October 2023
Volume 64, Issue 13
Open Access
Retina  |   October 2023
Variability in Capillary Perfusion Is Increased in Regions of Retinal Ischemia Due to Branch Retinal Vein Occlusion
Author Affiliations & Notes
  • Martin Hein
    Lions Eye Institute, Nedlands, Western Australia, Australia
    Centre for Ophthalmology and Visual Science, University of Western Australia, Perth, Australia
  • Andrew Mehnert
    Lions Eye Institute, Nedlands, Western Australia, Australia
    Centre for Ophthalmology and Visual Science, University of Western Australia, Perth, Australia
  • K. Bailey Freund
    Vitreous Retina Macula Consultants of New York, New York, New York, United States
    Department of Ophthalmology, New York University Grossman School of Medicine, New York, New York, United States
  • Dao-Yi Yu
    Lions Eye Institute, Nedlands, Western Australia, Australia
    Centre for Ophthalmology and Visual Science, University of Western Australia, Perth, Australia
  • Chandrakumar Balaratnasingam
    Lions Eye Institute, Nedlands, Western Australia, Australia
    Centre for Ophthalmology and Visual Science, University of Western Australia, Perth, Australia
    Department of Ophthalmology, Sir Charles Gairdner Hospital, Western Australia, Perth, Australia
  • Correspondence: Chandrakumar Balaratnasingam, Lions Eye Institute and University of Western Australia, 2 Verdun Street, Nedlands, WA 6009, Australia; balaratnasingam@gmail.com
  • Footnotes
     MH and AM contributed equally to the work presented here and should be regarded as equivalent first authors.
Investigative Ophthalmology & Visual Science October 2023, Vol.64, 30. doi:https://doi.org/10.1167/iovs.64.13.30
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Martin Hein, Andrew Mehnert, K. Bailey Freund, Dao-Yi Yu, Chandrakumar Balaratnasingam; Variability in Capillary Perfusion Is Increased in Regions of Retinal Ischemia Due to Branch Retinal Vein Occlusion. Invest. Ophthalmol. Vis. Sci. 2023;64(13):30. https://doi.org/10.1167/iovs.64.13.30.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose: To investigate alterations in macular perfusion variability due to branch retinal vein occlusion (BRVO) using a novel approach based on optical coherence tomography angiography (OCTA) coefficient of variation (CoV) analysis.

Methods: Thirteen eyes of 13 patients with macular ischemia due to BRVO were studied. Multiple consecutive en face OCTA images were acquired. Bias field correction, spatial alignment, and normalization of intensities across the images were performed followed by pixelwise computation of standard deviation divided by the mean to generate a CoV map. Region of interest–based CoV values, derived from this map, for arterioles, venules, and the microvasculature were compared between regions with macular ischemia and control areas of the same eye. Control areas were regions of the same macula that were not affected by the BRVO and had normal retinal vascular structure as seen on multimodal imaging and normal retinal vascular density measurements as quantified using OCTA.

Results: CoV increased by a mean value of 17.6% within the microvasculature of ischemic regions compared to the control microvasculature (P < 0.0001). CoV measurements of microvasculature were consistently greater in the ischemic area of all 13 eyes compared to control. There were no differences in CoV measurements between ischemic and control areas for arterioles (P = 0.13) and venules (P = 1.0).

Conclusions: Greater variability in microvasculature perfusion occurs at sites of macular ischemia due to BRVO. We report a novel way for quantifying macular perfusion variability using OCTA. This technique may have applicability for studying the pathophysiology of other retinal vascular diseases.

