Pearls and Pitfalls of Optical Coherence Tomography Angiography Imaging: A Review

Structural OCT can acquire volumetric images of the retina by executing successive single B-scans at different retinal locations. Differently, OCTA devices perform several repeated B-scans at the same retinal location and the obtained structural information is compared to detect signal changes secondary to flowing erythrocytes (motion contrast). Furthermore, each B-scan is composed of several A-scans that are acquired at sequential positions to generate the B-scan. The A-scan rate is dependent on the OCT instrument and typically ranges between 70,000 A-scans per second (spectral domain devices) and 100,000 A-scans per second (swept source instruments) among commercially available devices. Of note, faster acquisition speeds are available in various research instruments.

A crucial role in OCTA imaging is represented by the interscan time, which represents the delay between two B-scans repeated at the same retinal location. Therefore, while a shorter interscan time may decrease sensitivity to motion, this also reduces the occurrence of parasitic eye motions which can confound the motion signal [1].

Three different methodologies may be used to detect motion contrast using OCTA devices, as follows: (1) phase-based, (2) amplitude-based, and (3) complex amplitude-based, in which the OCTA algorithm uses both phase and amplitude information. The OCTA signal is normalized between 0 and 1. The interscan time and the background noise determine the slowest detectable flow. Erythrocytes must move a sufficient distance within repeated B-scans to be detected. In addition, differences in signal must be higher than the background noise to be detected as flow. This causes an intrinsic limitation to current OCTA technology which may not be able to distinguish the absence of flow from slow flow below a detectable threshold range [1].

OCTA may be captured with spectral domain OCT (SD OCT), which in commercial devices employs a wavelength at around 840 nm, or with swept-source OCT (SS OCT), which uses a longer wavelength (~ 1050 nm).

While OCT is considered a cross-sectional imaging modality, OCTA images are mainly displayed with en face visualization. Using the segmentation of the volumetric OCTA scans at specific depths, the flow data within any slab are summed or projected into a two-dimensional en face image that can be viewed and studied.

The generation of two-dimensional OCTA en face images may be obtained using different methodologies, including the “maximal intensity projection” approach which projects the maximal intensity pixel from each column onto a flat two-dimensional summary. While the advantage of this strategy is that it has more sensitivity for detecting smaller vessels and subtle flow, it may also yield more noise. Another approach to obtain two-dimensional OCTA images is through use of an “average intensity projection,” which is less susceptible to noise but may be less sensitive for small vessels.

Of note, the en face visualization of the OCTA images is susceptible to segmentation artifacts, especially in eyes with retinal and choroidal disorders [5].

Fluorescein angiography (FA) and indocyanine green angiography (ICGA) have traditionally been considered the gold standard for clinical assessment of the retinal and choroidal vasculature in vivo. Even though dye administration is generally safe in patients, serious allergic reactions may occur, and these techniques are considered invasive and require considerable patient cooperation. In contrast, OCTA provides noninvasive assessment of the retinal and choroidal vasculatures [6]. Moreover, OCTA has the additional advantage of depth resolution with improved visualization of the deeper vascular plexuses. In addition, both FA and ICGA have significant limitations in the assessment of the choroid [2]. Despite these limitations, dye-based imaging provides other advantages including dynamic estimation of dye transit and dye leakage in diseases affecting the blood-retinal barrier (e.g. diabetic retinopathy).

Low signal strength may significantly affect the quality of the OCTA images. The ratio between the signal (signal-to-noise ratio), which is the information component dependent on the imaged tissue, and the non-signal component (or noise), is a mathematical relationship which may influence the image quality. As an example, media opacities may reduce the signal-to-noise ratio by decreasing the OCT signal. A decreased signal-to-noise ratio leads to an adjustment of the grayscale range with a consequent increase in image noise and artifactitious OCTA signal. Averaging of multiple en face OCTA images may reduce noise and improve the vessels’ continuity, as well as significantly improve qualitative and quantitative measurements [7, 8]. False OCTA signals may be also generated by a patient’s movements [9]. Eye tracking systems have significantly decreased the impact of these artifacts in OCTA imaging.

Segmentation errors represent a significant and recurring OCTA artifact [5, 9]. As discussed above, en face OCTA images are displayed by locating two boundaries throughout the retinal and/or choroidal structure and vessels are therefore visualized using different available processes. These boundaries may be set by selecting predefined layers which can be differentiated on the basis of reflectivity, texture, or other attributes. The main problem with this strategy is that these parameters used to differentiate retinal and choroidal layers may be significantly modified in pathological conditions [5]. As an example, the automatic segmentation of vascular layers may be challenging in eyes with myopia, even in the absence of chorioretinal complications. Of note, many OCTA devices employ software enabling manual correction of segmentation errors and propagation of corrections throughout multiple B-scans, but this process may be extremely time-consuming.

