Detailed Vascular Anatomy of the Human Retina by Projection-Resolved Optical Coherence Tomography Angiography — TEI-tagged view

J. P. CampbellpersName, M. ZhangpersName, T. S. HwangpersName, S. T. BaileypersName, D. J. WilsonpersName, Y. JiapersName, D. HuangpersName (2017date). “Detailed Vascular Anatomy of the Human Retina by Projection-Resolved Optical Coherence Tomography Angiography.” Scientific Reportstitle·j.

Object of study — the retinal vascular plexusesterm·key of normal human eyes imaged by projection-resolved OCTAterm·mth

Research question — how to discriminate retinal vessel layers in depth despite flow projection artifactterm·dom

Contribution — a nomenclatureterm·key and segmentationterm·key scheme for 3-dimensional retinal vascular anatomy by OCTAterm·mth

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

Detailed Vascular Anatomy of the Human Retina by Projection-Resolved Optical Coherence Tomography Angiographyterm·mth Abstract Optical coherence tomography angiographyterm·mth (OCTAterm·mth) is a noninvasive method of 3D imaging of the retinal and choroidal circulations. However, vascular depth discrimination is limited by superficial vessels projecting flow signal artifact onto deeper layers. The projection-resolved (PR) OCTAterm·mth algorithm improves depth resolution by removing projection artifactterm·key while retaining in-situ flow signal from real blood vessels in deeper layers. This novel technology allowed us to study the normal retinal vasculatureterm·key in vivo with better depth resolution than previously possible. Our investigation in normal human volunteers revealed the presence of 2 to 4 distinct vascular plexusesterm·key in the retina, depending on location relative to the optic discterm·key and foveaterm·key. The vascular pattern in these retinal plexuses and interconnecting layers are consistent with previous histologic studies. Based on these data, we propose an improved system of nomenclatureterm·key and segmentationterm·key boundaries for detailed 3-dimensional retinal vascular anatomy by OCTAterm·mth. This could serve as a basis for future investigation of both normal retinal anatomy, as well as vascular malformations, nonperfusion, and neovascularization.

Our current understanding of the retinal vascular networks developed from pioneering work on primate histology 1 2 . From these early studies, we know that there are up to four retinal vascular networks in the macula ( Fig. 1 ) 3 . The superficial vascular plexusterm·dom (SVPterm·dom) is supplied by the central retinal artery and composed of larger arteries, arterioles, capillaries, venules, and veins vessels primarily in the ganglion cell layerterm·dom (GCLterm·dom). There are two deeper capillary networks above and below the inner nuclear layerterm·dom (INLterm·dom) referred to as the “intermediate” and “deep” capillary plexuses, or ICPterm·dom and DCPterm·dom, respectively, which are supplied by vertical anastomoses from the SVPterm·dom 1 2 . The fourth network is a regional layer called the radial peripapillary capillary plexusterm·dom (RPCPterm·dom). The radial peripapillary capillaries have a unique anatomic organization because they run in parallel with the NFLterm·dom axons, as opposed to the deeper vascular plexusesterm·key, which have a lobular configuration 4 5 . The functional significance of the RPCPterm·dom has been recognized due to its role supplying the densely packed nerve fiber layerterm·dom (NFLterm·dom) bundles in this region 4 5 6 . This organization of the retinal vascular plexusesterm·key has been confirmed in humans ex vivo using confocal microscopy 7 8 and in vivo using speckle variance OCTterm·mth 9 and adaptive optics confocal scanning laser ophthalmoscopy (AO-cSLO) 10 in research studies, technologies with a very narrow field of view and which are not widely available. Current clinical imaging of the ocular circulation has been dominated by fluorescein angiographyterm·mth (FAterm·mth), a 2D imaging modality that was developed in the 1950’s that captures the fluorescence signal from intravenously injected dye in the retinal and choroidal circulation, and many retinal conditions are defined by their characteristic FAterm·mth patterns. FAterm·mth has the ability to visualize breakdown of the blood-retinal barrier (with the finding of “leakage” of extravasated fluorescein), but is limited in its ability to resolve the depth of specific signal abnormalities, and cannot visualize the deeper retinal capillary layers (ICPterm·dom and DCPterm·dom) due to blocked fluorescence 11 12 .

