Quantitative Comparison of Retinal Capillary Images Derived By Speckle Variance Optical Coherence Tomography With Histology

Abstract

Purpose.: The purpose of this study was to correlate human retinal capillary network information derived from a prototype speckle variance optical coherence tomography (svOCT) device with histology to determine the utility of this instrument for quantitative angiography.

Methods.: A retina location 3 mm superior to the optic disk was imaged with svOCT in 14 healthy human eyes. Qualitative and quantitative features of capillary networks, including capillary diameter and density, were compared with perfusion-labeled histological specimens from the same eccentricity. Twelve human donor eyes with no history of eye disease were used for histological comparisons.

Results.: svOCT was able to clearly distinguish the morphological features of the nerve fiber layer capillary network, the retinal ganglion cell (RGC) layer capillary network, the capillary network at the border of the inner plexiform layer and superficial boundary of the inner nuclear layer, and the capillary network at the boundary of the deep inner nuclear layer and outer plexiform layer. The morphological features of these networks were highly comparable to those in previous histological studies. There were no statistical differences in mean capillary diameter between svOCT images and histology for all networks other than the RGC capillary network. Capillary density measurements were significantly greater in svOCT images, except in the RGC capillary network.

Conclusions.: svOCT has the capacity to provide histology-like anatomical information about human retinal capillary networks in vivo. It may have great potential as a research and diagnostic tool in the management of retinal vascular diseases. Further work is required to clarify the cause of some quantitative differences between svOCT and histology.

The high metabolic demands of retinal neurons^1–3 are satisfied by a complex and morphologically unique system of capillary networks.^4–8 Quantitative characteristics of each capillary network, with respect to capillary diameter, loop area, and density, are varied relative to neuronal layer and retinal eccentricity.^8,9 Collectively, the retinal circulation including the capillary bed systems and major vascular conduits span a total area of approximately 1000 mm².¹⁰ Due to the nature of terminal retinal arteries and the absence of collateral circulations, the retina is particularly vulnerable to ischemic injury. Because capillaries represent the major site of nutrient delivery and toxic substrate exchange between neurons and vasculature,^11–14 an understanding of capillary network distribution and their response to injury is especially important for our understanding of retinal vascular diseases. 

Our previous work provided an in-depth quantitative histological analysis of the different capillary networks in the normal human retina.^8,9 Subsequently, we were also able to demonstrate that the presence of cardiovascular morbidity had a selective influence on capillary networks prior to the onset of clinically detectable retinal disease.¹⁵ Specifically, there was significant nonuniformity in the magnitude of disease-induced exchange between capillary networks that supplied predominantly somal, axonal, and synaptic compartments within the retina. Therefore, it is likely that selective capillary network disease may also be critically linked to the pathogenesis of diabetic retinopathy and other microvascular diseases. 

Our ability to study distinct retinal capillary networks in real-time and in vivo has been limited.^16,17 Fluorescein angiography, although an excellent investigative tool, requires the administration of a contrast agent and is associated with many adverse effects.^18–22 Noninvasive techniques that have the capacity to perform angiography without the administration of a contrast agent are thus highly desirable. Although retinal capillaries have low optical contrast in scanning laser ophthalmoscopy (SLO), recent development of label-free flow contrast and forward scattering/offset pinhole techniques have been described in the literature using adaptive optics scanning laser ophthalmoscopy.^23–26 Label-free, depth-resolved optical angiography using optical coherence tomography (OCT) for visualization of capillary flow has also been reported. Various techniques for extracting flow contrast from OCT data have been described, including phase variance,²⁷ optical microangiography,²⁸ speckle variance (sv),²⁹ phase-resolved,³⁰ and split spectrum amplitude decorrelation angiography.³¹ Each of these techniques has its own advantage, but they all operate on similar principles. Variations in the intensity and/or phase of the OCT signal caused by particle motion are detected by pixel-to-pixel comparison across repeat B-scan images acquired at the same location. The strength of these techniques is that they are sensitive to the slow flow of blood cells in capillaries. A detailed review of flow contrast OCT comparing these techniques has recently been published.³² Although the image quality of the various implementations of flow contrast OCT are comparable for in vivo imaging, the svOCT technique, described by Mariampillai et al.²⁹ is computationally simple, permitting the images to be generated in real time during acquisition.^33,34 In this study, we used a custom-built svOCT device³⁵ to quantitatively study the different capillary networks in the human retina. We took advantage of real-time svOCT imaging capability for visual feedback to optimize the image quality during acquisition of images from research subjects. Comparisons are made to our previous quantitative histological report.⁸ We demonstrate the capability of this device for defining and providing detailed quantitative information of distinct capillary networks that may be useful in the clinical setting. 