Capillary blood flow is precisely coupled to the fluctuating energy demands of retinal neurons.1 Alterations in capillary perfusion are a hallmark feature of retinal vascular diseases and can precede the development of irreversible retinal structural alterations. As an example, a significant increase in retinal blood flow precedes the development of microaneurysms in experimental diabetic retinopathy (DR).2,3 Quantitative measures of blood flow can serve as a biomarker of retinal vascular disease, but the methodologic challenges associated with visualizing retinal capillary structures rapidly and noninvasively have limited the study of human capillary perfusion in the clinical setting. As such, our current knowledge regarding retinal perfusion and blood flow at the capillary level is predominantly derived from animal studies using complex custom-built imaging technology. For example, fluorescently labeled red blood cells and confocal microscopy are utilized to measure absolute retinal blood flow, and similar specialized techniques are used to measure single blood cell velocity in mice.4,5 While laser Doppler velocimetry studies have assessed blood flow rates in human retinas in the research setting for the past four decades, the technology is mostly limited to data collected from arterioles and venules with little information generated for the capillary microvasculature.6,7 More recently, technologies such as adaptive optics scanning laser ophthalmoscopy (AOSLO) have been able to visualize in vivo retinal blood flow in humans down to the capillary level with success in detecting changes due to DR.8 These techniques for measuring perfusion are all relatively challenging, are time-consuming, and require specialized equipment and skills not readily available in current clinical settings. 
Optical coherence tomography angiography (OCTA) is a major advancement in ophthalmic imaging that provides stratified visualization of retinal capillary plexuses. Our previous studies have shown that retinal capillary anatomic information detected using OCTA is comparable to the histologic ground truth.911 OCTA is a functional extension of structural OCT and uses repeated B-scans to detect motion contrast from red blood cell (RBC) movement and visualize perfused vasculature.12 The interscan time determines the slowest RBC flow speed that can be detected, and similarly, speeds above a certain threshold cannot be differentiated and the OCTA signal becomes saturated.13 Consecutive en face OCTA acquisitions, taken seconds apart, can theoretically record real-time changes in the appearance of the perfused retinal microvasculature over time.14 By extension, it is possible to quantify the spatial and temporal changes in perfused microvasculature by aligning the images and measuring the coefficient of variation (CoV) of signal intensity for each pixel.15 The major advantage of this technique for quantifying capillary perfusion is that it is rapid, safe, and noninvasive. 
Retinal ischemia is characterized by a restriction of blood flow that limits the delivery of vital nutrients and/or removal of toxic metabolites from the energy-intensive retina. As the retinal vasculature lacks autonomic innervation, local blood flow changes are autoregulated by myogenic and metabolic mechanisms.16 Vasoactive factors are released by retinal neurons, glial cells, and vascular endothelial cells during neurovascular coupling, but production of these factors may be altered during retinal ischemia.1,17 Ischemic retinopathies are a major cause of severe vison loss in adults and children.1820 The relationship between retinal ischemia and capillary perfusion is difficult to investigate because a myriad of physiologic and disease variables can modulate retinal blood flow. Such factors include age, sex, blood pressure, smoking, caffeine intake, and systemic vascular disease.21 As such, in conditions characterized by retinal ischemia, it has been difficult to discern if changes in capillary perfusion are due to the pathogenic effects of ischemia alone or to the influence of confounding physiologic and systemic variables that modulate retinal blood flow. The purpose of this report is to investigate microvasculature perfusion changes in eyes with BRVO that involve the macula. Retinal perfusion variability is quantified using CoV analysis of signal intensity in consecutive OCTA images. By using the uninvolved, contralateral nonischemic portion of the same macula as an internal control, we negate the influence of physiologic variables on retinal blood flow and isolate the influence of ischemia on capillary perfusion. This study provides new insights into the pathophysiology of retinal ischemia perfusion and serves as an exemplar experimental model for studying other ischemic retinopathies where early perfusion abnormalities have been implicated. 
Methods
Participants
This study was approved by the Human Research Ethics Committee of the University of Western Australia and has been carried out in accordance with the Code of Ethics and Ethical principles for medical research involving human participants of the World Medical Association (Declaration of Helsinki). Written consent was attained from all participants prior to enrollment in the study. 
Thirteen eyes with BRVO from 13 patients were recruited from the Lions Eye Institute in Perth, Western Australia, between March 2019 and July 2022. Eligible patients were at least 18 years of age with BRVO and macular ischemia diagnosed on clinical assessment and multimodal imaging. Exclusion criteria included (1) any concurrent or previous ocular disease that manifested retinal vascular abnormalities, (2) any previous ocular/retinal surgery or laser procedure, (3) poor-quality retinal imaging that precluded analysis of retinal vascular detail, (4) media opacity, (5) patients with diabetes mellitus (DM), or (6) presence of macular edema that may distort segmentation reliability. Patients receiving intravitreal anti–vascular endothelial growth factor (VEGF) therapy were included, but imaging was not performed within 6 weeks of treatment administration. Demographic and clinical information was obtained for each participant. Patients were instructed to avoid caffeine the day of testing. 
Retinal Multimodal Imaging
All patients underwent contemporaneous multimodal retinal imaging at the same visit, including fundus photography, fluorescein angiography (FA), and OCTA. 
  • 1. Fundus Photography
  • Standard retinal photographs of the macula, posterior pole, and optic disk were captured using the Canon CX-1 digital retinal camera (Retinal Imaging Control Software for CX-1, 4.6.0.5; Canon Medical Systems, Otawara, Japan). A BRVO was confirmed on color photography by venous compression at an arteriovenous crossing point with dilatation of retinal veins and retinal hemorrhages proximal to the point of compression.
  • 2. Fluorescein Angiography
  • Angiogram sequences were captured using the Heidelberg Spectralis OCT2 Module using the 30° and 55° lenses (Heidelberg Eye Explorer, 1.10.4.0; Heidelberg Engineering, Heidelberg, Germany). Multiple frames of the first 5 minutes were captured. A BRVO was confirmed on FA by a delay in arteriovenous transit time within the circulatory bed beyond the site of occlusion. FA of the macula was examined by a retina specialist (CB) to confirm the presence of macular ischemia.22 Eyes that did not demonstrate macular ischemia on the FA were not included in this study.
OCTA Image Acquisition
OCTA was acquired for each participant with the Optovue RTVue XR Avanti (Optovue, Inc., Freemont, CA, with AngioVue version 2018.1.0.43), which uses the split spectrum amplitude decorrelation angiography (SSADA) algorithm. This instrument has an A-scan rate of 70,000 scans per second using a light source centered on 840 nm and a bandwidth of 45 nm. Each OCTA volume contains 304 × 304 A-scans with two consecutive B-scans captured at each fixed position before proceeding to the next sampling location. The total acquisition time for a single volume was approximately 3 seconds. The system employs two strategies to reduce motion artifacts arising from eye blinks, saccades, and fixation changes. The first is real-time eye tracking, permitting the reacquisition of a portion of the OCTA volume when motion is detected. The second is to acquire a pair of consecutive OCTA volumes with B-scans orthogonal to one another and to spatially coregister and combine the two volumes as a software postprocessing step. The instrument software automatically segments the different retinal boundaries in the merged volume and generates a two-dimensional en face projection through the retinal thickness being evaluated. The resulting image, hereinafter called an OCT angiogram, can then be exported as a raw file. 
The scan area was 3 × 3 mm centered on the fovea. A total of 10 to 15 sequential OCT angiograms were acquired for each study eye over a maximum period of 5 minutes. Our preliminary data have shown that spatiotemporal variations in capillary perfusion can be reliably detected using this number of angiograms.14 For each volume, the OCT angiogram for the full retinal slab was exported as a raw file and the corresponding en face OCT projection exported as a PNG file. 
To ensure validity of quantitative measures, OCT angiograms were individually inspected and excluded based on the following criteria: (1) if the image demonstrated motion artifact in the form of doubling of vascular structures or sideways shearing, (2) OCTA images with a quality rating below 8, (3) if the center point was significantly deviated from the fovea, and (4) errors in automatic layer segmentation. After exclusion of inadequate scans, all eyes had more than eight volumes that could be used for analysis. 
Method for Quantifying Capillary Perfusion
The creation of a single OCTA volume involves acquiring multiple B-scans of the same retinal location. The movement of RBCs between the consecutive B-scans produces differences in reflectance patterns. Larger differences yield brighter voxels in the OCTA volume and in the OCT angiogram derived from it.23 This contrast in turn visualizes the perfused microvasculature. A sequence of consecutive OCT angiograms, taken seconds apart, can therefore record changes in the en face appearance of the perfused microvasculature over time. We have previously shown that the pixelwise CoV of intensity can be used to assess perfusion over consecutive OCTA scans.14 CoV is a dimensionless measure of the spread or dispersion of data about its mean, defined as the standard deviation divided by the mean (generally expressed as a percentage). It permits the comparison of the variation from one data series to another, even where means are very different. For example, this permits a comparison of the variation in OCTA values between arterioles, venules, and capillaries or between an ischemic region and a control region where we might expect the mean values to be different. It also permits comparison between patients/participants. Warner et al.24 similarly used CoV to compare RBC flow, measured using an adaptive optics scanning laser ophthalmoscope, between human retinal arterioles, venules, and capillaries. A nearly unbiased estimate of the population CoV can be obtained by multiplying the sample CoV by (1 + 1/4n), where n is the number of observations.25 
By computing the CoV for each pixel from a sequence of aligned OCT angiograms, one obtains a CoV map. Individual CoV values in the map provide information about temporal intensity variation, and sets of CoV values within a region of interest (ROI) additionally provide information about spatial variation. ROI-based statistics therefore provide quantitative information about spatiotemporal variation of perfusion. 
The CoV estimate at a given pixel depends on both the quality of the images acquired by the OCT imaging system (degrading factors include speckle noise, bias field, and motion artifacts) and the accuracy of the alignment of the individual OCT angiograms.26,27 For this reason, in addition to taking steps to mitigate the influence of these factors, we perform CoV analysis on sets of pixels belonging to vessel segments rather than on individual pixels. 
Supplementary Figure S1 is a flowchart of our vessel segment–based CoV analysis pipeline. The pipeline was implemented in FIJI28 and R.29 The pipeline takes as input a sequence of OCT angiograms (as raw files) and the corresponding sequence of en face OCT projections (as PNG files), both exported from the Optovue system. It then performs the following steps, illustrated in Figure 1 (steps 1–6), Figure 2 (steps 7 and 8), and Figure 3 (steps 9 and 10): 
  • 1. Bias Field Correction of The OCT Angiograms: The angiograms are often affected by intensity inhomogeneity known as the bias field. We use the method described Zang et al.27 to correct each angiogram based on an estimate of the bias field from the corresponding OCT projection (Figs. 1A–C).
  • 2. Spatial Alignment of The OCT Angiograms: The corrected angiograms are first roughly aligned (translation-only) using corresponding local key features derived from the scale-invariant feature transform.30 The alignment is then refined using intensity-based rigid, affine, and nonlinear registration (Figs. 1A–C).31 Supplementary Figure S2 illustrates the efficacy of the spatial alignment step.
  • 3. Normalization of Intensities Across The Angiograms: This is achieved using quantile normalization (Figs. 1A–C).
  • 4. Computation of The Pixel-Wise Mean, Standard Deviation, and (Unbiased) CoV Images (Figs. 1D, 1E).
  • 5. Segmentation of Single-Pixel-Thick Vessel Centerlines From The Mean Image32,33 (Fig. 1F).
  • 6. Generation of The Vessel Centerline CoV Map: This is a heatmap visualization of the CoV values along the vessel centerlines (Fig. 1G).
  • 7. Computation of The Mean CoV For Each Vessel Segment: Individual vessel segments are identified in the centerline segmentation from step 5 (Fig. 2). A vessel segment is defined to be a sequence of pixels between junctions and endpoints. Using the pixelwise vessel centerline CoV map (Fig. 2A), each vessel segment is assigned the mean CoV value of its constituent pixels creating the vessel segment CoV map (Fig. 2B).
  • 8. Visualization of The Vessel Segment CoV Map: This is a heatmap visualization showing the mean CoV value assigned to each vessel segment (Fig. 2B).
  • 9. Interactive Selection of ROIs:  
    • a. Before ROI Selection, Three Areas Are Defined (Fig. 3):  
      • (1) A rectangular “I” area locating the region of macular ischemia
      • (2) A rectangular “C” control area opposite the region of ischemia capturing macula unaffected by the BRVO
      • (3) A “large vessel mask” encompassing all pixels belonging to large vessels (large vessels are defined as Horton–Strahler orders 2 to 4). This “large vessel mask” is later used to isolate microvasculature in the “I” and “C” areas.
    • b. Six ROIs Are Then Defined (Fig. 3):  
      • (1) “VI” ROI encompassing the large venules within “I”
      • (2) “AI” ROI encompassing the large arterioles within “I”
      • (3) “VC” ROI encompassing the large venules within “C”
      • (4) “AC” ROI encompassing the large arterioles within “C”
      • (5) “MI” ROI encompassing microvasculature within “I”
      • (6) “MC” ROI encompassing microvasculature within “C”
  • 10. Computation of The (Unbiased) CoV of Vessel Segment CoV Values For Each of The Six ROIs: A given ROI contains several vessel segments, each of which has been assigned a CoV value derived from the OCTA signal intensity change over time (step 7). The CoV of these vessel segment CoV values is computed to yield a single number for the ROI. This number quantifies how the CoV values (derived from OCTA signal intensity) vary within the ROI, an estimate of perfusion variability within the ROI.
Figure 1.
 