Optical coherence tomographyterm·mth (OCTterm·mth) is a non-invasive imaging technology that has the high speed required for 3D volumetric imaging and the high spatial resolution to visualize individual layers in the retina. OCTterm·mth has become routine in the evaluation of retinal and optic nerve diseases, and is the most commonly used imaging modality in ophthalmology 13 . In a recently developed extension of OCTterm·mth, OCTterm·mth angiography (OCTAterm·mth) detects blood flow down to the capillary level by measuring change (decorrelation or variance) in OCTterm·mth signal in consecutive cross-sectional images (B-frames) taken at the same location. Because OCTAterm·mth uses motion as intrinsic contrast, extrinsic contrast such as intravenous injection of fluorescein dye is not needed, making this new modality more practical for routine clinical use. Since its clinical introduction, OCTAterm·mth has been used for en face visualization of blood flow in anatomic slabs, so that the retinal circulation can be displayed separately from the choroidal circulation, and the vascular anatomy can be further subdivided into more superficial and deeper retinal vascular plexusesterm·key, typically called the superficial and deep plexuses 14 15 . However, these early attempts to separate the retinal circulation into superficial and deep plexuses has been marred by segmentationterm·key boundaries that imperfectly segment the intermediate capillary plexusterm·dom ( Fig. 1 ), and flow projection artifactterm·dom 16 17 18 .

Flow projection artifactterm·dom is caused by the flickering shadowterm·dom from superficial blood flow being projected onto deeper tissue layers. Since in-situ flow and its shadow both produce OCTterm·mth signal variation, the OCTAterm·mth algorithms interpret them both as flow. Due to the projection artifactterm·key, superficial vessels acquire a long tail downward (also called tailing artifactterm·dom) on cross-sectional OCTAterm·mth. On en face OCTAterm·mth, the vascular patterns from more superficial plexuses are duplicated on the deeper layers, making it impossible to obtain a clean image of the deeper vascular plexusesterm·key, or to separately visualize the ICPterm·dom and the DCPterm·dom, and demonstrating signal in the normally avascular outer retina. This has complicated early attempts at using OCTAterm·mth to identify pathological neovascularization in (or under) the outer retina 19 , and quantify nonperfusion in the deeper layers 20 .

We previously reported a novel algorithm called “projection-resolved” OCTAterm·mth (PR-OCTAterm·key) 21 . Our hypothesis was that this algorithm would allow visualization of the four human retinal vascular plexusesterm·key known from histology, and allow quantitative comparison of vessel densities at each location in the retina, both in depth (within the layers of the retina), and in location between the optic nerve and the retinal periphery. We proposed to resolve the flow projection artifactterm·dom limiting existing technology, produce clean en face OCTAterm·mth of the deeper retinal capillary plexuses, and describe the inter-plexus anatomy 18 . Using this technology to characterize the retinal vascular plexusesterm·key in vivo , we propose a standard nomenclatureterm·key for these vascular networks, and provide guidance for rational segmentationterm·key algorithms, which respect the normal distribution of vascular networks in the human retina.

Results Human Subjects We examined one eye from each of nine human subjects, including 3 women and 6 men, 3 left eyes and 6 right eyes. The mean age was 30.2 (standard deviation [SD] 7.4). The mean refractive error was −1.3 diopters (SD 1.7). 4 of the subjects were Asian, and 5 were Caucasian. The mean systolic blood pressure was 113 mmHg (standard deviation 4), and none of the subjects had any medical or ocular comorbidities, and were on no topical or systemic medications.

