Rapid prototyping of a retinal multivascular network phantom for optical retinal vascular imaging equipment evaluation

Retinal vascular health holds paramount importance for healthy vision. Many technologies have been developed to examine retinal vasculature non-destructively, including fundus cameras, optical coherence tomography angiography (OCTA), fluorescein angiography (FA), and so on. However, there is a lack of a proper phantom simulating the critical features of the real human retina to calibrate and evaluate the performance of these technologies. In this work, we present a rapid, high-resolution, and economical technology based on 3D printed mold-based soft lithography and spin coating for the fabrication of a multivascular network and multilayer structural retinal phantom with the appropriate optical properties. The feasibility of the retinal phantom as a test device was demonstrated with an OCTA system and a confocal retinal ophthalmoscope. Experiment results prove that the retinal phantom could provide an objective evaluation of the OCTA and confocal retinal ophthalmoscope. Furthermore, the microfluidic phantoms enabled by this fabrication technology may support the development and evaluation of other techniques.


Introduction
The retinal vascular system supplies the inner retina with oxygen and nutrients and removes retina metabolism byproducts.Diseases with the retinal vascular system, such as hypertensive retinopathy, retinal vascular occlusion, diabetic retinopathy, and so on, can result in loss of vision.Moreover, it has been proved that the changes in the retinal vascular system can predict disorders that could induce morbidity and mortality, including hypertension, diabetes, coronary disease, renal disease, and stroke [1].Therefore, accurate mapping of the retinal vascular system has become one of the vital targets and intensive efforts have been devoted to technology development to provide more accessible high-resolution retinal imaging techniques [2].
Currently, the most commonly used optical retinal vascular imaging techniques in ophthalmology clinics include ophthalmoscope, digital fundus camera, fluorescein angiography (FA), and optical coherence tomography angiography (OCTA).Validation and comparison of the performance of these optical technologies is often needed during product development, application for regulatory approval, and periodic inspections during service.Therefore, a retinal phantom with a multivascular network, multilayer structure, corresponding physical dimensions, and appropriate optical properties of human eyes to support device development, regulatory approval, and periodic inspection is desired.
Consequently, many studies have been done to develop retinal phantoms with vascular models for the evaluation of corresponding techniques.Straight and thin microtubes or wires have been embedded in casted tissue phantoms, such as epoxy, aliphatic polyurethane, and polydimethylsiloxane (PDMS), to simulate vascular in tissue to calibrate retinal oximetry [3], OCTA [4,5], and Raman spectroscopy [6], respectively.However, with this technique, the microfluidic channels are usually straight and the minimum width is much larger than the real vasculature.Long Luu et al. reported a microfluidics-based retinal phantom with 36 µm-width microfluidic tortuous channels created with nanosecond laser inscription.Nevertheless, the technique used a lab-built 266 nm nanosecond laser inscription system which is inaccessible for the majority of labs, and hasn't been demonstrated with a multilayer structure similar to the human retina [7].The invention of 3D printing allows the manufacture of structures that are difficult with traditional manufacturing methods.Jianting Wang et al. demonstrated a 3D-printed phantom with fluid channels for the characterization of OCT, microscopy, and hyperspectral reflectance imaging [8].With the rapid development of the 3D printing technique, the same group reported a 3D-printed microfluidics-based retinal vascular phantom and proved its capability for calibrating a hyperspectral reflectance retinal oximetry system [9].Whereas, the human retina possesses a multilayer structure with different layer thickness and scattering properties, which is difficult to realize with 3D printing.Furthermore, the printed hollow channel width cannot achieve the claimed 3D printing resolution, which hinders the application in microfluidics [10].
Because of the limitations of directly 3D-printed retinal phantoms, soft lithography has always been a preferred option for microfluidic phantoms.With soft lithography, PDMS is commonly used to replicate the structure of a mold to form a chip with open channels.With plasma processing, the open channel could be bonded with substrates, such as glass, PDMS, or other materials, to form a microfluidic chip.The most frequently used molds in soft lithography are created with photolithography, allowing the accurate fabrication of complicated microfluidic systems.Shawn Mishra et al. fabricated a microfluidics-based retinal phantom to study the retinal progenitor cell migratory response to aid in transplant therapies for retinal diseases [11].On another aspect, Hyun-Ji Lee et al. developed a microfluidics-based retinal phantom for the evaluation of OCTA structural imaging and retinal vascular imaging [12].Nevertheless, the retinal phantom was produced with silicon wafers, the fabrication of which with photolithography or silicon dry etching is also complicated, time-consuming, and requires dedicated facilities and special expertise.
The combination of soft lithography and 3D printing, especially stereolithography (SLA) and digital light processing (DLP), empowers the fabrication of a simple, fast turnaround, and low-cost soft lithography mold prototyping with lower than ten-micrometer resolution [13,14].However, an unavoidable problem with SLA or DLP printed resin mold is the inhibition of PDMS curing.Studies reported customized postprocessing methods for 3D printed molds to solve the problem, such as sonication, heating, UV post-curing, solvents, salinization, or a coating [15][16][17][18][19][20][21][22][23][24].Venzac et al. conducted a systematic study about this problem and revealed that the fundamental reason behind the PDMS inhibition is the phosphine oxide-based photo-initiators leached out from the 3D-printed molds poisoning the PDMS catalyst [25].This inhibitor leaking could be eliminated via UV and/or thermal treatments.
In this work, we present a 3D printed mold-based soft lithography and spin coating combined technique for the rapid and economical prototyping of a microfluidic-based multilayer structural phantom.We demonstrate that this technique can reliably produce retinal phantoms that comprise a complex microfluidic multivascular network and a multilayer structure.The accuracy of the critical structures was carefully characterized.To showcase the functionality of the fabricated retinal phantom for the evaluation of the retinal imaging instruments, it was tested with an OCTA system and a commercial confocal retinal ophthalmoscope.