CoV analysis pipeline: steps 1 to 6 leading to the generation of the vessel centerline CoV map. The pipeline takes as input multiple consecutive OCT angiograms (A) and the corresponding en face OCT projection images (B). For each pair of OCTA and OCT images, bias field correction is performed on the OCTA image using an estimate of the bias field computed from the OCT image. The corrected OCTA images then undergo spatial alignment and normalization of intensity (C). Mean and SD projections of the resulting OCTA stack are generated (D and E, respectively). The vessel centerline map is created from the mean projection (F). The mean and SD projections are used to calculate the CoV for each pixel of the vessel centerline map. The resulting vessel centerline CoV map is visualized using a heatmap and (optionally) superimposed on the mean image (G).
Figure 1.
 
CoV analysis pipeline: steps 1 to 6 leading to the generation of the vessel centerline CoV map. The pipeline takes as input multiple consecutive OCT angiograms (A) and the corresponding en face OCT projection images (B). For each pair of OCTA and OCT images, bias field correction is performed on the OCTA image using an estimate of the bias field computed from the OCT image. The corrected OCTA images then undergo spatial alignment and normalization of intensity (C). Mean and SD projections of the resulting OCTA stack are generated (D and E, respectively). The vessel centerline map is created from the mean projection (F). The mean and SD projections are used to calculate the CoV for each pixel of the vessel centerline map. The resulting vessel centerline CoV map is visualized using a heatmap and (optionally) superimposed on the mean image (G).
Figure 2.
 
CoV analysis pipeline: steps 7 to 8 leading to the generation of the vessel segment CoV map. Steps 1 to 6 are detailed in Figure 1. (A) The pixelwise CoV map from Figure 1G. In this map, each individual pixel from the vessel centerline map is assigned a CoV value. (B) The vessel segment CoV map where each vessel segment is assigned a CoV value from the mean of its constituent pixel CoV values. A vessel segment is defined to be a sequence of pixels between vessel junctions and endpoints in the centerline image. The two insets demonstrate on two vessel segments how a mean CoV value (B, inset) is derived from its constituent pixel CoV values (A, inset). The scale bar describes the color representation of CoV values in panels A and B.
Figure 2.
 
CoV analysis pipeline: steps 7 to 8 leading to the generation of the vessel segment CoV map. Steps 1 to 6 are detailed in Figure 1. (A) The pixelwise CoV map from Figure 1G. In this map, each individual pixel from the vessel centerline map is assigned a CoV value. (B) The vessel segment CoV map where each vessel segment is assigned a CoV value from the mean of its constituent pixel CoV values. A vessel segment is defined to be a sequence of pixels between vessel junctions and endpoints in the centerline image. The two insets demonstrate on two vessel segments how a mean CoV value (B, inset) is derived from its constituent pixel CoV values (A, inset). The scale bar describes the color representation of CoV values in panels A and B.
Figure 3.
 
ROI selection for CoV analysis example on a single eye with BRVO. FA (A) and a mean projection of multiple consecutive aligned OCT angiograms (B) over the same macula. Areas of capillary dropout (yellow asterisks) on FA shown in panel A are correlated with areas of dropout on OCTA seen in panel B. The process of ROI selection is depicted in panel B. The ischemic “I” area (red rectangle) encompasses areas of capillary dropout and absence of fluorescence in the BRVO territory. The control “C” area (green rectangle) is a mirrored area flipped over the horizontal meridian to capture the unaffected macula. Radial arterioles (cyan “a”) and venules (magenta “v”) are defined within the ischemic and control areas (green vessel outline; B) and are segregated to create arteriole ROIs and venule ROIs within the ischemic and control area for comparison. (C, D) Magnified insets over the same area in panel B (orange dotted rectangle) and depict how an OCTA region (C) is represented as a vessel segment CoV map (D). A large vessel mask that encompasses large vessels (gray vessels, D) is used to isolate the microvasculature for analysis. This process generates six ROIs for comparison: microvasculature in control “MC” and ischemic “MI” areas, arterioles in control “AC” and ischemic “AI” areas, and venules in control “VC” and ischemic “VI” areas. The scale bar describes the color representation of CoV in panel D.
Figure 3.
 
ROI selection for CoV analysis example on a single eye with BRVO. FA (A) and a mean projection of multiple consecutive aligned OCT angiograms (B) over the same macula. Areas of capillary dropout (yellow asterisks) on FA shown in panel A are correlated with areas of dropout on OCTA seen in panel B. The process of ROI selection is depicted in panel B. The ischemic “I” area (red rectangle) encompasses areas of capillary dropout and absence of fluorescence in the BRVO territory. The control “C” area (green rectangle) is a mirrored area flipped over the horizontal meridian to capture the unaffected macula. Radial arterioles (cyan “a”) and venules (magenta “v”) are defined within the ischemic and control areas (green vessel outline; B) and are segregated to create arteriole ROIs and venule ROIs within the ischemic and control area for comparison. (C, D) Magnified insets over the same area in panel B (orange dotted rectangle) and depict how an OCTA region (C) is represented as a vessel segment CoV map (D). A large vessel mask that encompasses large vessels (gray vessels, D) is used to isolate the microvasculature for analysis. This process generates six ROIs for comparison: microvasculature in control “MC” and ischemic “MI” areas, arterioles in control “AC” and ischemic “AI” areas, and venules in control “VC” and ischemic “VI” areas. The scale bar describes the color representation of CoV in panel D.
Defining Ischemic and Control Areas for CoV Analysis
Our method for defining ischemia and selecting ROIs for quantitative analysis is shown in Figure 3. The selection of the ischemic area was first done by reviewing FA images to confirm the location of the BRVO. Ischemia was defined as sites where there was an absence of fluorescence within some of the capillary beds within the region of BRVO (Fig. 3). The opposite portion of the macula not affected by BRVO was used as an internal control. Once the site of macular ischemia was determined using FA, the same region was located on the corresponding mean OCT angiogram (produced in step 4 described in the previous section; Fig. 1D) and used to define the ischemic “I” area as a rectangle extending vertically from the midpoint of the foveal avascular zone (FAZ) to the superior or inferior half of the macula and extended horizontally to include areas of capillary dropout. The control “C” area was defined as the mirror image of the “I” area reflected over the horizontal meridian. 
While current OCTA visualizes retinal microvascular images down to the capillary level, it cannot reliably differentiate between capillaries and small arterioles or venules. Leveraging knowledge gained from vascular histology, we therefore define vessels of Horton–Strahler orders 2 to 4 to be radial arterioles or venules.11,34 A “large vessel mask” was defined to demarcate these large radial vessels. This mask was generated by simple global thresholding of the mean OCT angiogram. 
Color photography and FA images were used to distinguish arterioles from venules within the “I” area to define the “AI” ROI and “VI” ROI. The “AC” and “VC” ROIs were similarly defined within the “C” area. The set difference between “I” and “large vessel mask” was used to define the microvasculature “MI” ROI within “I.” The set difference between “C” and “large vessel mask” was used to define the microvasculature “MC” ROI within “C.” 
The resulting set of six ROIs (“AC,” “AI,” “VC,” “VI,” “MC,” “MI”) permits CoV values of arterioles, venules, and the microvasculature within ischemic “I” and control “C” areas to be compared. 
Vascular density measurements were calculated for the “I” and “C” areas from the vessel centerline segmentation (produced in step 5 in the previous section; see Supplementary Fig. S3). The vascular density was defined to be the number of pixels belonging to vessel centerlines divided by the total number of pixels within the “MI” and “MC” ROIs, respectively. Calculations were used to confirm ischemia. 
Statistical Analysis
For each participant, the analysis pipeline was used to compute the CoV of the vessel segment CoV values in each of the six ROIs (one value per vessel segment; see Fig. 2B). The CoV of the vessel segment CoV values within a ROI is a quantitative measure of the spatiotemporal variation in perfusion within the ROI. 
The CoV values for “MC,” “MI,” “VC,” “VI,” “AC,” and “AI” ROIs for each participant were taken to be measurements under different conditions. One-way repeated-measures analysis of variance (RM ANOVA) was performed to test the null hypothesis that there is no difference in mean CoV between the ROIs. A post hoc multiple comparison test, based on the paired-sample t-test, was performed to test the null hypothesis that there is no difference in the mean CoV between selected pairs of ROIs versus the alternate hypothesis that the difference is greater than zero (one-tailed test). 
Statistical analyses were performed using R with several additional packages.29 Notably, the rstatix and afex packages were used to perform RM ANOVA, and the emmeans package was used to perform the post hoc multiple comparison test with P value correction.3537 All tests were performed using a 5% level of significance. 
Results
Cohort Demographics
A total of 13 eyes from 13 patients were included in this study. The demographic and clinical characteristics of these patients are summarized in the Table. The mean age of patients was 61.5 years, and three patients were male. Mean time between diagnosis of BRVO and the acquisition of OCTA scans used in this study was 2.2 years. Mean BCVA was 20/30. Of the 13 patients, 9 had previously received treatment with anti-VEGF intravitreal injections for central macular edema, and 5 patients had hypertension or borderline hypertension that was well controlled with medication. 
Table.
 