Anatomic localization of vascular plexusesterm·key Using the PR-OCTAterm·key algorithm, we were able to detect the depth of vascular plexusesterm·key in the retina in four geographic regions temporal to the optic nerve (peripapillary, parafoveal, perifoveal, peripheral) by identifying depth-specific peaks in vessel densityterm·dom for each subject. This montage OCTAterm·mth ( Fig. 2 ) from one subject displays decorrelation signals from a single 2 dimensional image demonstrating the typical angiographic appearance of the inner retina with both large and small vessels. Overlaying the decorrelation signal onto a structural OCTterm·mth image ( Fig. 2 ), after application of the PR-OCTAterm·key algorithm, the specific depth and relative breadths of the four vascular plexusesterm·key are easily visualized.

Depth-resolved capillary densityterm·key profiles ( Fig. 3 ) were obtained in 4 anatomic areas. In the peripapillary region, the RPCPterm·dom and the SVPterm·dom demonstrate two peaks of capillary densityterm·key, but with no distinct trough separating them. In the parafoveal and perifoveal regions, there are three distinct peaks corresponding to the three vascular plexusesterm·key (SVPterm·dom, ICPterm·dom, and DCPterm·dom). There are 2 distinct troughs separating the 3 plexuses in the central macula ( Fig. 3 ), and the 3 plexuses could be visualized as distinct layers in cross-sectional OCTAterm·mth vessel densityterm·dom map ( Fig. 4 ). The 3 plexuses merge into one at the edge of the foveal avascular zoneterm·dom (FAZterm·dom). In the peripheral region, the ICPterm·dom and the DCPterm·dom merge into a single peak ( Fig. 3 ). The merged parafoveal capillary ring can be visualized in the en face OCTAterm·mth of all 3 plexuses ( Fig. 5 ). On the cross-sectional capillary densityterm·key map ( Fig. 4 ), the ICPterm·dom and DCPterm·dom can be seen to merge into a single layer approximately 6–7 mm temporal to the foveaterm·key.

En face visualization of vascular plexusesterm·key Using the depth locations of the retinal vascular plexusesterm·key defined by the investigation above ( Fig. 3 ), we produced en face OCTAterm·mth of the four vascular plexusesterm·key ( Fig. 5 ). Although the RPCPterm·dom and the SVPterm·dom appear to be a single complex in cross section ( Fig. 4 ), they present distinct patterns en face ( Fig. 5 ). The RPCPterm·dom has long radial capillaries that run parallel with the nerve fiber bundles, whereas the SVPterm·dom consists of a mixture of large and small vessels ( Fig. 5 ). In the macula, the SVPterm·dom has a pattern of centripetally branching pattern that terminate at capillary ring around the foveal avascular zoneterm·dom ( Fig. 5 ). The largest arcade vessels interweave between the NFLterm·dom and GCLterm·dom, and thus are captured in both the RPCPterm·dom and SVPterm·dom slabs. The ICPterm·dom and DCPterm·dom consist of thin layers of capillaries arranged in lobular patterns without directional preference.

En face visualization of inter-plexus layers In the parafoveal and perifoveal retina, the retinal layers are sufficiently thick for PR-OCTAterm·key to resolve not only the 3 distinct vascular plexusesterm·key, but also the capillary-free interconnecting layers between them. This is demonstrated in Fig. 6 using a high resolution 2 mm × 2 mm scan in the temporal macula. In these inter-plexus spaces, some interconnecting vascular segments can be observed. Comparison with a color fundus photograph showed that these interconnecting vessels arise from terminal arterioles and venules in the SVPterm·dom. The diving venules are seen more clearly than diving arteriole in the ICPterm·dom, DCPterm·dom, and interconnecting layers ( Fig. 6 ). In some instances, these diving venules are seen to arise from a radiating network of capillaries in the ICPterm·dom and DCPterm·dom.

