P = 0.014). Considering the mean VFS_macula of 1.47 ms^–1, large hyperTDs correspond to an approximately 27% decrease in blood flow speed, and medium hyperTDs to an approximately 21% decrease; VFS_macula was not significantly associated with the local flow impairment.

Table 3.

Fixed Effects of Variables on Local Blood Flow Speed Impairment (ΔVFS)

Variables	Estimate	P	95% CI
			Lower	Upper
VFSMacula	–0.13	0.51	–0.53	0.30
Age	–0.003 ms–1/yr	0.80	–0.030 ms–1/yr	0.023 ms–1/yr
GLD ≥250 μm	–0.40 ms–1	0.014	–0.73 ms–1	–0.084 ms–1
GLD between 63 and 250 μm	–0.31 ms–1	0.030	–0.59 ms–1	–0.031 ms–1
Intercept	0.45 ms–1	0.71	–1.96 ms–1	2.86 ms–1

CI = confidence interval; GLD = greatest linear dimension; VFS = variable interscan time analysis flow speed.

A linear mixed-effects model was used to account for multiple hypertransmission defects within the same eye and intereye correlations. Values in bold indicate P < 0.05.

Spatial Extent of CC Flow Impairment

Figure 5A illustrates how VFS was evaluated in regions surrounding the hyperTDs. Three 100 μm-wide zones were defined based on the traced masks. Figure 5B shows the local flow impairment ΔVFS inside the hyperTDs and at increasing distances around the lesions. For GLD <63 μm, ΔVFS was not significant at all distances as well as inside the hyperTD. Likewise for GLD between 63 and 250 μm, ΔVFS was not statistically significant beyond the lesion boundaries (0–100 μm, P = 0.047, P_adjusted = 0.080; 100–200 μm and 200–300 μm, P > 0.05). For GLD ≥250 μm, significant flow impairment was observed within 0 to 100 μm around the lesion (P = 0.0017, P_adjusted = 0.0061). At further distances, no significant flow impairment was observed. These results indicate that the spatial extent of CC flow impairment may extend beyond the lesion boundaries for large hyperTDs.

Figure 5.

Spatial extent of local blood flow speed impairment around hyperTDs. A, Schematic showing how regions are determined around hyperTDs. Three adjacent 100 μm wide zones are computationally identified from the traced hyperTD masks. B, Local blood flow speed impairment (ΔVFS) within and around hyperTDs. The plots represent hyperTDs categorized by size: top: GLD ≥250 μm, middle: GLD 63–250 μm, and bottom: GLD <63 μm. Error bars indicate standard deviation. Large hyperTDs (GLD ≥250 μm) exhibit significant and persistent local blood flow speed impairment, extending into the surrounding regions. GLD = greatest linear dimension; hyperTD = hypertransmission defect; VFS = variable interscan time analysis flow speed.

Traditional OCTA Metrics

Figure 6 shows results using traditional OCTA metrics: normalized OCTA and CC FD percentage. As with the VFS analysis, the difference between each hyperTD’s value and the macular average was analyzed. For OCTA signal, negative values indicate reduced perfusion compared with the macula. For FD, positive values indicate greater CC impairment.

Figure 6.

Traditional OCTA metrics for evaluating CC flow impairment in hyperTDs, shown as box plots. A, Difference in normalized OCTA signal, computed as OCTA_HyperTD – OCTA_Macula. B, Difference in CC FD, using a binarization threshold of 0.05. Positive values in FD indicate greater CC impairment in the hyperTD. Data are shown for different interscan times. Blue dots indicate the mean; red lines, the median; and × markers, outliers. CC = choriocapillaris; FD = flow deficit; hyperTD = hypertransmission defect; OCTA = OCT angiography.

All large hyperTDs (GLD ≥250 μm) consistently showed lower normalized OCTA signal than the macula across all interscan times (Fig 6A). Medium-sized hyperTDs (63–250 μm) also showed reduced OCTA signal in most cases. These decreases were statistically significant across all interscan times for both large- and medium-sized hyperTDs. In contrast, small hyperTDs (<63 μm) did not exhibit significant reductions in normalized OCTA signal at any time point.

The FD results followed a similar pattern: large- and medium-sized hyperTDs showed higher FD values than the macular average, whereas small hyperTDs did not show significant differences (Fig 6B). Tables S1 and Tables S2 (available at www.ophthalmologyscience.org) report the mean, SD, confidence intervals, and statistical significance for analyses using normalized OCTA signal and FD, respectively. Figures S4 (available at www.ophthalmologyscience.org) shows OCTA metrics without comparison to the macular average.

Discussion

This study demonstrates that hyperTDs in nonexudative AMD correlate with locally reduced CC blood flow speed measured by VISTA OCTA. Large hyperTDs with GLD ≥250 μm exhibited a significant decrease in CC flow speed compared with the macular average, and this decrease extended beyond the lesion boundary. In contrast, small hyperTDs (GLD <63 μm) showed no significant flow speed impairment, suggesting that the degree of vascular disruption may scale with lesion size. Because hyperTDs arise from RPE disruption, these findings suggest that RPE health and CC hemodynamics are locally correlated, but this cross-sectional study cannot determine the temporal sequence of events—whether decreased CC perfusion precedes the formation of a large hyperTD or formation of a large hyperTD precedes decreased CC perfusion.