Demographic Features of Cohort
Table.
 
Demographic Features of Cohort
Vessel Density Measurements in Ischemic and Control Areas
A paired sample t-test was performed to test whether the mean difference in vessel density between the ischemic microvasculature “MI” ROI and control microvasculature “MC” ROI is zero versus the alternative that it is less than zero. The mean difference of –3.46% ± 0.55% is statistically significant (P < 0.0001), suggesting that the mean vessel density is lower in the “MI” ROI (15.9% ± 2.8%) than the “MC” ROI (19.3% ± 1.1%). 
ROI CoV Measurements
Figure 4 shows a side-by-side boxplot of the CoVs for the “MC,” “MI,” “VC,” “VI,” “AC,” and “AI” ROIs for each patient. The “MC,” “VC,” and “AC” ROIs demarcate the microvasculature, large venules, and large arterioles, respectively, in the control “C” area. The “MI,” “VI,” and “AI” ROIs demarcate the microvasculature, large venules, and large arterioles, respectively, in the ischemic “I” area. Paired measurement data of these comparisons are provided in Supplementary Figure S4. For all 13 patients, the CoV value for the “MI” ROI is greater than that for the “MC” ROI. 
Figure 4.
 
Side-by-side boxplots of the vessel segment CoV for each ROI for all 13 patients with BRVO. The ROIs are defined with respect to an area of ischemia and a contralateral area not affected by BRVO used as an internal control. The ROIs “MC” demarcate the microvasculature (capillaries, small venules, and arterioles), “VC” large venules, and “AC” large arterioles, respectively, in the control area. The ROIs “MI” demarcate the microvasculature (capillaries, small venules, and arterioles), “VI” large venules, and “AI” large arterioles, respectively, in the area of ischemia. Paired measurements are explicitly shown in Supplementary Figure S4. *P < 0.0001.
Figure 4.
 