Regional vessel densityterm·dom variation in the retinal plexuses The densities of the retinal vascular plexusesterm·key show characteristic regional variations ( Fig. 7 , Table 1 ). The ICPterm·dom and DCPterm·dom have roughly constant density in the peripapillary and macular regions ( Table 1 ) except for the focal absence in the foveaterm·key ( Fig. 7 ). They merge into a single layer in the periphery with slightly higher combined density than the single plexuses. The SVPterm·dom density is highest in the peripapillary and macular regions (except for the foveal area) and decreases in the periphery (P < 0.05 peripheral v. peripapillary, parafoveal, or perifoveal density). In the periphery, the density of the merged ICPterm·dom/DCPterm·dom plexus is higher than the SVPterm·dom (P < 0.05). In the peripapillary, parafoveal, and perifoveal locations ( Table 1 ), the peak vessel densityterm·dom of the SVPterm·dom is higher than the ICPterm·dom and DCPterm·dom (P < 0.05). The RPCPterm·dom density varies dramatically with location. It is densest near the optic discterm·key edge and decreases outward ( Fig. 7 ). It also appears to range further from the disc along the arcuate nerve fiber bundles ( Fig. 5 ), but the limited imaging area in this study do not permit complete characterization of its distribution. The intravisit repeatability of vessel densityterm·dom measurements was 1.9% for the RPCPterm·dom, 2% for the SVPterm·dom, 1.7% for ICPterm·dom, and 3.3% for the DCPterm·dom. There were no detectable associations of peak vessel densityterm·dom (parameters shown in Table 1 ) with age, gender, or refractive error in this small population using multivariable linear regression.

Discussion The recent clinical introduction of OCTAterm·mth enabled 3-dimensional visualization of retinal and choroidal circulations that was not possible with 2-dimensional FAterm·mth images. Within the retina, the FAterm·mth image is dominated by the SVPterm·dom and the choroidal signal, with poor visualization of the ICPterm·dom and DCPterm·dom 12 22 23 . OCTAterm·mth en face images have been used to visualize vascular networks within the retina at various depth layer-by-layer using thin slabs 15 . Using this approach, 2 different vascular patterns were initially distinguished, and they were referred to as the superficial and deep plexuses based on their relative depth and vascular patterns by Lumbroso and others ( Figs 1 and 4 ) 11 15 . Spaide et al . further noted a radially oriented vascular network could be visualized near the optic discterm·key, corresponding to the radial peripapillary capillaries previously known from histology, but not well visualized by FAterm·mth 11 . These observations led to many early studies with OCTAterm·mth in both normal and diseased eyes 24 , however these analyses have been significantly limited due to flow projection artifactterm·dom 16 23 , which prevented: (1) the ability to separate the ICPterm·dom from DCPterm·dom, (2) precise visualization of abnormal neovascularization in avascular areas of the retina (as in macular degeneration), (3) accurate quantification of nonperfusion in normally vascularized areas of the retina (as in diabetes), and (4) visualization of the inter-plexus spaces. Indeed, with conventional OCTAterm·mth all retinal layers, including the avascular outer retina, appeared to contain flow signal, and no natural boundary could be defined between vascular plexusesterm·key. Additionally, due to flow projection artifactterm·dom, the true plexus boundaries between the superficial and deeper vasculature could not be well characterized, which led to segmentationterm·key at the INLterm·dom/IPL border on SD-OCT (as shown in Figs 1 and 4 ). Unfortunately, this anatomic boundary yields unpredictable segmentationterm·key of the ICPterm·dom between the superficial and deep slabs.