Projection artifacts (decorrelation tails; Fig. 2) may affect the visualization of the deeper vascular layers (ICP, DCP, and CC). In order to image the deeper structures, the OCT beam passes through the retinal layers, and vessels included in more superficial layers may be falsely detected in the deeper layers, even in the absence of real flow in these slabs. Since the transmitted light diffuses through flowing blood, the deeper layers are thus reached by fluctuating light which can create a decorrelation between two successive B-scans at the same location. Several strategies have been evolved to limit projection artifacts [10‐12]. One of the easiest strategies to limit projection artifacts is to subtract the SCP OCTA en face image from the deeper layers [10, 11, 13‐15].

A shadowing artifact occurs when the OCT beam is attenuated or blocked, thereby impeding its passage to the deeper layers of the retina/choroid. This may occur due to the presence of different structures, including vitreous opacities, hemorrhage, and drusen. SS OCTA systems have partially overcome this limitation, since a longer wavelength limits attenuation from media opacities and improves penetration into the deeper retinal and choroidal layers [2].

OCTA is still limited in the visualization of the larger vessels of the choroid. This limitation is secondary to the signal attenuation due to scattering by the retinal pigment epithelium (RPE) and by the CC vessels [2]. As a result, choroidal vessels are typically displayed as silhouettes with complete loss of signal at greater depths. These limitations have been partially resolved with ultrahigh-speed SS OCTA instruments [16, 17]. An improved visualization of the medium-sized choroidal vessels may also be obtained in highly myopic eyes using the projection artifacts in the sclera [18].

While quantifiable metrics applied to images obtained using different OCTA devices were demonstrated to be not comparable [19] and quantification is strictly dependent on the OCTA en face image’s size [20], different studies have used quantitative variables to quantify the retinal and choroidal vasculature with OCTA.

OCTA en face images are composed of pixels which may be in a grayscale range of possible values from 0 to 255. In order to obtain quantitative analysis, a threshold may be applied to these images. Thresholding is used to create binary images [in which pixels over the applied threshold are displayed as white (or black) and pixels falling under the threshold are shown as black (or white, respectively)] from grayscale images. Moreover, thresholding may be obtained using different methodologies that can significantly influence the final quantification [21]. The binarized OCTA images may be also skeletonized, in order to obtain an image in which vessels are visualized as tracings of 1 pixel in width [22].

Binarized and skeletonized OCTA images (Fig. 3) may be used to obtain different quantitative metrics:

Previous studies on the CC have investigated the distribution, size, and percentage of flow voids (also called signal voids), which are small dark regions that alternate with granular bright areas, the latter thought to represent CC flow, and the former more likely to be secondary to CC vascular dropout or flow reduction [2, 25, 26].

OCTA has significantly expanded our knowledge on the organization of the retinal vessels. The retinal vascularization may be separated into the following plexuses: the SCP, the MCP, the DCP, and the RPCP [27‐30]. The RPCP is unique in that it is mainly located within the superficial nerve fiber layer surrounding the optic disc.

Since the introduction of OCTA, a number of models have been proposed regarding connections among the retinal vascular plexuses [29‐32].

The SCP is a dense meshwork of vessels accommodated in the retinal ganglion cell layer (GCL) [18]. The small capillaries arise from SCP arterioles and drain into SCP venules. The GCL slab enhances visualization of the capillary-free zones around the arteries and arterioles and helps differentiate arterial and venous systems [32]. Using OCTA, SCP arterioles were demonstrated to be located more superficially than SCP venules [30]. The SCP capillaries were shown to be arranged in a centripetal pattern in the macula and to converge on the parafoveal capillary ring [29]. In the periphery, the superior and inferior circulations converge in an interlaced comb pattern [29].

The SCP capillaries are connected with the other two layers of vessels at the inner plexiform layer/inner nuclear layer (IPL/INL) border (MCP) and inner nuclear layer/outer plexiform layer (INL/OPL) border (DCP). While the MCP colocalizes with the bipolar cell processes and amacrine cells, the DCP colocalizes with horizontal cells. These two plexuses are thus close to the high-oxygen-demand synapses of the IPL and OPL, respectively. The capillaries of the MCP and DCP have a lobular configuration without directional preference within their laminar planes [29]. Because the MCP and DCP consist of capillaries of uniform size and are separated by only a small distance, they are often grouped together as the deep vascular complex (DVC).

Around the foveal avascular zone (FAZ), these three retinal capillary plexuses converge to form a single parafoveal capillary loop and collectively define the borders of the FAZ [29]. Although some OCTA studies have separately investigated the FAZ area in different retinal capillary plexuses, recent papers have opted to assess this measurement in a single segment (whole retinal thickness). This choice was due to histologic and technical reasons: (1) there is a strong body of evidence suggesting the retinal plexuses merge at the edge of the FAZ, which may be thus considered a singular structure throughout the entire foveal thickness [33]; and (2) assessing the FAZ size at different segments may lead to increased variability of measurements [34]. The FAZ size was demonstrated to increase with age [35, 36] and disease (e.g. myopia, hypertension) [37, 38].