Side-by-side boxplots of the vessel segment CoV for each ROI for all 13 patients with BRVO. The ROIs are defined with respect to an area of ischemia and a contralateral area not affected by BRVO used as an internal control. The ROIs “MC” demarcate the microvasculature (capillaries, small venules, and arterioles), “VC” large venules, and “AC” large arterioles, respectively, in the control area. The ROIs “MI” demarcate the microvasculature (capillaries, small venules, and arterioles), “VI” large venules, and “AI” large arterioles, respectively, in the area of ischemia. Paired measurements are explicitly shown in Supplementary Figure S4. *P < 0.0001.
The data were checked to verify that it satisfied the test assumptions for the one-way RM ANOVA in relation to outliers, normality, and sphericity. Normality was assessed using the Shapiro–Wilk's test (P > 0.05 for each ROI). Mauchly's test indicated that the assumption of sphericity had been violated (P < 0.03); therefore, degrees of freedom were corrected using the Huynh–Feldt estimate of sphericity (ε = 0.075).38 The result was significant, F(5.35, 64.15) = 11.21, P < 0.001, generalized η2 = 0.430, suggesting that the mean CoV is not the same for all ROIs. 
Given the violation of sphericity, the Bonferroni method was chosen for the correction of P values for the post hoc multiple comparison test.39 The test was performed to compare microvasculature, arteriole, and venule ROIs between the control and ischemic areas: 
  • 1. Microvasculature (“MI” Vs. “MC”):
  • CoV is higher in “MI” (28.7% ± 0.7%) than “MC” (24.4% ± 0.7%), and the mean difference of 4.3% ± 0.6% is statistically significant (P < 0.0001). This corresponds to a proportional increase of (4.3/24.4 * 100) = 17.6% in “MI” compared to “MC.”
  • 2. Arterioles (“AI” Vs. “AC”):
  • The mean difference of 4.5% ± 2.0% is not statistically significant (P = 0.13), suggesting that there is no difference in CoV between the ischemic arteriole ROI “AI” and the control arteriole ROI “AC.”
  • 3. Venules (“VI” Vs. “VC”):
  • The mean difference of –0.75% ± 2.1% is not statistically significant (P = 1.00), suggesting that there is no difference in CoV between the ischemic venule ROI “VI” and the control venule ROI “VC.”
Discussion
The purpose of this study was to investigate the patterns of retinal perfusion abnormalities in BRVO. The main findings of this report are as follows: (1) perfusion properties of the macular circulation can be noninvasively quantified using OCTA-derived CoV measurements, and (2) variations in macular perfusion are increased within the microvasculature in regions containing macular ischemia in BRVO. 
Retinal perfusion is a dynamic process that is fine-tuned to the rapidly evolving energy demands of retinal neurons during photopic, scotopic, and disease states.40,41 Unlike many other vascular systems in the human body, the retinal circulation is largely devoid of autonomic innervation and relies on myogenic, metabolic, and rheologic mechanisms for autoregulation.16,42,43 Changes in retinal perfusion can therefore reflect structural injury to pericytes, smooth muscle cells, vascular endothelia, glia, or disease-induced upregulation of vasoactive metabolites such as nitrous oxide and lactate.4449 Measuring retinal perfusion therefore has tangible benefits for monitoring retinal vascular disease progression but may also serve as a harbinger of vascular perturbation prior to the onset of irreversible retinal structural damage. 
Several in vivo techniques have been previously used to study retinal perfusion in human eyes. Doppler velocimetry was described over 4 decades ago and has been used in patients with DM to study blood flow in large order vessels. The study by Nagaoka et al.6 found that retinal blood flow was significantly lower in retinal arterioles in patients with type 2 DM compared to nondiabetic control participants. Using a different method, the difference in absorption of light between oxyhemoglobin and deoxyhemoglobin was leveraged to generate oxygen saturation maps of the retina by Hardarson and colleagues.50 Using this technique, they showed an increase in oxygen saturation in arterioles and venules over time in patients with DR. Although Doppler velocimetry and oxygen saturation maps have provided valuable insights into the pathophysiology of retinal vascular diseases, the information attained from such techniques has largely been confined to large-order retinal vessels. Given that the major site of nutrient and waste exchange in the retina occurs at the level of the microvasculature, there remains an unmet need to develop novel methods to visualize the retinal capillary system.51 More recently, Pi et al.52 applied visible light OCT to measure total oxygen metabolic rate in the various capillary beds of the rat retina. This noninvasive technique has been a major leap forward in studying oxygen saturation in retinal capillaries. However, as the authors have stated in their study, this technique requires refinement before it can be applied in humans to reduce discomfort due to photoreceptor bleaching. 
OCTA is a noninvasive technique that provides rapid imaging of the retinal circulation. It involves rapidly acquiring repeated OCT B-scans at the same location. For static tissue, these will be identical, but in the case of motion from blood flow, the signal varies between them. The repetition time, or interscan time, between B-scans governs the range of detectable flows. A major limitation of conventional OCTA is that it is unable to quantify blood flow because of the relatively long interscan times used.53 Through a number of clinical–histology correlation reports, we have shown that OCTA can provide histology-like anatomic detail of retinal capillaries.9,10 A plethora of studies have characterized the OCTA features of the macular circulation in retinal vascular diseases.5457 However, these studies have almost exclusively been confined to reporting the quantitative and qualitative aspects of disease-induced structural alterations rather than focusing on dynamic changes to the retinal circulation such as perfusion. An exception to this is variable interscan time analysis (VISTA), a technique described by Ploner and colleagues58 to visualize relative blood flow speeds using OCTA. Using a custom-built OCTA device with a 400-kHz A-scan rate and a variable interscan time of 1.5m s and 3.0 ms between B-scans, the authors were able to calculate blood flow rates within the macula. The authors of the study did not specifically measure relative blood flow changes within regions of ischemia but did show that blood flow rates were reduced in retinal neovascularization and increased in the trunk of choroidal neovascular membranes. Further applications of VISTA found altered relative blood flow speeds in proximity of DR vascular abnormalities such as microaneurysms, intraretinal microvascular abnormalities, and neovascularization with a globally slower flow speed in DR eyes compared to control.59 A limitation of the VISTA technique is that it requires a high A-scan rate, not yet available in commercial machines.53 
The CoV technique we describe takes a different approach, seeking to quantify retinal perfusion in terms of the variation in OCTA signal across sequential OCTA acquisitions of the retina. The CoV of the intensity values for a single point in an OCTA volume or en face projection is a dimensionless quantitative measure of the variation of motion contrast at that location. In a vessel where the intensity varies very little, indicative of consistent flow, the CoV will be small. In a vessel that has fluctuating, increased, decreased, or completely intermittent flow, because of normal autoregulation or diseased state, the CoV will be higher. The CoV of the set of CoV values within the vessel network in a defined area is in turn a quantitative measure of the spatiotemporal variation of perfusion in that area. We found that the spatiotemporal variation in the ischemic microvasculature was consistently higher in every BRVO eye compared to the control microvasculature of the same eye. This provides a proof of concept for this method of quantifying perfusion variability, and with further understanding of this technique and perfusion variability in health and disease, it may prove effective for rapidly and noninvasively detecting abnormal variability in retinal perfusion in other retinal vasculopathies with currently available technology. This may have implications for early detection or prognostication of retinal vascular disease as it enables the additional assessment of functional vascular changes, which is understood to occur in early disease, as opposed to solely structural abnormalities as is currently done.2,60 However, such an analysis can be confounded by the occurrence of artifact, noise, and physiologic variables that make it difficult to isolate the influence of disease on OCTA-derived measures of macular perfusion. We have accounted for these confounding variables in our experimental model by confining our analysis to ischemic and control regions within the same eye. Arguably, in such a model, the effects of noise and physiologic variables such as heart rate and blood pressure are expected to have the same effect on control and disease regions with the only major pathogenic difference between the two areas being the occurrence of ischemia. Our CoV analysis method also intrinsically minimizes the impact of artifact and noise because the analysis is based on CoV values averaged over vessel segments rather than values for individual pixels, which are more prone to the influence of noise. Additionally, it corrects for intensity inhomogeneity (bias field) and admits the possibility of discarding one or more OCTA scans because of poor image quality (e.g., based on an instrument-provided quality metric and/or other quality metrics such as sharpness or signal-to-noise ratio). It should be noted that CoV measures will be influenced by the technique and device used to obtain the OCTA data. For example, the A-scan rate of the imaging device governs the detectable range of motion change in the scanned tissue.53 The OCTA data in this study were acquired using the Optovue RTVue XR Avanti system, which uses the SSADA algorithm. Using phantom experiments, Tokayer et al.61 established that a linear relationship exists between SSADA values and absolute flow velocity over a limited range, the values saturate for faster flows, and the range of speeds and transition to saturation depend on the time scale of SSADA measurement. 
In this study, we found it is possible to quantify perfusion properties of the retina with OCTA. Our findings are consistent with that of previous investigators who have also investigated retinal blood flow and perfusion.62 Bedggood and Metha63 utilized adaptive optics flood-illuminated ophthalmoscopy techniques and pixel-intensity cross-correlation to characterize microvascular flow in three healthy human participants. In their study, they found spatial variations in blood flow speeds at eccentricities 1° to 7° from fixation. When we studied perfusion changes within areas of macular ischemia, we found that CoV measurements increased by approximately 18% within the microvasculature when compared to the same-eye control. Post hoc analysis revealed that this increase was consistent in all 13 eyes of the study (Supplementary Fig. S4). No significant change was seen within arterioles and venules. 
Histologic studies have shown that ischemia due to BRVO is characterized by loss of pericytes and endothelial cells, acellularity of the capillary bed, and loss of neuroglial structures in the inner retinal layers.6466 Key retinal elements that control retinal perfusion include these pericytes, smooth muscle cells, and glia.17,67,68 Smooth muscle α-actin is a key protein that confers contractile properties to some of these cells in response to regional changes in glutamate, lactate, and nitric oxide concentrations.69,70 Tomasek and colleagues71 utilized knockout mice to demonstrate that the lack of smooth muscle α-actin in pericytes and smooth muscle cells altered the properties of the blood–retina barrier, including vascular permeability. It is therefore plausible that histologically proven loss of cellular elements within the collective apparatus that serves to fine-tune retinal perfusion can lead to rapid fluctuations in blood flow as correlated clinically by increased microvasculature CoV measurements. The relative absence of pericytes and smooth muscle cells around venules may explain why CoV measurements within retinal veins were not significantly altered within regions of ischemia.17 The lack of autonomic innervation of retinal arterioles limits the ability to rapidly couple blood flow to metabolic demands, this may explain why no alteration in arteriole CoV measurements was detected in ischemic regions.72 The process of autoregulation predominantly occurs at the level of capillaries, and our findings here are consistent with that concept.44 In addition to loss of smooth muscle α-actin–expressing cells, the upregulation of VEGF within regions of ischemia may also account for changes in CoV measurements. Previous studies have shown that VEGF can alter the permeability of vascular structures as well as alter blood flow within the circulation.73,74 Other factors and metabolites affected in response to ischemia such as lactate and nitric oxide may similarly affect capillary perfusion as found in this report.44,46 One published abstract has investigated a similar principle to ours with quantification of CoV to map perfusion heterogeneity between DR and control patients, finding variation in regions of both DR and healthy patients (Yuan PHS, et al. IOVS 2020;61:ARVO E-Abstract PB0020). The method described quantified CoV on a pixel-wise basis, reporting the mean and standard deviation over the whole image. In comparison, our method computes the mean CoV for vessel segments (which decreases susceptibility to artifacts) and also computes ROI-based CoV (better sensitivity to localized differences). 
The major advantages of this CoV technique are that it is safe, is noninvasive, uses a commercially available machine, and therefore can be readily applied in the clinical setting. With sufficient workflow automation, this technique may also prove to be relatively rapid. Having established perfusion variability is increased in ischemic regions of BRVO, this CoV measure may be investigated longitudinally as a biomarker in clinical studies to determine how it may stratify the risk of developing sight-threatening complications of BRVO such as macular edema and neovascularization. Further application of this technique may provide novel insights regarding the pathophysiology of other ischemic retinopathies where it is known that perfusion changes occur well before structural abnormalities are visible.2,60 A limitation of this report is the sample size and the inclusion of patients with good visual acuity who were able to fixate during imaging and reduce artifact. In certain populations, poor image quality due to reduced visual acuity and inability to fixate may preclude application of this CoV technique. Another limitation is that no comparison of this technique to other in vivo modalities that more directly quantify blood flow was made in this study. We acknowledge that it would be important to validate this technique against technologies such as AOSLO and live in vivo microscopy in human and animal models of retinal ischemia.8 We note, however, that our previously published CoV results using the same technique in normal participants are consistent with those obtained by Warner et al.24 using AOSLO.15 Both techniques demonstrate that venules and arterioles have similar CoV values and that CoV values for capillaries are higher. Nevertheless, the analysis demonstrated a consistent increase in CoV measurements of the microvasculature in the ischemic region of every eye compared to control. The impacts of artifact and noise were attempted to be accounted for by including a control region within the same eye and performing rigorous alignment, bias field correction, intensity normalization, and vessel segment analysis. Potential confounders such as recent anti-VEGF therapy, uncontrolled hypertension, and recent caffeine intake were accounted for. 