Retinal phantom design
The retina is a semi-transparent tissue that lines the internal posterior surface of the eye.The eleven-layer structure of the retina, the thickness of each layer, and the scattering properties of each layer have been thoroughly characterized with high-quality OCT images in previous studies [26][27][28][29].In this work, we also included the eleven-layer retina structure with distinct layer thicknesses and scattering properties for each layer [28,30].
In the analysis of the retinal vascular network with OCTA in ophthalmic clinics, the complex vascular system is commonly divided into two networks based on the depth, i.e. superficial vascular complex (SVC) and deep vascular complex (DVC) [31][32][33].In this work, we designed a multivascular network with the combination of two vascular networks, i.e.SVC and DVC, based on previously published ultra-wide retinal OCTA images [34,35].The designed SVC and DVC networks are shown in Fig. 1.The SVC network includes the optic disc, fovea, retinal arteries, retinal veins, and 2 nd and 3 rd -order branches of the retinal arteries and veins.A microvascular network was designed around the fovea to form the foveal avascular zone (FAZ).The widths of the vessels in SVC are 200 µm, 150 µm, 100 µm, 80 µm, 50 µm, and 30 µm.The DVC was designed as a network of microvascular with a vessel width of 20 µm and a pattern cycle of 250 µm.The FAZ area of the SVC and DVC was 0.25 mm 2 and 0.43 mm 2 [36,37].The SVC was embedded within the retina nerve fiber layer (NFL), and the DVC was embedded within the outer plexiform layer (OPL).During the assembling process described in 2.4, the SVC and DVC networks are stacked in a sequence with the FAZ and the fluid inflow and outflow channels in the two networks aligned with each other, respectively.

3D printing of the mold
The molds used in this study were produced with a 3D printer based on projection micro stereolithography (S140, BMF, China) [13].Both the lateral and axial printing resolution of this 3D printer are 10 µm.Before printing, two 3D disc models with a diameter of 49 mm and a thickness of 1 mm were designed.The SVC and DVC vascular networks shown in Fig. 1. were added to the center of the top surfaces of the disc models as a 10 µm-thick layer, respectively.The two discs with the protruded vessel networks were then sliced with a layer thickness of 10 µm to produce the image stacks for 3D printing.After 3D printing, the printed molds were first rinsed with alcohol to remove the extra fluidic resin on the surface.To further remove the photo-inhibitors to facilitate PDMS molding, the printed molds were further cured with UV for 3 hours and heated at 120°C for 3 hours [25].Figure 2 shows the picture of the successfully demolded PDMS phantoms with the SVC (Fig. 2