The PR-OCTAterm·key algorithm provides a new and more accurate way to investigate detailed retinal vascular anatomy in humans in vivo 18 . In this article, we used this new technology to advance retinal vascular imaging in several significant ways. First, we demonstrate that PR-OCTAterm·key can visualize four unique vascular plexusesterm·key in the human retina with distinct vascular patterns, which vary based on depth and location from the nerve. Secondly, using PR-OCTAterm·key one can visualize the inter-plexus spaces and the interconnecting vascular segments in vivo . Thirdly, PR-OCTAterm·key preserves in situ flow signal in deeper vessels, resulting in visualization of continuous capillary networks in the ICPterm·dom and DCPterm·dom. This was not possible with earlier attempts at projection suppression using slab-subtraction algorithms, which produced numerous gaps that disrupted the continuities of vascular networks deeper layers 16 21 25 . Finally, we were able to perform quantitative analysis of capillary densityterm·key using PR-OCTAterm·key, and establish the relative and absolute troughs in vessel densityterm·dom that form natural boundaries between retinal vascular plexusesterm·key. These boundaries were defined in terms of retinal anatomic layers that could be easily segmented from structural OCTterm·mth images. Based on what could be delineated on PR-OCTAterm·key, we propose a system of terminology for the retinal circulation ( Figs 1 , 4 and 5 ) that we hope will be useful for future clinical studies. These definitions are consistent with the detailed anatomy known from histologic studies 2 4 12 , but are more detailed than what was used in previous OCTAterm·mth literature. This PR-OCTAterm·key vascular anatomic scheme is described below.

The superficial vascular complexterm·dom consists of the SVPterm·dom and the RPCPterm·dom. The SVPterm·dom is a network of both large and small vessels connected directly to the retinal arteries and veins, and supplies all other vascular plexusesterm·key. In the macula, the SVPterm·dom is arranged in a centripetal pattern that converges on the parafoveal capillary ring ( Fig. 5 ). In the periphery, the superior and inferior circulations converge in an interleaved comb pattern similar to the retinal vasculatureterm·key shown by FAterm·mth 12 23 . Primate histology demonstrated that the SVPterm·dom supplies all other vascular plexusesterm·key through vertical pre-capillary arterial segments, which typically ascend to the NFLterm·dom (in the case of the RPCPterm·dom) and descend to the deeper layers (ICPterm·dom and DCPterm·dom) off of the 1 st order branches from the main arcades and terminal extensions of the SVPterm·dom arterial segments 2 4 . The post-capillary segments often drain directly into the main venous arcades 3 . We can now visualize these vascular networks using PR-OCTAterm·key including the interconnecting vessels and the laminar networks in the ICPterm·dom and DCPterm·dom ( Fig. 6 ). The SVPterm·dom vessel densityterm·dom decreases as a function of distance from the optic nerve ( Fig. 7 ) as the surrounding GCLterm·dom decreases in thickness ( Fig. 2 ). These results correlate well with what is known from primate histology 2 .

The RPCPterm·dom represents a unique capillary plexus in the peripapillary region with three defining characteristics from the original histology, which can be uniquely visualized within the SVC by segmenting on the NFLterm·dom/GCLterm·dom interface 4 5 . First, as opposed to the deeper SVPterm·dom capillary plexus, the RPC travel along relatively longer and straighter paths than other plexuses ( Fig. 5 ) 4 . Second, they have fewer anastomoses within the plane of the RPCPterm·dom ( Fig. 5 ) but are fed by pre-capillary arterial segments from the SVPterm·dom and drained by post-capillary venous segments into the SVPterm·dom. Third, though in these data we focused analysis on the maculopapillary axis, histology demonstrates that these vessels are located only in the posterior pole where they run alongside the dense NFLterm·dom axons in an asymmetric butterfly distribution around the optic nerve and the vascular arcades 4 . Prior work has suggested that this layer nourishes the NFLterm·dom, and therefore is densest in the peripapillary region and reduces in thickness as the NFLterm·dom decreases, which is consistent with our findings ( Figs 5 and 7 ) 2 4 5 .

As a whole, the SVC presents an extremely dense vascular network in the peripapillary region ( Fig. 5 ) that decreases in density with distance from disc along the maculopapillary axis ( Fig. 7 ). This agrees with previous histologic evidence – Snodderly et al . reported the percentage of retinal area covered by capillaries and found that it was highest in the peripapillary region, often exceeding 60% 2 .