Histological studies have long demonstrated CC loss in AMD as well as its link with RPE atrophy.36, 37, 38, 39, 40, 41 Techniques such as Ulex europaeus agglutinin staining of endothelial cells and light microscopy of vascular lumens have revealed patterns of CC loss, including “ghost vessels” that lack endothelial cells, but retain lumen structures.⁴² While these postmortem methods offer high-resolution structural insight, they cannot capture functional changes of CC in vivo and macular changes in CC perfusion that are a consequence of aging.⁴³ As a result, CC dysfunction has historically been inferred from structural loss, rather than directly measured.

The advent of SS-OCT at 1050 nm has enabled in vivo CC studies. The longer wavelength of SS-OCT enables deeper penetration below the RPE compared with spectral-domain OCT at 840 nm, and its higher A-scan rate improves OCTA quality.⁴⁴ Among these studies, FDs have been widely used to quantify CC flow impairment. Representing regions with OCTA signal below a defined threshold, FD has been shown to predict GA progression and has been locally associated with nascent GA development.45, 46, 47, 48, 49, 50 Traditional FD analyses are limited by their binary nature and the lack of blood flow speed information, making it difficult to detect hemodynamic alterations in the CC. However, the changes in CC FDs shown in Figure 6 closely parallels the changes observed in the VFS analyses, suggesting that changes in CC FDs could serve as a surrogate for VFS assessment if sufficiently short interscan times are used.

We hypothesize that CC dysfunction develops along a continuum, beginning with reduced blood flow speed before progressing to structural dropout, but it would appear that the structural changes precede the reduction in blood flow given the hyperTD size dependence of the decreased VFS analyses. In addition, there was no difference in the macular VFS in eyes with and without hyperTDs suggesting that the initial cause of the localized decreased VFS was secondary to the localized anatomic changes in the retina/RPE complex and not a global decrease in ocular perfusion beyond the expected age-dependent decrease in ocular perfusion.^43,51

Variable interscan time analysis expands the scope of CC analysis by capturing quantitative flow speed information from OCTA saturation dynamics. In doing so, it overcomes the limitations of traditional OCTA, which focuses mainly on presence-versus-absence of detectable flow. Beyond advancing our understanding of AMD pathophysiology, this capability may help identify hyperTDs at greater risk of expansion, which is particularly important given their high variability in growth rates.⁵²

To the best of our knowledge, this is the first study to quantify changes in the CC blood flow speed associated with disease progression in AMD, offering a novel perspective on vascular changes in the disease. Hemodynamic studies of the CC have been limited, with most methods being either invasive or lacking depth resolution. For example, tracking fluorescence-labeled red blood cells, as demonstrated in albino rats, is not feasible in humans.^53,54 Noninvasive modalities such as pneumotonometry, laser Doppler flowmetry, and laser speckle flowgraphy can measure choroidal flow, but cannot isolate CC hemodynamics.⁵⁵ Adaptive optics studies have provided detailed measurements of blood flow velocity in a range of retinal vasculature, from capillaries to arterioles.^56,57 However, adaptive optics studies of CC blood flow are limited because the CC is obscured by the RPE.

The VISTA OCTA protocol scans the majority of the macula, enabling macular blood flow speed to serve as a patient-specific reference. This is valuable because a large intersubject variance in CC blood flow speed is expected; previous studies on ophthalmic artery flow, which perfuses the choroid and subsequently the CC, have reported significant variance among healthy subjects.^58,59 Our method of analysis, comparing hyperTD flow speed against the patient’s own macular average, is limited in that it does not account for overall variations in macular blood flow speed, which may associate with hyperTD progression. However, it may reduce the effects of intersubject differences, enhancing the detection of localized flow speed disruptions.

In prior work, we provided technical recommendations for performing VISTA on the retinal vasculature.¹⁷ Similar principles apply to the CC; VISTA requires a short interscan time to capture OCTA saturation. Given the average macular CC VFS (∼1.5 ms^–1), an interscan time under ∼2 ms is recommended. At typical scan parameters (e.g., 10 μm spacing, 75% scanner duty cycle, 3 mm × 3 mm field of view), this corresponds to an A-scan rate of approximately 200 kHz. While this exceeds the capabilities of most current commercial systems, several next-generation wide-field SS-OCT devices offer A-scan rates ≥200 kHz. Therefore, CC VISTA imaging should be feasible using standard 3 mm × 3 mm protocols on upcoming commercial OCTA platforms.