Spin coating of the layer structure
We used PDMS (Sylgard 182, Dow Corning, Midland, MI) to simulate the human retina.The mass ratio between the base and curing agent is 10:1.The refractive index of the PDMS mixture is 1.41.To simulate the scattering property of the human retina, titanium dioxide nanopowder (TiO 2 , T164497, Aladdin, China) was dispersed into the curing agent using a probe sonicator (JY92-IIDN, Scientz, China) for 30 mins and then mixed with the base.Afterward, the sample was placed in a vacuum chamber for 30 mins to remove the air bubbles.
The microfluidic multilayer retina phantom was fabricated in three parts.The first part includes the NFL layer and SVC channel, the second part includes the GCL, IPL, and INL, and the third part includes the OPL to RPE with DVC embedded in the OPL.To fabricate the microfluidic channel with layer structure, the 3D printed mold was used as the spin coating substrate.For the second part which does not include a vessel channel, a 2-inch silicon wafer was used as the coating substrate.During spin coating, the PDMS sample with different concentrations of TiO 2 nanopowder was dropped in the center of the mold and spun for 1-2 mins with different speeds to create thin films with different thicknesses and optical properties [28,30].After spinning, the PDMS was cured on a heat plate at 120°C for 10 mins.The mold with the cured PDMS was cooled for the fabrication of the next layer.This process was repeated for all the designed layers as shown in Fig. 3.

Model eye assembling
Because the first part only composites the NFL layer and SVC channel, it is too thin to be demolded.A 0.4 mm-thick transparent PDMS layer was added to the top of the first part by pouring the PDMS onto the mold.For the third part, a 2 mm-thick TiO 2 doped PDMS layer was created after the RPE layer to simulate the choroid and also facilitate the connection of the retina phantom with the inflow and outflow tubing.After curing, all three parts were demolded by tearing the coated film off.
During assembly, the second and third parts were bonded with plasma treatment firstly, on which punch was carried out to build the inlet and outlet.Then the first part was aligned and bonded with the previously bonded part.Silicon tubing with an inner diameter of 0.8 mm and an outer diameter of 1.8 mm was inserted into the inlet and outlet to allow the simultaneous fluid pump of the SVC and DVC networks.The model eye used in this study includes a lens group with a focal length of 19.5 mm (51D) and a cylindrical housing, as shown in Fig. 3(c).To facilitate the assemble of the flat retina phantom and the model eye and mitigate the spherical aberration in the meantime, the lens group comprises a plano-convex lens and a plano-concave lens.The fluidic multilayer retinal phantom was placed at the posterior of the model eye in contact with the plane surface of the plano-concave lens for the instrument calibration.

Critical parameter measurement
The critical parameters of the 3D-printed molds and retinal phantom were cautiously characterized.The widths of the microfluidic channels on the 3D-printed molds were measured using a measuring microscope (MM400N, Nikon, Japan) with a 10 X objective lens and 0.1 µm measuring resolution.It is known that fabricating perfect perpendicular straight walls with an SLA 3D printer is difficult.Therefore, to obtain consistent width measurements, the largest width of each channel was measured.The widths of the vessel channels in the PDMS phantom parts were measured in the same way before the plasma bonding process with the open channel facing up.Images of the SVC and DVC of the demolded PDMS parts were taken with the same microscope.FAZs in SVC and DVC were segmented and the pixel numbers of segmented FAZs were counted and converted into areas.To characterize the PDMS channel shrinkage caused by the high-temperature cure process, the SVC and DVC channel heights of the 3D-printed molds and the demolded PDMS phantom parts were measured using a confocal microscope (inVia, Renishaw plc, UK) with an axial resolution of <1 µm and a 50 X objective lens.During measurement, the focal point of the microscope was adjusted to the adjacent area and the center of the open channels sequentially.The channel heights were calculated as the differences between the two focal heights.The thickness of each one of the 11 layers was measured as the optical path length divided by the PDMS refractive index with a calibrated OCT system (Telesto, Thorlabs, US).