The comparison of ischemic and control areas in BRVO eyes described here provides proof of principle that OCTA-derived CoV measurements can be used to quantify retinal perfusion variability in human participants. 
Acknowledgments
Supported by National Health and Medical Research Council of Australia Investigator Grant (APP1173403), Perth Eye Foundation, Perth, Australia. 
Disclosure: M. Hein, None; A. Mehnert, None; K.B. Freund, Genentech (C), Zeiss (C), Heidelberg Engineering (C), Allergan (C), Bayer (C), and Novartis (C); D.-Y. Yu, None; C. Balaratnasingam, Allergan (C), Bayer (C), Novartis (C), Roche Pharmaceuticals (C) 
References
Pournaras CJ, Rungger-Brandle E, Riva CE, Hardarson SH, Stefansson E. Regulation of retinal blood flow in health and disease. Prog Retin Eye Res. 2008; 27: 284–330. [CrossRef] [PubMed]
Alder VA, Su EN, Yu DY, Cringle S, Yu P. Overview of studies on metabolic and vascular regulatory changes in early diabetic retinopathy. Aust N Z J Ophthalmol. 1998; 26: 141–148. [CrossRef] [PubMed]
Cringle SJ, Yu D-Y, Alder VA, Su EN. Retinal blood flow by hydrogen clearance polarography in the streptozotocin-induced diabetic rat. Invest Ophthalmol Vis Sci. 1993; 34: 1716–1721. [PubMed]
Joseph A, Guevara-Torres A, Schallek J. Imaging single-cell blood flow in the smallest to largest vessels in the living retina. Elife. 2019; 8: E45077. [CrossRef] [PubMed]
Kornfield TE, Newman EA. Measurement of retinal blood flow using fluorescently labeled red blood cells. Eneuro. 2015; 2: 0005–0015.2015. [CrossRef]
Nagaoka T, Sato E, Takahashi A, Yokota H, Sogawa K, Yoshida A. Impaired retinal circulation in patients with type 2 diabetes mellitus: retinal laser Doppler velocimetry study. Invest Ophthalmol Vis Sci. 2010; 51: 6729–6734. [CrossRef] [PubMed]
Grunwald JE, Riva CE, Sinclair SH, Brucker AJ, Petrig BL. Laser Doppler velocimetry study of retinal circulation in diabetes mellitus. Arch Ophthalmol. 1986; 104: 991–996. [CrossRef] [PubMed]
Burns SA, Elsner AE, Sapoznik KA, Warner R, Gast TJ. Adaptive optics imaging of the human retina. Prog Retin Eye Res. 2019; 68: 1–30. [CrossRef] [PubMed]
An D, Balaratnasingam C, Heisler M, et al. Quantitative comparisons between optical coherence tomography angiography and matched histology in the human eye. Exp Eye Res. 2018; 170: 13–19. [CrossRef] [PubMed]
Yu PK, Mehnert A, Athwal A, Sarunic MV, Yu DY. Use of the retinal vascular histology to validate an optical coherence tomography angiography technique. Transl Vis Sci Technol. 2021; 10: 29. [CrossRef] [PubMed]
Balaratnasingam C, An D, Freund KB, Francke A, Yu DY. Correlation between histologic and oct angiography analysis of macular circulation. Ophthalmology. 2019; 126: 1588–1589. [CrossRef] [PubMed]
Spaide RF, Fujimoto JG, Waheed NK, Sadda SR, Staurenghi G. Optical coherence tomography angiography. Prog Retin Eye Res. 2018; 64: 1–55. [CrossRef] [PubMed]
Arya M, Rashad R, Sorour O, Moult EM, Fujimoto JG, Waheed NK. Optical coherence tomography angiography (OCTA) flow speed mapping technology for retinal diseases. Expert Rev Med Devices. 2018; 15: 875–882. [CrossRef] [PubMed]
Yu DY, Cringle SJ, Yu PK, et al. Retinal capillary perfusion: spatial and temporal heterogeneity. Prog Retin Eye Res. 2019; 70: 23–54. [CrossRef] [PubMed]
Yu DY, Mehnert A, Balaratnasingam C, et al. An assessment of microvascular hemodynamics in human macula. Sci Rep. 2023; 13: 7550. [CrossRef] [PubMed]
Yu DY, Su EN, Cringle SJ, Yu PK. Isolated preparations of ocular vasculature and their applications in ophthalmic research. Prog Retin Eye Res. 2003; 22: 135–169. [CrossRef] [PubMed]
Balaratnasingam C, An D, Hein M, Yu P, Yu D-Y. Studies of the retinal microcirculation using human donor eyes and high-resolution clinical imaging: insights gained to guide future research in diabetic retinopathy. Prog Retin Eye Res. 2022; 94: 101134. [CrossRef] [PubMed]
Sim DA, Keane PA, Zarranz-Ventura J, et al. The effects of macular ischemia on visual acuity in diabetic retinopathy. Invest Ophthalmol Vis Sci. 2013; 54: 2353–2360. [CrossRef] [PubMed]
Heidary F, Gharebaghi R. Outcomes of retinopathy of prematurity. Med Hypothesis Discov Innov Ophthalmol. 2016; 5: 112–114. [PubMed]
Noma H, Funatsu H, Mimura T, Shimada K. Influence of ischemia on visual function in patients with branch retinal vein occlusion and macular edema. Clin Ophthalmol. 2011; 5: 679–685. [PubMed]
Pournaras CJ, Rungger-Brändle E, Riva CE, Hardarson SH, Stefansson E. Regulation of retinal blood flow in health and disease. Prog Retin Eye Res. 2008; 27: 284–330. [CrossRef] [PubMed]
Early Treatment Diabetic Retinopathy Study Research Group. Classification of diabetic retinopathy from fluorescein angiograms. ETDRS Report Number 11. Ophthalmology. 1991; 98: 807–822. [CrossRef] [PubMed]
Spaide RF, Fujimoto JG, Waheed NK. Image artifacts in optical coherence tomography angiography. Retina. 2015; 35: 2163–2180. [CrossRef] [PubMed]
Warner RL, Gast TJ, Sapoznik KA, Carmichael-Martins A, Burns SA. Measuring temporal and spatial variability of red blood cell velocity in human retinal vessels. Invest Ophthalmol Vis Sci. 2021; 62: 29. [CrossRef] [PubMed]
Gillbricht M, Robert R, Sokal F, Rohlf J. Biometry: The Principles and Practice of Statistics in Biological Research. Folkestone, UK: W.H. Freeman And Company Ltd; 1969.
Gong G, Zhang H, Yao M. Speckle noise reduction algorithm with total variation regularization in optical coherence tomography. Opt Express. 2015; 23: 24699–24712. [CrossRef] [PubMed]
Zang P, Liu G, Zhang M, et al. Automated motion correction using parallel-strip registration for wide-field en face OCT angiogram. Biomed Opt Express. 2016; 7: 2823–2836. [CrossRef] [PubMed]
Schindelin J, Arganda-Carreras I, Frise E, et al. Fiji: an open-source platform for biological-image analysis. Nat Methods. 2012; 9: 676–682. [CrossRef] [PubMed]
R-Development-Core-Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation For Statistical Computing; 2011.
Lowe DG. Distinctive image features from scale-invariant keypoints. Int J Comput Vis. 2004; 60: 91–110. [CrossRef]
Clayden J, Modat M, Presles B, Anthopoulos T, Daga P. Rniftyreg: Image Registration Using the ‘Niftyreg’ Library. R Package; 2020, https://cran.r-project.org/web/packages/RNiftyReg/RNiftyReg.pdf. Accessed September 6, 2022.
Wagner T, Hiner M, Raynaud X. Thorstenwager/Ij-Ridgedetection: Ridge Detection 1.4.0 (Version V1.4.0). Zenodo; 2017, https://zenodo.org/records/845874. Accessed September 6, 2022.
Steger C. An unbiased detector of curvilinear structures. IEEE Trans Pattern Anal Machine Intell. 1998; 20: 113–125. [CrossRef]
An D, Balaratnasingam C, Heisler M, et al. Quantitative comparisons between optical coherence tomography angiography and matched histology in the human eye. Exp Eye Res. 2018; 170: 13–19. [CrossRef] [PubMed]
Kassambara A. Rstatix: Pipe-Friendly Framework for Basic Statistical Tests. R Package Version 0.7.0. 2021, https://cran.r-project.org/web/packages/rstatix/index.html. Accessed September 6, 2022.
Singmann H, Bolker B, Westfall J, Aust F, Ben-Shachat M. Afex: Analysis of Factorial Experiments. R Package Version 1.0-1; 2021, https://cran.r-project.org/web/packages/afex/afex.pdf. Accessed September 6, 2022.
Lenth R. Emmeans: Estimated Marginal Means, aka Least-Squares Means. R Package Version 1.7.4-1; 2022, https://cran.r-project.org/web/packages/emmeans/emmeans.pdf. Accessed September 6, 2022.
Haverkamp N, Beauducel A. Violation of the sphericity assumption and its effect on type-I error rates in repeated measures ANOVA and multi-level linear models (MLM). Front Psychol. 2017; 8: 1841. [CrossRef] [PubMed]
Maxwell SE. Pairwise multiple comparisons in repeated measures designs. J Educ Stat. 1980; 5: 269–287. [CrossRef]
Harris A, Ciulla TA, Chung HS, Martin B. Regulation of retinal and optic nerve blood flow. Arch Ophthalmol. 1998; 116: 1491–1495. [CrossRef] [PubMed]
Courtie E, Veenith T, Logan A, Denniston AK, Blanch RJ. Retinal blood flow in critical illness and systemic disease: a review. Ann Intens Care. 2020; 10: 152. [CrossRef]
Arciero J, Harris A, Siesky B, et al. Theoretical analysis of vascular regulatory mechanisms contributing to retinal blood flow autoregulation. Invest Ophthalmol Vis Sci. 2013; 54: 5584–5593. [CrossRef] [PubMed]
Iadecola C. The neurovascular unit coming of age: a journey through neurovascular coupling in health and disease. Neuron. 2017; 96: 17–42. [CrossRef] [PubMed]
Riva CE, Grunwald JE, Petrig BL. Autoregulation of human retinal blood flow. Invest Ophthalmol Vis Sci. 1986; 27: 1706–1712. [PubMed]
Yu CY, Boyd NM, Cringle SJ, Su EN, Yu DY. Vasoactive response of isolated pulpal arterioles to endothelin-1. J Endod. 2004; 30: 149–153. [CrossRef] [PubMed]
Yu DY, Su EN, Cringle SJ, Morgan WH, Mcallister IL, Yu PK. Local modulation of retinal vein tone. Invest Ophthalmol Vis Sci. 2016; 57: 412–419. [CrossRef] [PubMed]
Gonzales AL, Klug NR, Moshkforoush A, et al. Contractile pericytes determine the direction of blood flow at capillary junctions. Proc Natl Acad Sci USA. 2020; 117: 27022–27033. [CrossRef] [PubMed]
Hamilton NB, Attwell D, Hall CN. Pericyte-mediated regulation of capillary diameter: a component of neurovascular coupling in health and disease. Front Neuroenergetics. 2010; 2: 5. [CrossRef] [PubMed]
Metea MR, Newman EA. Glial cells dilate and constrict blood vessels: a mechanism of neurovascular coupling. J Neurosci. 2006; 26: 2862–2870. [CrossRef] [PubMed]
Hardarson SH, Stefánsson E, Bek T. Retinal oxygen saturation changes progressively over time in diabetic retinopathy. PLoS One. 2021; 16: E0251607. [CrossRef] [PubMed]
Tuma RF, Duran WN, Ley K. Handbook of Physiology: Microcirculation. 2nd ed. San Diego, CA: Academic Press (Elsevier); 2008.
Pi S, Hormel TT, Wei X, et al. Retinal capillary oximetry with visible light optical coherence tomography. Proc Natl Acad Sci USA. 2020; 117: 11658–11666. [CrossRef] [PubMed]
Richter D, Fard AM, Straub J, Wei W, Zhang Q, Wang RK. Relative retinal flow velocity detection using optical coherence tomography angiography imaging. Biomed Opt Express. 2020; 11: 6710–6720. [CrossRef] [PubMed]
Hwang TS, Jia Y, Gao SS, et al. Optical coherence tomography angiography features of diabetic retinopathy. Retina (Philadelphia, Pa). 2015; 35: 2371–2376. [CrossRef] [PubMed]
El Ameen A, Cohen SY, Semoun O, et al. Type 2 neovascularization secondary to age-related macular degeneration imaged by optical coherence tomography angiography. Retina. 2015; 35: 2212–2218. [CrossRef] [PubMed]
Coscas GJ, Lupidi M, Coscas F, Cagini C, Souied EH. Optical coherence tomography angiography versus traditional multimodal imaging in assessing the activity of exudative age-related macular degeneration: a new diagnostic challenge. Retina. 2015; 35: 2219–2228. [CrossRef] [PubMed]
Kashani AH, Lee SY, Moshfeghi A, Durbin MK, Puliafito CA. Optical coherence tomography angiography of retinal venous occlusion. Retina. 2015; 35: 2323–2331. [CrossRef] [PubMed]
Ploner SB, Moult EM, Choi W, et al. Toward quantitative optical coherence tomography angiography: visualizing blood flow speeds in ocular pathology using variable interscan time analysis. Retina. 2016; 36(suppl 1): S118–S126. [PubMed]
Arya M, Filho MB, Rebhun CB, et al. Analyzing relative flow speeds in diabetic retinopathy using variable interscan time analysis OCT angiography. Ophthalmol Retina. 2021; 5: 49–59. [CrossRef] [PubMed]
Alder VA, Su EN, Yu D-Y, Cringle SJ, Yu PK. Diabetic retinopathy: early functional changes. Clin Exp Pharmacol Physiol. 1997; 24: 785–788. [CrossRef] [PubMed]
Tokayer J, Jia Y, Dhalla AH, Huang D. Blood flow velocity quantification using split-spectrum amplitude-decorrelation angiography with optical coherence tomography. Biomed Opt Express. 2013; 4: 1909–1924. [CrossRef] [PubMed]
Riva CE, Grunwald JE, Sinclair SH, Petrig BL. Blood velocity and volumetric flow rate in human retinal vessels. Invest Ophthalmol Vis Sci. 1985; 26: 1124–1132. [PubMed]
Bedggood P, Metha A. Mapping flow velocity in the human retinal capillary network with pixel intensity cross correlation. PLoS One. 2019; 14: E0218918. [CrossRef] [PubMed]
Kornzweig AL, Eliasoph I, Feldstein M. Occlusive disease of retinal vasculature: a clinicopathological study. Arch Ophthalmol. 1964; 71: 542–551. [CrossRef] [PubMed]
Frangieh GT, Green WR, Barraquer-Somers E, Finkelstein D. Histopathologic study of nine branch retinal vein occlusions. Arch Ophthalmol. 1982; 100: 1132–1140. [CrossRef] [PubMed]
Hamilton AM, Kohner EM, Rosen D, Bird AC, Dollery CT. Experimental retinal branch vein occlusion in rhesus monkeys. I. Clinical appearances. Br J Ophthalmol. 1979; 63: 377–387. [CrossRef] [PubMed]
An D, Chung-Wah-Cheong J, Yu DY, Balaratnasingam C. Alpha-smooth muscle actin expression and parafoveal blood flow pathways are altered in preclinical diabetic retinopathy. Invest Ophthalmol Vis Sci. 2022; 63: 8. [CrossRef] [PubMed]
An D, Chandrasekera E, Yu DY, Balaratnasingam C. Non-proliferative diabetic retinopathy is characterized by nonuniform alterations of peripapillary capillary networks. Invest Ophthalmol Vis Sci. 2020; 61(4): 39. [CrossRef] [PubMed]
Yu PK, An D, Balaratnasingam C, Cringle SJ, Yu DY. Topographic distribution of contractile protein in the human macular microvasculature. Invest Ophthalmol Vis Sci. 2019; 60: 4574–4582. [CrossRef] [PubMed]
Kim SJ, Kim SA, Choi YA, Park DY, Lee J. Alpha-smooth muscle actin-positive perivascular cells in diabetic retina and choroid. Int J Mol Sci. 2020; 21(6): 2158. [CrossRef] [PubMed]
Tomasek JJ, Haaksma CJ, Schwartz RJ, et al. Deletion of smooth muscle alpha-actin alters blood-retina barrier permeability and retinal function. Invest Ophthalmol Vis Sci. 2006; 47: 2693–2700. [CrossRef] [PubMed]
Laties AM. Central retinal artery innervation. Absence of adrenergic innervation to the intraocular branches. Arch Ophthalmol. 1967; 77: 405–409. [CrossRef] [PubMed]
Rissanen TT, Korpisalo P, Markkanen JE, et al. Blood flow remodels growing vasculature during vascular endothelial growth factor gene therapy and determines between capillary arterialization and sprouting angiogenesis. Circulation. 2005; 112: 3937–3946. [CrossRef] [PubMed]
Ali M, Bracko O. VEGF paradoxically reduces cerebral blood flow in Alzheimer's disease mice. Neurosci Insights. 2022; 17: 26331055221109254. [CrossRef] [PubMed]
Figure 1.
 