Traditional metrics of normalized OCTA signal and FD also revealed local flow impairment in hyperTDs with GLD >63 μm, consistent with the VFS findings (Fig 6). When all 88 hyperTDs were analyzed, the fixed effect of being a large hyperTD (GLD ≥250 μm) was statistically significant for all interscan times using normalized OCTA signals (Fig S5, available at www.ophthalmologyscience.org), similar to the VFS results in Table 3. While the fixed-effect estimate decreased with increasing interscan time, the reduction was less pronounced for normalized OCTA than for FDs. This difference may stem from the fact that the normalized OCTA signal retains continuous signal variations related to blood flow, whereas FDs involve binarization, which may decrease sensitivity to subtle changes.^60,61

At shorter interscan times, the OCTA signal is sensitive to blood flow speed, because the signal has not yet reached saturation.⁶² At longer interscan times, OCTA becomes saturated and is primarily determined by the amount of moving red blood cells rather than their speed. Therefore, even without the full VISTA framework, short interscan time OCTA may still detect flow impairments that arise from changes in flow speed. Collectively, these findings suggest that traditional OCTA signals with shorter interscan times may be more sensitive to flow-related changes than those with longer interscan times, as supported by previous studies.^60,61,63,64

Several limitations of this study beyond the limited sample size should be acknowledged. First, the cross-sectional design prevents studying the causality between CC flow impairment and hyperTD enlargement. Longitudinal studies are necessary to investigate whether reduced CC flow precedes or follows hyperTD expansion. Additionally, the method of local analysis, comparing hyperTD flow speed against the macular average, does not account for intereye variation in macular blood flow speed, which may itself be associated with hyperTD progression. We have also not assessed the repeatability of VISTA and VFS measurements over small regions that are the size of hyperTDs. Larger studies that characterize CC flow speed in normal aging and AMD, as well as evaluate VISTA repeatability across different types of AMD lesions, are needed. Our study used a prototype SS-OCT system operating at a 600 kHz A-scan rate, which is not yet widely available. However, as discussed, CC VISTA imaging should be feasible with a 2-ms interscan time, which can be achieved with a 200 kHz A-scan rate.

Although previous studies have demonstrated high interreader agreement for hyperTD identification across various sizes, this study did not perform a separate interreader agreement analysis, which is a clear limitation of our approach.^9,10 Nonetheless, our study used a dense OCT scan pattern, which should provide the same or higher level of detail compared with prior studies. In addition, the 63 μm threshold, which is adopted from the small drusen criterion, remains an arbitrary size threshold for hyperTDs. In a post hoc analysis where GLD was modeled as a continuous variable, the fixed effect of GLD was not statistically significant (Table S3, available at www.ophthalmologyscience.org). Moreover, previous studies have reported that CC perfusion decreases with increasing eccentricity (distance from the fovea).^43,65 In our post hoc analysis including eccentricity as a covariate, eccentricity did not show a statistically significant association with CC blood flow speed impairment in hyperTDs (Table S4, available at www.ophthalmologyscience.org). Nonetheless, the eccentricity of a hyperTD lesion may still influence its progression. It is notable that small hyperTDs showed high variance in VFS measurements. This variability may reflect both physiological heterogeneity of CC flow speed (less evident at larger scales due to spatial averaging) and the inherently higher susceptibility of small regions to noise because they include fewer data points.

Further studies on smaller hyperTDs are needed to determine the size threshold that significantly correlates with local CC flow impairment. A curiosity in the data is the apparent plateau in VFS once the lesions exceed about 400 μm (Fig S6, available at www.ophthalmologyscience.org) suggesting that the signal arises from the larger choroidal vessels in Sattler’s layer and not the CC. Therefore, future studies may consider placing an upper size limit on large hyperTDs, as extremely large lesions may represent a different stage of choroidal vasculature changes. Furthermore, stratifying lesions according to the degree of outer retinal atrophy observed in structural B-scans may allow for a more detailed understanding of the relationship between retinal structural changes and CC flow alterations. For example, Figure S7 (available at www.ophthalmologyscience.org) shows a hyperTD without complete RPE atrophy from the same eye presented in Figure 2.

Finally, our traditional OCTA metric analysis used only normalized OCTA and split-spectrum amplitude decorrelation. While widely adopted, other methods such as optical microangiography apply different artifact correction strategies, which are especially relevant in CC imaging. Our findings should be interpreted within the context of the algorithm used.^66,67 It is likely that different OCTA analysis methods will have different sensitivity for assessing alterations in CC versus retinal vasculature. Therefore, we believe it is important to store the repeated OCT B-scans used to calculate OCTA to enable retrospective analysis using different algorithms.

In conclusion, this study demonstrates a local correlation between RPE atrophy and CC hemodynamic impairment in AMD by showing that large hyperTDs are associated with reduced CC flow speed. By using VISTA to quantify these hemodynamic alterations, SS-OCT provides a platform to study the continuum of CC dysfunction, which we hypothesize precedes complete CC loss. Longitudinal studies using VISTA may offer insight into the factors influencing hyperTD persistence or resolution, improve patient selection for interventional trials, and guide the development of therapies aimed at preserving or restoring CC perfusion in AMD.

Manuscript no. XOPS-D-25-00612.