OCTA, retinal imaging and FA evaluation
The model eye was used to evaluate the performance of a home-built ophthalmic OCTA system.The OCTA system is based on a spectral domain OCT with a central wavelength of 850 nm and a 3 dB bandwidth of 97 nm.The A-line rate of the OCT is 120 kHz.The lateral resolution and axial resolution of the system are 13.5 µm and 5 µm, respectively.The depth range of the imaging system is 2.1 mm in air.The model eye with the retina phantom was mounted on an adjustable frame and the retina phantom was placed at the OCT imaging focal plane.During imaging, the flow speed of the intralipid at the import was set as 1 mm/s with a syringe pump (Cole Palmer 78-9100C, Illinois, US).The retinal phantom was sampled with 2048 data points per A-line, 800 A-lines per B-scan across a length of 13 mm in the fast scan direction, and 800 B-scans per volume over a range of 13 mm in the slow scan direction with three consecutive B-scans at the same position.The collected OCT data were processed with the split-spectrum amplitude-decorrelation angiography (SSADA) algorithm [38].Due to the invisibility of the 0.4 mm-thick transparent PDMS layer over the NFL of the retinal phantom in the OCT and OCTA retinal images, the thickness of this transparent layer was not calculated in the retinal phantom parameter calculation.
The red-free retinal imaging and FA imaging evaluation with the retina phantom were validated with a commercial confocal retina ophthalmoscope (Apollo CRO Plus, Microclear Medical Instruments Co.LTD, China).The ophthalmoscope has a resolution of 5 µm and a field of view of 150 o with a one-shot single capture.During the experiment, the vessel channels in the retina phantom were injected with fluorescein sodium.The red-free retinal image and FA image of the retina phantom were taken with a 488 nm scanning laser source.

Results
Figure 4 shows the photos of the assembled retina phantom.To test the connection of the SVC and DVC vascular networks, red ink was injected into the microfluidic channels.As shown in Fig. 4(b) the SVC and DVC vascular networks were highlighted with red ink, indicating the microfluidic channels were functioning well after the assembly process shown in Fig. 3. Table 1 shows the measurements of the critical parameters.By controlling the spin coating speed, the layer thickness of the phantom achieves the designed thickness with a standard deviation under 3 µm.Table 1(b) shows the vessel channel widths of the 3D-printed molds and the demolded PDMS parts.Compared to the channel widths of the 3D-printed molds, the channel widths in the demolded PDMS parts shrunk by an average of 2.25%.Table 1(c) demonstrates the areas of the FAZ in the SVC and DVC networks in the demolded PDMS parts.The measured vessel widths and FAZ areas achieve the designed dimension with acceptable standard deviations, proving that the 3D printed molds can be used to fabricate vascular networks with high accuracy.The channel heights for the 3D-printed SVC and DVC molds are 12.11 ± 0.16 µm and 12.28 ± 0.16 µm, respectively.The SVC and DVC channel heights of the demolded PDMS parts are 11.6 ± 0.19 µm and 11.75 ± 0.28 µm, which indicates that the PDMS channel height shrunk for 4.25% on average due to the high-temperature cure process.OCTA imaging evaluation function of the retinal phantom was tested with the home-built ophthalmic OCTA system described in 2.6.Figure 5 shows the 3D view of the cropped OCT and OCTA volume data of the retina phantom obtained with the home-built system.In the 3D OCT volume image, the layer structure of the retina phantom is captured, which can be seen on the cross-sectional surface of the 3D volume dataset.Although there is a slight scattering difference between the intralipid and the layer structure, the vascular networks cannot be distinguished.