Materials and Methods

This study was approved by the human research ethics committees at The University of Western Australia and The University of British Columbia. All imaging of live patients was performed with informed consent at the Eye Care Centre in Vancouver. All human tissue was handled according to the tenets of the Declaration of Helsinki. 

Human Donor Eyes

A total of 12 eyes from 10 human donors were used for histological study with confocal scanning laser microscopy. The eyes used in this study were also used in our previous work.⁸ Donor eyes were acquired from the Lions Eye Bank (Lions Eye Institute, Western Australia) following removal of corneal buttons for transplantation. All human donor eyes used in this study had no known history of ocular disease. 

Perfusion Labeling of Retinal Capillary Networks

The method used for targeted endothelial labeling of retinal capillary networks was identical to that described in our previous reports.^36,37 Briefly, following cannulation of the central retinal artery with a micropipette, the following sequence of solutions were perfused through the retinal circulation: (1) 20-minute washout of blood with oxygenated Ringer's solution and 1% bovine serum albumin; (2) 30-minute fixation with 4% paraformaldehyde in 0.1 M phosphate buffer; (3) 7-minute wash with 0.1% Triton X-100 in buffer; (4) 30-minute phosphate buffer wash; (5) 2-hour perfusion labeling of microfilaments; and (6) 30-minute phosphate buffer wash. Specimens were then immersed in 4% paraformaldehyde solution overnight. We labeled the f-actin microfilaments with a mixture of phalloidin conjugated to Alexa Fluor 546 (30 U; Life Technologies Australia, Mulgrave, Victoria, Australia). Nuclei were labeled with bis-benzamide H 33258 (1.2 μg/mL; Sigma-Aldrich Corp., St. Louis, MO, USA) or iodide dye (YO-PRO-1; 6.6 μM; Thermo Scientific, Waltham, MA, USA). 

Tissue Preparation and Confocal Scanning Laser Microscopy

Following perfusion labeling, eyes were dissected at the equator to allow viewing of the posterior retina. A few cuts were made to the retina to enable flat mounting in glycerol. Images were captured from locations of the retina 3 mm superior to the optic disk. Two C1 microscopes (Nikon, Minato-ku, Tokyo, Japan) equipped with either three lasers (wavelengths of 405 nm, 488 nm, and 532 nm) or four lasers (wavelengths of 405 nm, 488 nm, 545 nm, and 637 nm) were used for confocal imaging. Both of the microscopes were equipped with EZ-C1 software (version 3.20). A 20× (NA 0.75) lens was used for confocal imaging. Using a motorized stage, we captured a series of z-stacks for each specimen beginning from the vitreal surface at the level of the inner limiting membrane to the outer retina. Each z-stack consisted of a depth of optical sections collected at 0.35-μm increments along the z-plane. Images of different wavelengths were acquired sequentially. 