CoV analysis pipeline: steps 1 to 6 leading to the generation of the vessel centerline CoV map. The pipeline takes as input multiple consecutive OCT angiograms (A) and the corresponding en face OCT projection images (B). For each pair of OCTA and OCT images, bias field correction is performed on the OCTA image using an estimate of the bias field computed from the OCT image. The corrected OCTA images then undergo spatial alignment and normalization of intensity (C). Mean and SD projections of the resulting OCTA stack are generated (D and E, respectively). The vessel centerline map is created from the mean projection (F). The mean and SD projections are used to calculate the CoV for each pixel of the vessel centerline map. The resulting vessel centerline CoV map is visualized using a heatmap and (optionally) superimposed on the mean image (G).
Figure 1.
 
CoV analysis pipeline: steps 1 to 6 leading to the generation of the vessel centerline CoV map. The pipeline takes as input multiple consecutive OCT angiograms (A) and the corresponding en face OCT projection images (B). For each pair of OCTA and OCT images, bias field correction is performed on the OCTA image using an estimate of the bias field computed from the OCT image. The corrected OCTA images then undergo spatial alignment and normalization of intensity (C). Mean and SD projections of the resulting OCTA stack are generated (D and E, respectively). The vessel centerline map is created from the mean projection (F). The mean and SD projections are used to calculate the CoV for each pixel of the vessel centerline map. The resulting vessel centerline CoV map is visualized using a heatmap and (optionally) superimposed on the mean image (G).
Figure 2.
 