Table 1. The critical parameter measurements of the retina phantom. (a) PDMS phantom layer thicknesses, (b) vessel channel widths in the 3D-printed molds and the demolded PDMS parts, (c)
FAZ areas in the demolded PDMS parts 1 2 OCTA imaging evaluation function of the retinal phantom was tested with the home-built 3 ophthalmic OCTA system described in 2.6.Figure 5 shows the 3D view of the cropped OCT 4 and OCTA volume data of the retina phantom obtained with the home-built system.In the 3D 5 OCT volume image, the layer structure of the retina phantom is captured, which can be seen on 6 the cross-sectional surface of the 3D volume dataset.Although there is a slight scattering 7 difference between the intralipid and the layer structure, the vascular networks cannot be 8 distinguished.Whereas, in the processed OCTA dataset, the vascular networks are resolved 9 distinctively, as shown in Fig. 5 (b).Fig. 6 (a) shows the on-face view image of the 3D OCTA 10 dataset with color-coded depth information.Fig. 6 (b-c) presents the cross-sectional OCT and 11 OCTA B-scan at the location indicated by the blue dashed line in Fig. 6 (a).Similar to Fig. 5., 12 the layer structure can be seen distinctively and the cross-section of the vessel channels is 13 vaguely distinguishable in the OCT B-scan image (Fig. 6 (b)).However, in the processed 14 OCTA B-scan, the vessel cross-sections can be seen clearly without any interference from the 15 layer structure.Whereas, in the processed OCTA dataset, the vascular networks are resolved distinctively, as shown in Fig. 5(b).Figure 6(a) shows the on-face view image of the 3D OCTA dataset with color-coded depth information.Figure 6(b-c) presents the cross-sectional OCT and OCTA B-scan at the location indicated by the blue dashed line in Fig. 6(a).Similar to Fig. 5., the layer structure can be seen distinctively and the cross-section of the vessel channels is vaguely distinguishable in the OCT B-scan image (Fig. 6(b)).However, in the processed OCTA B-scan, the vessel cross-sections can be seen clearly without any interference from the layer structure.To enable the analysis of SVC and DVC vascular networks separately, the retinal vascular network was separated by projecting it through the NFL to GCL and IPL to INL, corresponding to the SVC and DVC networks respectively, as shown in Fig. 7(b), (c).The FAZs in the SVC and DVC vascular networks were measured.The diameters of the FAZs in the SVC and DVC networks are 556 µm and 722 µm, and the corresponding areas are 0.24 mm 2 and 0.41 mm 2 , respectively.The measured results agree with the measured FAZ in Table 1(c).
Figure 8(a) and (b) present the red-free retina image and the FA image of the retina phantom, respectively.The red-free retina image has a high contrast and both the SVC and DVC vascular   networks can be acutely differentiated.In the FA image, the SVC vascular network is much more conspicuous compared to the DVC vascular network.It is also observed that the specular reflection from the lens surfaces inside the model eye causes a strong saturation in both the red-free retina image and the FA image.

Discussion
In this study, we demonstrated a microfluidic multilayer phantom with complex vascular networks spreading at different depths simulating the human retina fabricated with a high-resolution, economical, and efficient technology.Although attempts have been made to create microfluidic retinal phantoms with laser micromachining and 3D printing in order to characterize techniques including Doppler optical coherence tomography, Doppler ultrasound, hyperspectral reflectance imaging, and so on [7,9], it is difficult to realize the combination of retinal multilayer structure and microfluidic vessel channels with these fabricating technologies to fully mimic the human retina.Recently, Hyun-Ji Lee et al. demonstrated a retina phantom with microfluidic channels and a multilayer structure for the evaluation of OCTA [12].Whereas, the fabrication of such retina phantom relies on silicon wafers, which requires a photomask with the desired pattern and a series of steps, including substrate cleaning and preparation, photoresist application, exposure and development, etching, implantation, and photoresist removal.The process is expensive, time-consuming, and requires specialized equipment and a clean room to produce.For the next design iteration, a new photomask and all the subsequent steps will be needed again.The soft lithography based on 3D-printed molds presented in this work provides an option to create fast turnaround and high-resolution microfluidic phantoms for the evaluation of the corresponding technologies.
In our work, the 3D printer has a resolution lateral and axial resolution of 10 µm and costs 60 k USD.Whereas, the price for 3D printers with 20 µm ∼ 30 µm resolution only costs several thousands of dollars and requires minimum maintenance.In addition, 3D printers have become an accessible technology that many labs are already equipped with.Therefore, a simple method of creating high-resolution microfluidics with them would lower the threshold of microfluidics and prompt more applications.However, the surface flatness of the 3D printed mold is not as good as the conventional microfluidic mold produced with photolithography, which may influence the spin coating process and the layer thickness consistency.The layer thickness measurement method adopted in this study, using a calibrated OCT imaging system with an axial resolution of 5 µm, may not be sufficient to characterize this inconsistency.Nevertheless, because the multilayer phantom will mainly be used to evaluate an OCT system's ability to characterize the layer structure of the human retina, the result of this work proves that the phantom fabricated with the 3D printed molds could fulfill this errand.
In the retinal phantom we created with this technology, the minimal vessel width of 20 µm was achieved.It is observed that the width accuracy of the 20 µm is slightly deteriorated compared to the wider channels due to the approaching of the 3D printing resolution of 10 µm.The limits of the minimal channel width can be down to several micrometers depending on the resolution of the 3D printer.Compared to the 3D-printed molds, vessel channel widths and heights in the PDMS phantom shrunk by an average of 2.25% and 4.25%, respectively.The channel width shrinkage is similar to the previously published work [39,40].The higher channel height shrinkage may be due to the thin layer thickness created by the spin coating process.Furthermore, complex biomimicking features, such as the FAZ in the retina, were created with high precision in this retina phantom, which could be used to evaluate the critical parameter measurement performance of the tested instrument.
OCTA imaging is sensitive to the blood flow speed.In this work, the SVC and DVC networks were pumped simultaneously and the intralipid flow speed was controlled at the import location with a syringe pump.Albeit the rough control of the flow, all the vessel channels were resolved successfully.In future work, we will develop the technology to measure the flow speed in the vessel channels to better study the relationship between the flow speed in the vessel and the ability of OCTA to distinguish the vessel.Additionally, the SVC and DVC could be pumped in parallel with different syringe pumps to enable the independent flow speed control of the two networks.
Moreover, this phantom could be a powerful tool to evaluate the performance of different OCTA algorithms, especially in the reduction of projection artifacts which can be seen in Fig. 6(c).
Furthermore, the results of this work prove that the fluorescein sodium injected retina phantom can be used to evaluate the retinal imaging and FA imaging function of a confocal retina ophthalmoscope.However, the fluorescence intensity of the fluorescein sodium is different when it is injected into the blood vessel of a patient or the retinal phantom [41].Therefore, the fluorescence intensity difference needs to be considered during the evaluation.However, the exploration experiment in this work proves that the retina phantom could be used to evaluate the imaging resolution and the field of view of a confocal retina ophthalmoscope or fundus camera.It is observed that the specular reflection at the lens surfaces in the model eye impacts the image quality as shown in Fig. 8, which can be improved by using lenses with anti-reflection coatings in the future.