Speckle Variance OCT Imaging of Human Subjects

svOCT images of human subjects were acquired using a graphics-processing unit (GPU)-accelerated svOCT clinical prototype machine.^34,38 Details of the acquisition system have previously been published.³⁵ The OCT system was based on a 1060-nm swept source with 100 kHz A-scan rate (Axsun Technologies, Inc., Billerica, MA, USA) and 500 mega samples per second digitizer (AlazarTech, Inc., Pointe-Claire, QC, Canada). The light source spectrum had a full-width half-maximum (FWHM) bandwidth of 61.5 nm, which corresponded to a coherence length of ∼6 μm in tissue. Sample arm optics were configured to deliver a beam of ∼1.5-mm diameter at the subject's pupil, with the fast axis of galvanometer-mounted mirrors (6210H; Cambridge Technology, Inc., Bedford, MA, USA) oriented for a vertical scan. The size of the focal waist on the retina was estimated to be ω_o = ∼7.3 μm (calculated using Gaussian optics), using the Gullstrand-LeGrand model of the human eye,^39,40 corresponding to a lateral FWHM of ∼8.6 μm, calculated as FWHM = Display Formula . The scan area was sampled by a 300 × 300 (×3) pixel grid on a ∼1- × 1-mm field (corresponding to ∼3.3 μm between A-scans) of view in ∼ 3.15 seconds. Patient alignment was performed using a wider field, high speed, low-resolution en face imaging mode with the OCT system, providing visual targets to guide the subject's fixation so that the center of the scan area was ∼3 mm superior to the optic nerve head (ONH). These intensity-only OCT intensity images were not saved. The larger vessels provided landmarks for the acquisition of the ∼1- × 1-mm² svOCT data at this location. In order to have data suitable for accurate capillary analysis, the svOCT data were saved in a region avoiding large blood vessels. For the sv calculation, three repeat acquisitions at each B-scan location were acquired. En face visualization of the retinal microvasculature was processed and displayed in real-time using our open-source svOCT code program developed for GPU.³⁴ The GPU software permitted dynamic selection of the retinal layers used for generating the en face svOCT image. Real-time processing to improve the svOCT image quality included brightness and contrast adjustment to eliminate low values of sv, and filtering to remove streak artifacts.³⁴ In vivo scan dimensions of the retina were calculated using a reduced eye model (single refractive surface),⁴¹ adjusted for the eye length of each participant measured using the IOL Master 500 (Zeiss, Oberkochen, Baden-Württemberg, Germany).Scan dimensions of the retina were calculated as the length of an arc traced by the OCT beam as it was scanned in angle (corrected for Snell's law refraction) assuming a circle of radius equal to the subject's eye length. The index of refraction for eye medium (vitreous) was approximated as n = ∼1.33. The svOCT en face images were cropped to match the dimensions of the images used for the ex vivo confocal microscopy analysis. Images were acquired from both eyes of 7 patients.  

Segmentation of Capillary Networks

Our previous report demonstrated the presence of four different capillary networks in the retina in the region at 3 mm superior to the human optic disk.⁸ These networks are located in (1) nerve fiber layer (NFL); (2) retinal ganglion cell (RGC) layer; (3) the border of inner plexiform layer (IPL) and superficial boundary of the inner nuclear layer (INL); and (4) the boundary of deep INL and outer plexiform layer (OPL) (Fig. 1). In our previous study we colocalized the position of nuclei with respect to endothelial microfilaments to define different capillary networks. 

Figure 1

 

Region of study. Transverse histological retinal section was stained with toluidine blue (A), and a B-scan image was acquired using svOCT (B) to illustrate the various retinal layers at the eccentricity located 3 mm superior to the optic disk. Colored dashed lines demarcate the retinal layers where different capillary networks have previously been colocalized and illustrate how individual capillary networks were stratified from OCT and histological images. Although OCT and histology images were acquired from the same eccentricity, the influence of normal anatomical variation and the effects of postmortem changes in the histological specimen has resulted in some discrepancies in the dimensions of retinal layer thicknesses between the two images. Orange dashed lines indicate NFL; red dashed lines, RGC capillary network; yellow dashed lines, capillary network at IPL/sINL border; green dashed lines, capillary network at the dINL/OPL border. Scale bar: 50 μm.