Recently, using higher-axial-resolution OCTA technology, Nesper and Fawzi [30] identified distinct vascular connections from large-caliber arterioles and venules in the SCP to each of the three retinal capillary plexuses in healthy subjects. This organization accounts for the high oxygen demand of the plexiform layers that need highly oxygenated (arteriolar) blood in the MCP and DCP. Furthermore, they provided further evidence for the presence of vortices in the DCP, these draining into conduits which are straightly connected to venules in the SCP. Interestingly, collateral vessels crossing the horizontal raphe were identified at the DCP level. These crossing vessels were speculated to represent an alternative pathway of least resistance in the event of occlusive diseases. Nesper and Fawzi’s observations would appear to support the theory that each of the three plexuses has its own arteriolar supply and venular drainage, which would theoretically allow each neurovascular unit to have independent control of its vascular supply under physiologic conditions.

This parallel organization of the retinal circulation suggested by Nesper and Fawzi has been challenged by Freund and Sarraf [31] who have suggested a predominantly series-based arrangement of the retinal circulation, with the bulk of the venous drainage of the retina originating at the level of the DCP. The presence of collateral vessels in the setting of retinal venous occlusive disease at the level of the DCP, but not at the SCP, would appear to support this contention.

Recently, using a full-spectrum probabilistic OCTA with a novel algorithm for three-dimensional projection artifact removal (PAR), Hirano and colleagues [39] investigated the retinal microcirculation. Since full-spectrum OCTA algorithms improve axial resolution, the authors provided important insights into the three-dimensional architecture of the retinal vasculature. Remarkably, unlike the MCP and DCP, the location of the SVP significantly varied among retinal regions. In all parafoveal regions tested, the SVP was featured by a small peak at the NFL-GCL junction and a larger, broader peak within the GCL. Conversely, the perifovea had only a single SVP peak. Furthermore, this peak was closer to the ILM in the perifoveal nasal, superior, and inferior quadrants. The authors presumed that this displacement was secondary to the thicker NFL associated with the arcuate and papillo-macular bundles in these regions.

The nerve fibers in the retina are unmyelinated and they thus require large amounts of energy to maintain ion concentration gradients [40]. The retinal nerve fiber layer contains the radial peripapillary capillary network to supply oxygen and metabolites to the optic nerve axons. The RPCP is composed of long, straight vessels with few apparent feed points and anastomoses [32]. These capillaries arise from peripapillary retinal arterioles, extend radially from the optic disc in parallel with the nerve fiber axons, farthest along the temporal arcades [41].

Using widefield OCTA, Jia and colleagues [42] examined the RPCP flow characteristics in ten eyes of ten healthy patients and illustrated an association between perfusion in this vascular network and nerve fiber layer thickness. The perfusion of these regions may thus serve as a surrogate measure for the metabolic activity of the ganglion cell axons.

Given that previous imaging techniques have significant limitations for the evaluation of the CC, OCTA has evolved into a key technology for investigation of the CC [2, 25, 26].

En face OCTA images of the CC are generated by segmenting a thin slab with a thickness of 10–30-µm starting at Bruch’s membrane. In these OCTA CC en face images, small dark regions (called flow voids or signal voids) [43, 44] alternate with granular bright areas, the latter thought to represent CC flow, and the former more likely to be secondary to CC vascular dropout or flow reduction [44]. Signal voids seem to represent the inter-capillary spaces which have been characterized in histological studies on the CC [45].

However, histological images of the CC are slightly similar to the OCTA versions. With more advanced OCTA technology, we expect to have an improved and more truthful visualization of the CC. However, an enhanced visualization of the CC network may even be obtained with commercially available devices using multiple en face image averaging [46]. The registration and averaging of sequential en face OCTA images was shown to transform the poorly defined granular appearance of the CC displayed with OCTA into a morphologic pattern that more closely represented the meshwork pattern observed on histology. Notably, not only did averaging improve CC visualization, but more precise quantitative measures of the CC were obtained.

In the CC OCTA images, the number and size of the signal voids have been shown to follow a mathematical relationship of a power law: log (number of signal voids) equals a scaling factor times the log (size of signal voids) plus a constant [47]. The presence of a power law relationship presumes that the CC is characterized by a propensity for capillary segments to be affected in the vicinity of an already nonfunctional segment.

As explained above, OCTA is still limited in the visualization of the medium- and larger-sized choroidal vessels. This limitation is mainly secondary to scattering by the pigment in the RPE and by the vessels in the CC, and consequent signal attenuation. Recently, Maruko and colleagues [48] used an SS OCTA device to image the choroid in healthy subjects, and subsequently employed a novel algorithm to improve the visualization of the choroidal vessels. They demonstrated that removal of the projection artifacts of CC can make the choroidal blood flow visible. Importantly, the choroidal blood flow area was correlated with choroidal thickness.