Conclusion
An economic, quick, and efficient high-resolution microfluidic multilayer phantom fabrication technology was introduced in this work.A microfluidic multilayer high-resolution retina phantom with SVC and DVC vascular networks spreading at different depths simulating the human retina fabricated with this technology was presented.Assembled with a model eye, the retina phantom was imaged with a home-built ophthalmic OCTA system and a commercial confocal retina ophthalmoscope.Results prove that the retina phantom produced with this technology could be used to evaluate the performance of an OCTA system and a confocal retina ophthalmoscope or a fundus camera to examine the accuracy of the measured retina and retina vessel parameters.Moreover, the microfluidic phantoms enabled by this fabrication technique may support the development and evaluation of other techniques.

Fig. 1 .
Fig. 1.The designed vascular networks.(a) The superficial vascular complex (SVC) vascular network, which includes optic disc, foveal avascular zone, vessels, and fluid inflow and outflow channels; (b) the deep vascular complex (DVC) vascular network, which consists of a 20 µm-wide vessel network, foveal avascular zone, and fluid inflow and outflow channels.

Fig. 3 .
Fig. 3.The schematic drawing of retinal phantom fabrication.(a) Layer structure spin coating.The retinal layer structure is separated into three parts and created by spin coating PDMS with different TiO 2 concentrations; (b) The assembling process.The three separated layer structures were aligned, plasma bonded, and punched to form a microfluidic retina phantom; (c) Combination of the retinal phantom with a model eye.The retina phantom is assembled at the bottom of a model eye for retinal imaging instrument test.

Fig. 4 .
Fig. 4. Photos of the retina phantom.(a) The retina phantom in a Petri dish, (b) The retina phantom injected with red ink

Fig. 5 .
Fig. 5. OCT and OCTA volume images of the retinal phantom in a model eye.(a) 3D view of

Fig. 5 .
Fig. 5. OCT and OCTA volume images of the retinal phantom in a model eye.(a) 3D view of the cropped OCT volume images, (b) 3D view of the cropped OCTA images.

Fig. 6 .
Fig. 6.Cross-sectional OCT and OCTA images of the retinal phantom in a model eye.(a) On-face view OCTA image of the retina phantom, (b), (c) OCT and OCTA B-scan images at the location indicated by the blue dashed line in (a).

Fig. 7 .
Fig. 7. Separation of the vascular networks in the on-face view OCTA image based on depth.(a) On-face view of all the vascular networks, (b) the SVC vascular network, (c) the DVC vascular network.

Fig. 8 .
Fig. 8. Commercial confocal retina ophthalmoscope images of the retinal phantom in a model eye.(a) The red-free retina image, (b) FA image.

Funding.
National Key Research and Development Program of China (2021YFC2401401); Fundamental Research Fund (AKYZD2407).