Analyses of OCT volumes were performed using saved data, which were processed into intensity and sv images using the GPU-accelerated program.³⁴ The inner limiting membrane (ILM) and Bruch's membrane/retinal pigment epithelium (BM/RPE) complex were automatically segmented based on the intensity data by using a three-dimensional Graph Cut-based algorithm tool implemented in Matlab.⁴² The quality of the segmentation and absence of artifacts was confirmed visually by one of the authors (JX). Given the small scan area and distance from the ONH, the retinal layers were reasonably uniform in thickness for each individual volume. Any curvature or tilt in the retinal surface was captured by the segmentation algorithm. The retinal vascular layers were semiautomatically delineated for each volume by applying a manually determined offset to the Graph Cut segmentation of the BM/RPE surface. The appropriate offset for each vascular layer boundary was determined by visualizing the placement of the segmenting surface as an overlay on an intersecting OCT intensity B-scan image selected near the middle of the volume. The semiautomatic delineations of the vascular layers on a representative OCT B-scan are shown in Figure 1, with the enumerated list above corresponding to the regions bounded by the following colors. The en face images of the four capillary networks were generated by projecting the svOCT-processed data within the depth region selective for a particular vascular layer. The value of the offsets used to define the boundaries was fine-tuned based on the qualitative appearance of the en face svOCT images. En face image data were used for direct comparison of the vascular networks between svOCT and histology. 

Image Analysis

All images were processed and analyzed using ImagePro Plus (version 7.1; Media Cybernetics, Rockville, MD, USA) and Image J (version 1.43; available in the public domain at http:/rsb.info.nih.gov/ij by the National Institutes of Health, Bethesda, MD, USA) software. Images in this report were prepared using Photoshop (version 12.1; Adobe Systems, Inc., San Jose, CA, USA) and Illustrator CS5 (version 12.1.0, Adobe Systems, Inc.). Images were pseudocolored with ImagePro Plus. Z-projection of images that comprised each of the four capillary networks was performed prior to quantitative analysis. Images were also inverted prior to measurement to allow better identification of capillary margins. Measurements were performed as follows (Figure 2). 

Figure 2

 

Quantitative capillary network measurements. Representative histological image (A) and svOCT image (B) of the deep inner nuclear layer and outer plexiform layer capillary network illustrate how capillary diameter and number of vessels per 100 μm were determined. Each image was first inverted (C, D) to allow easier identification of vessel margins. A grid was then placed on each image to partition it into nine equal segments (yellow lines). The perpendicular distance across the maximum chord axis of each vessel was used to measure capillary diameter in each segment (green line). Number of capillary intersections per 100 μm was determined by counting the total number of vessels intersecting the yellow line, dividing by 636.5 (the total length and width of each image in micrometers), and then multiplying by 100. Intercapillary distance was derived from this calculation. Scale bar: 100 μm.

Capillary Diameter.

Defined as the perpendicular distance across the maximum chord axis of each vessel. Each image was partitioned into 9 equal areas, and 5 measurements were acquired from each area to ensure representative sampling. Capillary diameter measurements were acquired from regions of the image where capillary margins were clearly defined. Care was taken to ensure that, whenever possible, capillary diameter measurements were acquired from all regions of each square within the grid. This step avoided oversampling of one particular region. A total of 2,160 diameter measurements (540 per layer) were taken from the 12 donor eyes, and a total of 2,520 diameter measurements (630 per layer) were taken from the 14 svOCT patient eyes. 

Capillary Density.

Density was measured using two indices: number of vessels per 100 μm and intercapillary distance.⁴³ Our measurements were performed on ×20 objective magnification confocal images, collected using the 1,024 by 1,024 pixel resolution. As a result, the settings on our confocal system would scan a square measuring 636.5 × 636.5 μm when collecting the images. The number of vessels per 100 μm and the intercapillary distances from the vessel intersections were calculated using the following formulae:  

 

 

 

where length is 636.5 μm.  

A total of 1,316 density measurements were taken from 12 donor eyes, and a total of 1829 density measurements were taken from 14 svOCT patient eyes. 

Manual tracing (Fig. 3) was performed in 5 eyes from svOCT images and 5 eyes that were imaged with confocal scanning laser microscopy. Capillary density was determined by calculating the percentage of the sample area occupied by vessel lumens. Results were compared between svOCT and histological images. Manual tracings were performed only on the NFL capillary network and the dINL/OPL capillary network because these networks were previously shown to be planar, with a relatively one-dimensional trajectory. The morphologies of the other networks are three-dimensional and are therefore more challenging to manually trace. Results of the comparisons between svOCT and histology, using manual tracing techniques, were compared with the results of capillary density using the indices described in point 2 above. This analysis allowed us to determine if the number of vessels per 100 μm and intercapillary distances were reliable indices for comparing capillary density between histology and svOCT. 