CoV analysis pipeline: steps 7 to 8 leading to the generation of the vessel segment CoV map. Steps 1 to 6 are detailed in Figure 1. (A) The pixelwise CoV map from Figure 1G. In this map, each individual pixel from the vessel centerline map is assigned a CoV value. (B) The vessel segment CoV map where each vessel segment is assigned a CoV value from the mean of its constituent pixel CoV values. A vessel segment is defined to be a sequence of pixels between vessel junctions and endpoints in the centerline image. The two insets demonstrate on two vessel segments how a mean CoV value (B, inset) is derived from its constituent pixel CoV values (A, inset). The scale bar describes the color representation of CoV values in panels A and B.
Figure 2.
 
CoV analysis pipeline: steps 7 to 8 leading to the generation of the vessel segment CoV map. Steps 1 to 6 are detailed in Figure 1. (A) The pixelwise CoV map from Figure 1G. In this map, each individual pixel from the vessel centerline map is assigned a CoV value. (B) The vessel segment CoV map where each vessel segment is assigned a CoV value from the mean of its constituent pixel CoV values. A vessel segment is defined to be a sequence of pixels between vessel junctions and endpoints in the centerline image. The two insets demonstrate on two vessel segments how a mean CoV value (B, inset) is derived from its constituent pixel CoV values (A, inset). The scale bar describes the color representation of CoV values in panels A and B.
Figure 3.
 
ROI selection for CoV analysis example on a single eye with BRVO. FA (A) and a mean projection of multiple consecutive aligned OCT angiograms (B) over the same macula. Areas of capillary dropout (yellow asterisks) on FA shown in panel A are correlated with areas of dropout on OCTA seen in panel B. The process of ROI selection is depicted in panel B. The ischemic “I” area (red rectangle) encompasses areas of capillary dropout and absence of fluorescence in the BRVO territory. The control “C” area (green rectangle) is a mirrored area flipped over the horizontal meridian to capture the unaffected macula. Radial arterioles (cyan “a”) and venules (magenta “v”) are defined within the ischemic and control areas (green vessel outline; B) and are segregated to create arteriole ROIs and venule ROIs within the ischemic and control area for comparison. (C, D) Magnified insets over the same area in panel B (orange dotted rectangle) and depict how an OCTA region (C) is represented as a vessel segment CoV map (D). A large vessel mask that encompasses large vessels (gray vessels, D) is used to isolate the microvasculature for analysis. This process generates six ROIs for comparison: microvasculature in control “MC” and ischemic “MI” areas, arterioles in control “AC” and ischemic “AI” areas, and venules in control “VC” and ischemic “VI” areas. The scale bar describes the color representation of CoV in panel D.
Figure 3.
 
ROI selection for CoV analysis example on a single eye with BRVO. FA (A) and a mean projection of multiple consecutive aligned OCT angiograms (B) over the same macula. Areas of capillary dropout (yellow asterisks) on FA shown in panel A are correlated with areas of dropout on OCTA seen in panel B. The process of ROI selection is depicted in panel B. The ischemic “I” area (red rectangle) encompasses areas of capillary dropout and absence of fluorescence in the BRVO territory. The control “C” area (green rectangle) is a mirrored area flipped over the horizontal meridian to capture the unaffected macula. Radial arterioles (cyan “a”) and venules (magenta “v”) are defined within the ischemic and control areas (green vessel outline; B) and are segregated to create arteriole ROIs and venule ROIs within the ischemic and control area for comparison. (C, D) Magnified insets over the same area in panel B (orange dotted rectangle) and depict how an OCTA region (C) is represented as a vessel segment CoV map (D). A large vessel mask that encompasses large vessels (gray vessels, D) is used to isolate the microvasculature for analysis. This process generates six ROIs for comparison: microvasculature in control “MC” and ischemic “MI” areas, arterioles in control “AC” and ischemic “AI” areas, and venules in control “VC” and ischemic “VI” areas. The scale bar describes the color representation of CoV in panel D.
Figure 4.
 
Side-by-side boxplots of the vessel segment CoV for each ROI for all 13 patients with BRVO. The ROIs are defined with respect to an area of ischemia and a contralateral area not affected by BRVO used as an internal control. The ROIs “MC” demarcate the microvasculature (capillaries, small venules, and arterioles), “VC” large venules, and “AC” large arterioles, respectively, in the control area. The ROIs “MI” demarcate the microvasculature (capillaries, small venules, and arterioles), “VI” large venules, and “AI” large arterioles, respectively, in the area of ischemia. Paired measurements are explicitly shown in Supplementary Figure S4. *P < 0.0001.
Figure 4.
 
Side-by-side boxplots of the vessel segment CoV for each ROI for all 13 patients with BRVO. The ROIs are defined with respect to an area of ischemia and a contralateral area not affected by BRVO used as an internal control. The ROIs “MC” demarcate the microvasculature (capillaries, small venules, and arterioles), “VC” large venules, and “AC” large arterioles, respectively, in the control area. The ROIs “MI” demarcate the microvasculature (capillaries, small venules, and arterioles), “VI” large venules, and “AI” large arterioles, respectively, in the area of ischemia. Paired measurements are explicitly shown in Supplementary Figure S4. *P < 0.0001.
Table.
 
Demographic Features of Cohort
Table.
 
Demographic Features of Cohort
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×