The DVCterm·dom consists of the ICPterm·dom above the INLterm·dom and the DCPterm·dom below the INLterm·dom. The capillaries of the ICPterm·dom and DCPterm·dom have a lobular configuration with no directional preference within their laminar planes. They consist of capillaries of uniform size ( Fig. 5 ), with the exception of larger vessels that interconnect the plexuses ( Fig. 6 ). From primate histology we know that these plexuses represent terminal anastomotic capillary networks supplied by vertically oriented interconnecting arteries and veins from the SVPterm·dom 3 , which we can now demonstrate ( Fig. 6 ). Confocal microscopy in ex vivo specimens has demonstrated that the ICPterm·dom is situated among bipolar cell processes, and the DCPterm·dom localizes to the region near the horizontal cells in the outer INLterm·dom 26 , presumably due to local metabolic needs in the retina 3 . In addition, we demonstrated that these plexuses have relatively constant capillary densityterm·key from the optic nerve to the periphery along the maculopapillary axis ( Fig. 7 ), and merge approximately 6–7 mm temporal to the foveaterm·key, two novel findings that are facilitated by the wider field of view of OCTAterm·mth compared to prior technologies 9 10 and the improved in situ visualization of flow using PR-OCTAterm·key.

We have developed a system for segmenting the retinal circulation into 2 complexes and 4 plexuses using anatomic layers that could be delineated on structural OCTterm·mth ( Fig. 5 ). The appropriate segmentationterm·key scheme to be used for the display and quantitative analysis of en face PR-OCTAterm·key depends on the anatomic region. In the peripheral retina, the SVC and DVCterm·dom are the 2 distinguishable layers. In the macula, the DVCterm·dom can be divided into the ICPterm·dom and the DCPterm·dom to further improve the detection of capillary dropout and other vascular pathologies. In the peripapillary region, the SVC can be divided into the RPCPterm·dom and the SVPterm·dom. Around the foveal avascular zoneterm·dom (FAZterm·dom), the retinal plexuses converge to form a single parafoveal capillary loop and collectively define the borders of the FAZterm·dom. This is clearly visualized on PR-OCTAterm·key ( Fig. 4 ) in agreement with early histological studies in primates 2 12 , which suggests that early reports highlighting FAZterm·dom area differences between superficial and deep plexuses may not be meaningful. Therefore, we recommend that the FAZterm·dom size be measured using an en face projection that include all retinal plexuses, rather than separating the FAZterm·dom area into superficial and deep FAZterm·dom areas, as has been done in early OCTAterm·mth analyses 23 .

There are a number of limitations to this analysis. First, we report the values for only a small population of subjects. Further studies using PR-OCTAterm·key are needed to establish a normative database of retinal plexus that account for relevant demographic variable such as age, gender, race/ethnicity, and refractive error/axial eye length, in addition to biological variables such as oxygen saturation, caffeine use, and physical activity. Second, the analysis of PR-OCTAterm·key in diseased eyes are beyond the scope of this report. In some diseases, retinal anatomic boundaries may be difficult to identify and prevent detailed analysis of single plexuses.

The retina vascular complexes and plexuses that we have identified and measured with PR-OCTAterm·key have applications in retinal and optic nerve disease. In glaucoma 27 28 29 30 , and other neurodegenerative diseases 24 , structural OCTterm·mth of the peripapillary NFLterm·dom and the macular GCCterm·dom have been helpful in diagnosis and monitoring. Atrophy and thinning of these structures are objective and precisely quantifiable disease indicators. However, retinal edema can confound thickness measurements, and in advanced glaucoma these structures are dominated by residual glial tissue and no longer serve as reliable disease indicators 31 . Optical coherence tomography angiographyterm·mth of the vascular supplies to these structure is a promising alternate approach to diagnose and monitor these diseases 6 32 33 34 35 . Using PR-OCTAterm·key, we have clearly identified the location of the RPCPterm·dom in the peripapillary region, and the SVC in the macula. Accurate measurement of vascular perfusion in these plexuses could improve the assessment of glaucoma and other optic nerve diseases.

