Quantitative analysis of choriocapillaris in non-human primates using swept-source optical coherence tomography angiography (SS-OCTA)

: The choriocapillaris is a unique vascular plexus located posterior to the retinal pigment epithelium. In recent years, there is an increasing interest in the examination of the interrelationship between the choriocapillaris and eye diseases. We used several techniques to study choroidal perfusion, including laser Doppler flowmetry, laser speckle flowgraphy, and optical coherence tomography angiography (OCTA), but with the latter no standardized algorithm for quantitative analysis has been provided. We analyzed different algorithms to quantify flow voids in non-human primates that can be easily implemented into clinical research. In-vivo, high-resolution images of the non-human primate choriocapillaris were acquired with a swept-source OCTA (SS-OCTA) system with 100kHz A-scan/s rate, over regions of 3 × 3 mm 2 and 12 × 12 mm 2 . The areas of non-perfusion, also called flow voids, were segmented with a structural, intensity adjusted, uneven illuminance-compensated algorithm and the new technique was compared to previously published methods. The new algorithm shows improved reproducibility and may have applications in a wide array of eye diseases including age-related macular degeneration (AMD).

image the choroid providing volumetric information on the microvasculature, but quantification is still challenging.
Swept-Source OCTA (SS-OCTA) has been used in a wide variety of recent studies as a non-invasive tool to visualize the choroid in-vivo [8][9][10][11]. Compared to spectral-domain OCTA, SS-OCTA offers several advantages including reduced fringe washout, lower sensitivity roll-off, increased imaging range with improved detection efficiency and dual balanced detection [12]. Commercial SS-OCT/OCTA systems for imaging the posterior pole of the eye operate at wavelengths around 1060 nm, which provides improved penetration over commercial spectral domain OCT/OCTA systems that operate at wavelengths around 850 nm [13]. The advantage of using SS-OCT/OCTA to investigate choroidal thickness has been well documented [14][15][16] and some commercial SS-OCT/OCTA systems have implemented automated choroidal segmentation. Enface choroid circulation analysis still remains challenging given the lack of distinct boundary between the choroidal layers, as well as low signal from choroidal vessels in OCT/OCTA scans, mainly due to the signal loss of the light penetrating through the RPE [17].
The choriocapillaris is the anterior-most choroidal layer underneath the RPE. This is a thin, monolayer vascular network that provides metabolic supply to the photoreceptors, carries out waste from the RPE, and regulates the temperature of the posterior retina [1]. The geometry of the choriocapillaris perfusion map is non-homogenous and several histological studies [18][19][20][21] have shown the decrease of vascular perfusion from the posterior pole to the periphery. In OCTA images, unlike choroidal vessels, choriocapillaris is seen as bright pixels, while the non-perfused areas are shown as dark regions, and referred to as flow voids. The sizes and patterns of flow voids can be altered by age-related macular degeneration [22][23][24], myopia [25], and hypertension [26]. However, the knowledge of how to optimally use OCTA to evaluate choriocapillaris remains very limited. Different algorithms have been developed to quantify the flow voids, but the comparison between these approaches is lacking. There is a disagreement in terms of the density of flow void in normal human eyes from various studies using OCTA [26][27][28][29], which may be related to different segmentation methods, different instruments used, the compensation mechanisms for projection artifacts from the superficial blood vessels, and the different thresholding algorithms. In addition, most of the studies focused on a relatively small region centered at the fovea, and have not provided an insight to more peripheral regions of the choriocapillaris.
In the present study, we used SS-OCTA derived choriocapillaris images from non-human primates to evaluate and compare different algorithms. So far, to the best of our knowledge, no study has been conducted to image the choriocapillaris in non-human primates using OCTA. The rationale behind this approach was two-fold. On the one hand, the similarity of ocular vascular network to human, the presence of a macula and cone photoreceptor predominance [30] make the non-human primates an attractive animal model to investigate the role of the choroid in AMD-like pathology. Importantly, many studies have reported signs of AMD in older non-human primates and some studies also observed early onset drusen [31]. On the other hand, the images are less influenced by motion artifacts, which can affect quantification of flow voids. Motion-related artifacts make comparison of different algorithms difficult, because limited reproducibility can either be a result of poor image quality or of poor algorithm performance.
Two different protocols (3 × 3 mm 2 and 12 × 12 mm 2 ) were used and a new algorithm was developed to analyze the choriocapillaris plexus. This algorithm accounts for the dark appearance of deeper choroidal vessels and the uneven illuminance that is particularly important when wider fields are used. ignal was onkeys, a repeatability test was conducted where each imaging protocol was repeated three times with 3-minute acquisition interval. The intraclass correlation coefficient (ICC) was calculated and reported. Repositioning of the monkey eye was performed through live fundus display between volume acquisition cycles to ensure the proper alignment of the imaging beam.

Image processing
The RPE layer was automatically segmented by a built-in segmentation tool from the review software (PLEX Elite 9000 Angiography Analysis, Carl Zeiss Meditec, Dublin, CA, USA), and manual correction was performed when necessary. The choriocapillaris was then segmented within a thin slab (31-40 µm) below the RPE, and a maximum projection in the axial direction was performed to extract the enface choriocapillaris map. Both angiographic and morphological enface images were exported from the review software and transferred to MATLAB for further analysis (MathWorks, Inc., Natick, MA, USA). Indeed, the choriocapillaris was found to be thinner than 10 µm in histology [20], but usually it is more a question whether it is practical to reliably segment such a thin slab from a commercial system. A summary of the thicknesses of choriocapillaris slab segmentations from other groups is shown in Table 1, and 10 µm thin slab segmentation used here is lower than the average.      Table 3 sho flow void segm the different usly, the algor oposed in this p mal difference s less than 4% Cs vary signific Li [43], only t sent algorithm od reproducib ty. Our results ifferent perform ferent algorithm are required It is also not ty using the pro = 0.07). 6 Figure 7 r eyes of one n data are evalu was probably calculated den void density d densities from The images e to 1024 × 10 exported imag (Fig. 8(a-c)) a of 7 μm, 18 μ le 3. Statistical re ng different segme es (N = 11) were a from Methods lli et al. [24] lo et al. [34] paide [26] . (mean-1SD) [29] (mean-1.5SD) [29 Proposed represents an ex non-human pri uated using our y caused by th nsities of flow difference is on m repeated scan 7. Repeated measu oriocapillaris due t hown in a-c) and g n in d-f) and j-l). t of pixel resol xported from t 024 pixels, wh ge was downs and spatial poo μm and 30 μm,    [46,47].
The flow can also be e primates and lateral resolu human data n take into con arising from removed by underneath th artifacts may be simulated b the near futur

Conclusio
In summary, w healthy nonsegmentation choriocapillar algorithms th observed an algorithm can diseases with 10. Wide field flow density map. (b) F ally in the density ased density o cence angiogra ris network as y network is f provide good periphery [45].
flow void dis othesis that inc oven, although d poor visual a ported by two ood flow is a void analysis employed in fu allowed to und tion, sampling need to be add sideration othe drusen. How excluding reg he drusen [29] lead to incorre by artificially e.