Figure 3

 

Capillary density determined by manual tracing. A representative histological image (A) and svOCT image (B) of the deep inner nuclear layer and outer plexiform layer capillary network and their respective manually traced images (C, D) illustrate how capillary density was calculated. Following manual tracing, the area occupied by capillaries was expressed as a percentage of the total area of tissue. In histological images, the capillary lumens demonstrated a smooth contour whereas nonuniform varicosities (arrowheads) were seen in svOCT images. Scale bar: 100 μm.

Statistical Analysis

Sigmastat version 3.1 software (SPSS, Chicago, IL, USA) was used to calculate all data in terms of means and standard errors. R software (R Foundation for Statistical Computing, Vienna, Austria)⁴⁴ was used to analyze multiple measurements from single eyes as well as measurements from right and left eyes of the same individual. We performed analysis of variance (ANOVA) to determine differences between various capillary networks (NFL, RGC, IPL/sINL, and dINL/OPL) and also between histology and svOCT eyes. We used linear mixed modeling to examine differences in retinal layer measurements. Statistical significance was defined as a P value of <0.05. 

Results

General

The mean donor age of the control histology group was 39.71 ± 3.68 years; measurements were made from 9 left and 3 right eyes of 1 female and 9 male donors. The average postmortem time before eyes were perfused was 14.80 ± 1.68 hours. 

The mean age of the svOCT research volunteer group was 45.58 ± 5.30 years. We imaged 14 eyes from 1 female and 6 male subjects with no history of eye disease. There was no statistically significant difference in age between the two groups (P = 0.455). 

Capillary Network Topography in Histology and svOCT Images

Figure 4 illustrates the highly comparable morphological appearance of capillary networks between histological and svOCT images. As noted in our previous work,⁸ capillaries in the NFL network were largely linear in organization and ran parallel to the direction of RGC axons in the NFL. Beneath the NFL capillary network was the RGC network, where not only capillaries but also arterioles and venules were seen. The IPL/sINL capillary network was three-dimensional in structure and was observably similar in density to the NFL capillary network. The deepest network was located at the level of the dINL/OPL and was the most easily distinguishable layer in histology and svOCT images. This layer was observed to have multiple closed loops and was also laminar in structure. Overall, we found that svOCT images had more background noise than that which was apparent on histology. 

Figure 4

 

Comparison of capillary network morphometry between histology and svOCT. Histology images (left) and svOCT images (right) show comparable nerve fiber layer capillary network morphological features (A, B), the retinal ganglion cell layer capillary network (C, D), the capillary network at the border of the inner plexiform layer and superficial boundary of the inner nuclear layer (E, F), and capillary network at the boundary of the deep inner nuclear layer (G) and outer plexiform layer (H). Scale bar: 100 μm.

Quantitative Analysis of Capillary Diameter

The average capillary diameter for all networks in the histology and svOCT data was 8.26 ± 0.03 μm (n = 2160) and 8.80 ± 0.04 μm (n = 2520) respectively. Mean capillary diameter values for each layer in svOCT and histology images are shown in the Table. Within the histology images, capillary diameters were significantly different among networks (P < 0.001). Post hoc analysis revealed that the mean diameter was significantly different among all networks (P = < 0.001), except for the NFL and dINL/OPL networks (P = 0.227) and RGC and IPL/sINL network (P = 0.740). Within the svOCT images, capillary diameters were significantly different between all networks (all P < 0.014), except for the NFL and IPL/sINL networks (P = 0.426). 

Table

 

Quantitative Capillary Diameter and Density Measurements for Histology and svOCT Images

 

Comparisons between histology and svOCT data revealed no significant differences in capillary diameters in the NFL network (P = 0.891), the IPL/sINL networks (P = 0.151), and the dINL/OPL networks (P = 0.168). There were significant differences in capillary diameter in the RGC network between histology and svOCT images (P < 0.001). Capillary diameter was greater in svOCT images. 