Using PR-OCTAterm·key to separate the retinal circulation into the SVC, ICPterm·dom, and DCPterm·dom in the macula, and SVC and DVCterm·dom in the periphery, may be useful in the assessment of capillary dropout, microaneurysms, dilated shunt vessels, and other vascular pathologies associated with diabetic retinopathy 20 27 , and other ischemic diseases of the retina. There are also retinal diseases that are hypothesized to affect single plexuses, such as paracentral acute middle maculopathy 28 , and OCTAterm·mth has demonstrated localized nonperfusion in this disease, even in the presence of a normal FAterm·mth 29 30 . Using PR-OCTAterm·key to visualize capillary dropout in the ICPterm·dom and DCPterm·dom, as well abnormal vessels in all retinal layers, could be helpful in the assessment of these diseases. As FAterm·mth transformed our understanding of retinal pathophysiology, OCTAterm·mth may lead to an even deeper understanding and ultimately provide earlier and more specific identification of disease, and better outcomes for patients. By demonstrating the anatomic organization of the retinal vascular plexusesterm·key in normal human eyes, this work using PR-OCTAterm·key provides a rational schema for nomenclatureterm·key and segmentationterm·key of the retinal vasculatureterm·key in future OCTAterm·mth publications.

Materials and Methods Image Acquisition and Processing Participants were recruited from the Casey Eye Institute. The research protocol was approved by the Oregon Health & Science University institutional review board and in compliance with the Declaration of Helsinki and HIPAA regulations. Informed consent was obtained from all human subjects. A commercial spectral-domain 70 kHz OCTterm·mth instrument (RTVue-XR, Optovue) was used to obtain OCTAterm·mth scans in healthy eyes. A commercial version of the split-spectrum amplitude-decorrelation angiographyterm·mth (SSADAterm·mth) algorithm was used to detect blood flow 36 . The measurements were taken with the subject at rest in normal room air in a seated position. Blood pressure measurements were taken and confirmed to be normal. The axial resolution of the OCTterm·mth system was 5 μm (full-width half-maximum) and the transverse beam spot diameter was 15 μm. Split-spectrum processing reduces the axial resolution to 20 μm for SSADAterm·mth flow signal 36 . The axial voxel height was 3.1 μm and each axial ( z ) scan has 512 voxels. Each volumetric scan has 304 × 304 transverse points. Each set of volumetric image data was obtained by orthogonal registration and merging of 1 x -priority and 1 y -priority scans. The OCTAterm·mth data from different participants were registered based on OCTterm·mth structural information. Specifically, the x dimension ( Fig. 2 ) was linearly scaled to register the disc and foveaterm·key center (from the middle of the optic discterm·key to the middle of the foveaterm·key), each major structural layer (ILM + NFLterm·dom, GCLterm·dom + IPL, INLterm·dom, OPL) was also scaled along z dimension ( Fig. 2 ) to register the individual layers of all participants. The subject depicted in Fig. 2 image was used as the reference subject for x and z scaling for all other subjects, and was the youngest subject without refractive error. Thus, the variability in the x and z dimensions was minimized so that we could identify the anatomic location of the capillary plexuses as a function of the structural boundaries, rather than arbitrary depth measurements, which would vary by retinal thickness. The PR-OCTAterm·key algorithm identifies voxels with in-situ flow as those where reflectance-normalized decorrelation values are higher than all shallower voxels in the same axial scan line 25 Fig. 8 provides a brief illustrated explanation of this algorithm. En face and cross-sectional angiograms were obtained by slab projection. Structural OCTterm·mth images were obtained by averaging reflectance signal over the slab thickness 37 . Angiograms were obtained by maximum flow projection over the slab thickness.

Depth-resolved capillary densityterm·key calculation In order to visualize the depth of the retinal vascular plexusesterm·key ( Fig. 3 ), the OCTAterm·mth volumes were aligned at the RPE level. Then flow pixels at each depth ( z ) plane in 0.1 mm × 0.8 mm ( x × y ) mm sample areas were then counted and divided by the total number of pixels to yield a capillary densityterm·key (units = percent flow per area). For the purposes of capillary densityterm·key measurement in all plexuses, the areas occupied by larger vessels (arteries and veins) were excluded from the analysis, by filtering the signal using a signal intensity based thresholding algorithm to create a large vessel mask from the en face angiogram of the SVPterm·dom (using a Gaussian window (20 × 20 pixels), then thresholding at a SSADAterm·mth decorrelation value of 0.12) and keeping connected regions larger than 50 voxels. The same procedure was applied to the entire group. The purpose of this mask was to remove the measurement of the large vessels from estimates of “capillary densityterm·key”. For the cross-sectional capillary densityterm·key map ( Fig. 4 ) each color-coded pixel in the cross section presents the capillary densityterm·key in a super-voxel 0.1 mm wide ( x ) by 0.8 mm deep ( y , perpendicular to the image plane) by 0.01 mm high ( z ) in units of % volume occupied by flow voxels within the super-voxel. The repeatability of this measurement for within-visit intra-subject repeatability was evaluated using the coefficient of variation.

Segmentationterm·key boundaries for en face visualization of vascular plexusesterm·key Using data from the depth-resolved capillary densityterm·key profile ( Fig. 3 ), and comparing to the structural SD-OCT signal, we empirically defined segmentationterm·key boundaries that separated the vascular plexusesterm·key using structural SD-OCT boundaries. These boundaries ( Fig. 5 ) were automatically processed (with manual verification), the same for all patients and were used throughout this article. The superficial vascular plexusterm·dom (SVPterm·dom) slab was defined as the inner 80% of the ganglion cell complexterm·dom (GCCterm·dom), which is commonly used for macular evaluation using SD-OCT (defined as the NFLterm·dom + GCLterm·dom + inner plexiform layer [IPL]). We then isolated the RPCPterm·dom from the SVPterm·dom by segmenting on the NFLterm·dom/GCLterm·dom interface. Thus, the SVC (SVPterm·dom + RPCPterm·dom) consisted of all vessels between the ILM and inner 80% of GCCterm·dom. The intermediate capillary plexusterm·dom (ICPterm·dom) was segmented between the outer 20% of the GCCterm·dom to the inner 50% of the inner nuclear layerterm·dom (INLterm·dom). The deep capillary plexus was segmented between the outer 50% of the INLterm·dom and the outer plexiform layer (OPL). Thus, the DVCterm·dom (ICPterm·dom + DCPterm·dom) is found in the anatomic layers consisting of the outer 20% of the GCCterm·dom (outer IPL) to the OPL. In order to visualize the inter-plexus space, we created single voxel-thick en face OCTAterm·mth images through the inter-plexus spaces in the IPL and INLterm·dom ( Fig. 6 ). The space between the SVPterm·dom and ICPterm·dom was found in the IPL, and the space between the ICPterm·dom and the DCPterm·dom was found in the INLterm·dom.

Data analysis Statistical analysis was performed using Stata version 11.0 (College Station, TX). Peak vessel densities were compared using the t-test, and multivariate linear regression was performed with a significance of P < 0.05 being considered statistically significant.

Additional Information How to cite this article : CampbellpersName, J. P. et al . Detailed Vascular Anatomy of the Human Retina by Projection-Resolved Optical Coherence Tomography Angiographyterm·mth. Sci. Rep. 7 , 42201; doi: 10.1038/srep42201 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.