Quantitative Analysis of Capillary Density

Mean values for the indices used to measure capillary density are provided in the Table for both svOCT and histology images. Within histology images, the intercapillary distance and vessels per 100 μm were significantly different between all networks (all P < 0.006), except between NFL and IPL/sINL networks (all P > 0.137) and RGC and dINL/OPL networks (all P > 0.368). Within the svOCT images, these capillary density indices were also significantly different among all networks (all P < 0.045), except between NFL and IPL/sINL networks (all P > 0.394). 

Density comparisons between histology and svOCT eyes demonstrated no differences in the RGC network for these indices (P = 0.497). For the remaining capillary networks, there were significant differences between svOCT and histology images for all of these indices (all P < 0.012). Density was greater in svOCT images. 

Results of Manual Tracing

The NFL capillary network demonstrated a mean density of 10.20 ± 0.02% in the histology images and 16.97 ± 0.02% in the svOCT images. These density measurements were significantly different between the two imaging modalities (P = 0.044). The dINL/OPL networks demonstrated a mean density of 16.04 ± 0.01% in the histology images and 25.61 ± 0.01% in the svOCT images. These density measurements were also significantly different between the two imaging modalities (P = 0.006). 

Discussion

The major findings of this study are as follows. (1) The morphological characteristics of human retinal capillary networks seen on svOCT imaging are comparable to what has previously been demonstrated histologically and is consistent with published reports that have used other OCT-based angiography techniques.^45–47 (2) With the exception of the RGC capillary network, there were no differences in mean capillary diameter measurements derived from svOCT and histology images for any network. (3) Capillary density measurements are significantly greater in svOCT images than those of histology for all networks, with the exception of the RGC network. 

The retina is particularly vulnerable to ischemic injury due to the absence of collateral circulation.^48,49 We have previously shown that, in the presence of systemic cardiovascular disease, morphological changes to capillary networks precede functional visual changes.¹⁵ An in depth understanding of the anatomy of human retinal capillary networks and the ability to detect change in vivo may therefore potentially aid in the clinical management of retinal vascular disorders. The value of fluorescein angiography, magnetic resonance imaging, and adaptive optics techniques to stratify the retinal circulation into various capillary networks remains unclarified.^16,50,51 svOCT is a noninvasive, noncontact technique that is able to provide angiographic information without the administration of contrast. 

The customized svOCT device detailed in this report has the capability of providing both structural and functional information simultaneously in real time and may therefore have many potential clinical applications in ophthalmology. svOCT detects the presence of motion based on the changes of the speckle patterns in OCT images. Acquisition of the three B-scans required to generate a single svOCT depth profile requires only ∼ 0.01 second, which is fast enough to be largely free of artifact. 

We used a fundus location 3 mm superior to the optic disk to correlate svOCT images to histology because the microscopic characteristics of different capillary networks in this region have previously been quantified.⁸ We found that svOCT was able to reliably stratify the retinal capillary circulation into the NFL network, RGC network, the capillary network bordering the IPL and superficial border of the INL, and the capillary network bordering the OPL and deep boundary of INL. Similar to our previous histological study, we observed that the innermost and outermost capillary networks on svOCT demonstrated a laminar configuration whereas IPL and deep INL networks demonstrated a complex three-dimensional configuration. 

When quantitative comparisons were made between svOCT and histology we found no differences in mean capillary diameter values for all networks, with the exception of the RGC capillary network. The mean diameter in svOCT images for the latter network was larger. Additionally, we observed the contour of capillary lumens to be different between svOCT and histology images. In histological images, the capillary walls were smooth and undulating, whereas the appearance of the capillary thickness in the svOCT images in all networks demonstrated nonuniform varicosities. The width of a capillary in the svOCT image is sensitive to the blood cells at the particular vessel location being imaged within the time length of the three B-scans used to calculate sv (∼0.01 second). We speculate that the irregular morphological appearance of capillaries in svOCT images reflect pulsatile flow and the non-uniform distribution of erythrocytes and leukocytes along the capillary length. Previous modeling studies have shown that red blood cells within a capillary bed travel in groups and that the velocities of red blood cells are not constant.⁵² It may also reflect regional pericyte constrictions within the length of a capillary segment.⁵³ A closer study of the distribution of the capillary varicosities on svOCT images may aid our understanding of mechanisms that control oxygen and nutrient delivery in the human retina. 

Comparisons between svOCT and histology yielded similar results for capillary density if manual tracing techniques or the indices of capillary density were used. This suggests that the number of vessels per 100 μm and intercapillary distances were reliable indices of capillary density. There were significant differences in capillary density measurements between svOCT images and histology for all networks, with the exception of the RGC network. Density measurements in svOCT images were consistently greater than histology. The resolving power of the svOCT is not as high as the images of donor retinal specimens acquired with the confocal scanning laser microscope. Although the coherence length of the OCT system is ∼6 μm, shadow artifacts due to blood vessels in the anterior planes may affect the resolution realized in the svOCT images. Some of the vessels visualized in the svOCT images in the deeper layers may therefore be due to these artifacts. This could potentially have resulted in some degree of double counting during quantitative analysis and may partly explain the increased density measurements in svOCT images. However, the influence of shadow artifact on the NFL is expected to be negligible as there are no capillary structures anterior to it. As svOCT density measurements were greater than histology in this network, it suggests that the difference is likely to be a true finding rather than one due to artifact. Additionally, difference in density measurements between svOCT and histology may also have been influenced by the limited sample size of the study. Previous work has shown that there is some degree of interindividual variation in retinal layer thickness measurements among normal human eyes.⁵⁴ Variations in retinal layer thickness measurements are likely to translate into variations in capillary density measurements among normal human eyes. With a larger sample size, it is possible that the differences between histology and svOCT may become insignificant. 

This study demonstrates that svOCT has the potential to provide in vivo, histology-like information about the different capillary networks in the human retina. It could therefore serve as a powerful tool for studying retinal vascular diseases. The capacity of this device to produce images in real-time is also particularly advantageous for patient assessment in the clinical setting. A major advantage of this study is that we used perfusion-labeling techniques of postmortem samples to ensure complete histological identification of the retinal circulation. We have previously shown that this technique ensures precise labeling of the capillary bed and therefore serves as a “gold standard” for comparing in vivo imaging technology.³⁶ However, we acknowledge several limitations of this study. The volume acquisition time with the current svOCT prototype is longer than the acquisition time of regular intensity-only volumes with current commercially available spectral domain OCT devices. Consequently, motion artifact may degrade the quality of capillary images in those patients who are unable to reliably fixate. During the acquisition of the volume, slight eye motions can also contribute to spatial distortions along the slow axis. Additionally we observed greater background noise in svOCT images than in histology, and this may potentially induce image artifact or obscure fine capillary detail in some networks. It will be important to resolve the reason for the appearance of varicosities along the capillary length in svOCT images as this may provide vital information about retinal vascular homeostasis. Investigating whether svOCT has the capacity to detect changes in retinal blood flow may also have great relevance to clinical ophthalmology. 

Acknowledgments

The authors thank the staff of the Lions Eye Bank of Western Australia, Lions Eye Institute, for providing human donor eyes; the staff of DonateWest, the Western Australian agency for organ and tissue donation, who facilitated the recruitment of donors into the study by referral and completion of consent processes; and Dean Darcey for expert technical assistance. 

Supported by National Health and Medical Research Council of Australia, the Sir Charles Gairdner Hospital Clinical Staff Education Fund, and in part by the Michael Smith Foundation for Medical Research, the Canadian Institutes of Health Research, the Natural Sciences and Engineering Research Council of Canada, and Brain Canada. 

Disclosure: P.E.Z. Tan, None; C. Balaratnasingam, None; J. Xu, None; Z. Mammo, None; S.X. Han, None; P. Mackenzie, None; A.W. Kirker, None; D. Albiani, None; A.B. Merkur, None; M.V. Sarunic, None; D.-Y. Yu, None