Adaptive optics (AO) enhances the capabilities of ophthalmic imaging modalities such as optical coherence tomography (OCT) and scanning laser ophthalmoscopy (SLO) to resolve individual human photoreceptors and other retinal cells in vivo [1,2]. Operating at this uniquely high resolution requires effective measurement tools to fully characterize AO imaging performance from preclinical device development to clinical usage. However, no standard performance test method for ophthalmic AO devices currently exists. Lateral spatial resolution is the primary figure of merit which objectively quantifies AO performance, but measurement under meaningful conditions is challenging and very rarely reported in the numerous ophthalmic AO studies described in the literature.

Major medical imaging modalities (e.g., x-ray, CT, MRI, ultrasound) routinely use physical models known as phantoms to standardize system performance and optimize image quality throughout the lifetime of a device [3]. Phantoms are typically fabricated of stable inorganic materials which approximate the energy absorption/scattering properties of tissue while producing relevant and reproducible images that can be carefully analyzed against the phantom’s ground truth characteristics. Our group and others have produced retina-simulating phantoms for ophthalmic OCT systems with an accurate representation of the thickness and reflectivity of individual retinal layers, as well as realistic morphology of the foveal pit and optic nerve head [4–6]. Valente and Vohnsen recently produced another type of retina phantom, using photolithography to specifically create waveguiding structures which replicate the Stiles–Crawford effect [7]. The explosive growth in additive manufacturing (3-D printing) is now providing solutions for creating biologically relevant structures with even more complex and detailed geometry [8–10]. Here we present nanoscale 3-D printed phantoms of cylindrical structures modelling the geometric size and arrangement of cone photoreceptor cells at varying retinal eccentricities, thereby acting as application-specific AO lateral resolution targets. We have approximated the refractive indices of the cone outer segments (OSs) and the surrounding interphotoreceptor matrix, which is known to affect photoreceptor waveguiding behavior [11,12]. The phantoms were encapsulated in a sturdy package which also provides a diffusely reflecting surface needed for robust AO wavefront sensing. The phantoms are coupled to a custom model eye assembled with off-the-shelf components representing an emmetropic human eye. This Letter builds upon our initial report of similar phantoms with limited capability for multimodal AO resolution assessment in a more cumbersome model eye arrangement, demonstrated on one research-grade AO system [13]. We now report refined, more thoroughly characterized phantom and model eye designs suitable for multimodal AO imaging, along with quantitative resolution performance results on two AO systems.

The photoreceptor-like structures were fabricated via a two-photon polymerization (2PP) process with a direct laser writing (DLW) system (Photonic Professional GT, Nanoscribe GmbH, Eggenstein-Leopoldshafen, Germany). This system uses a highly focused femtosecond-pulsed laser to selectively photopolymerize liquid-based photoresist at spatially controlled locations with feature sizes as small as 200 nm. The laser beam is rapidly scanned in x- and y-directions via galvanometer-mounted mirrors, and the sample is moved to different z-planes with a motorized translation stage. All movements are synchronized by system software which reproduces a standard mechanical drawing of the structures.

As depicted in Fig. 1(a), 2PP DLW was performed with the system’s $63\times /1.4\mathrm{NA}$ microscope objective immersed in a droplet of IP-S photoresist (a proprietary Nanoscribe formulation) on the unpolished surface of a silicon wafer. After the patterned laser exposure, the photoresist was submerged in a developer solution to remove unexposed material from the wafer. The polymerized cylindrical structures were then sputter coated with a $\sim 1\mathrm{nm}$ gold film to increase their optical reflectivity—especially necessary for SLO—and to facilitate characterization with scanning electron microscopy (SEM) (6390LV, JEOL USA Inc., Peabody, MA). Since the IP-S resist is autofluorescent, we also used fluorescence microscopy to inspect the final phantoms.

 

Fig. 1. (a) Illustration of 3-D printing with immersion-mode 2PP DLW. (b) and (c) SEM images of structures before PDMS encapsulation (left) and fluorescence images of final phantoms (right) for (b) 0.6° and (c) 1.5°/3°/5° eccentricities. Some collapsed structures are visible in the 0.6° images. (d) and (e) Comparison of the measurements from phantom and human retina for (d) cone center-to-center spacing and (e) OS diameter.

Download Full Size | PDF

The 3-D printed structures approximated reported human anatomical values of the center-to-center spacing [14] and diameter [15] of the cone OSs at four different retinal eccentricities—0.6°, 1.5°, 3°, and 5°—realized in four separate blocks in two separate phantoms: 0.6° eccentricity in one phantom and the three remaining eccentricities in the second phantom [Figs. 1(b) and 1(c)]. The graphs in Figs. 1(d) and 1(e) compare anatomical dimensions to SEM measurements of the phantoms. The height of all structures was 5–10 μm, much shorter than actual photoreceptors, but overall a more mechanically stable design. A few structures still collapsed, especially at the edges of the densely packed 0.6° phantom, but these could be ignored during AO image analysis, since the central region was quite intact. The 1.5°/3°/5° phantom had larger diameter structures than what has been reported for the human retina; this was a tradeoff between accuracy and structural stability. The second phantom also shows an attempt to model foveal (0° eccentricity) photoreceptors, but these very tightly packed structures fused together during the print process, and the result was a monolithic structure with small protrusions at the top.

Figures 2(a)–2(d) illustrate the phantom fabrication steps after 2PP DLW and gold sputter coating. We encapsulated the structures in a $\sim 60\mathrm{\mu m}$ thick spin-coated film of polydimethylsiloxane (PDMS) to model the extracellular matrix with refractive index differential to IP-S ($n_{\mathrm{IP}\text{-}\mathrm{S}}=1.50$, $n_{\mathrm{PDMS}}=1.41$) comparable to that found in the human retina between the OS ($n=1.43$) and the interphotoreceptor matrix ($n=1.34$), as reported by Snyder and Pask [11]. The gold on the outer surface of each cylinder acts as a partial reflector and, therefore, influences scattering and waveguiding along with the refractive index mismatch. To produce a diffuse reflection which fills the AO pupil for optimal wavefront sensing, we added a textured outer surface to the encapsulated structures by replicating the texture from the non-adhesive side of common household tape (Scotch Magic Tape, 3M, Maplewood, MN). This was achieved by applying a small droplet of uncured PDMS atop the encapsulation layer and pressing the back side of the tape (adhered to a glass slide) against the droplet to spread it into a thin film ($<5\mathrm{\mu m}$) while it was heat cured. The phantom was then peeled off the silicon wafer, during which a few of the structures did not transfer from the silicon wafer into the PDMS encapsulant, and confirmed via the fluorescence images in Figs. 1(b) and 1(c). After peeling, a $\sim 25\mathrm{\mu m}$ film of PDMS was spin coated on the inner surface to create separation between the embedded structures and the final surface of the model eye onto which the phantom was mounted. Thus, the gold-free ends of the cylinders (in contact with the silicon wafer during 2PP) were the first surfaces for incoming light during AO imaging, and the textured surface was at the rear.

 

Fig. 2. (a)–(d) Phantom fabrication steps: (a) gold-coated structures on silicon substrate, (b) structures encapsulated in PDMS, (c) texture added to the outer surface, (d) encapsulated structures peeled from substrate and additional PDMS film added to the inner surface. (e) Model eye assembly with the inner surface of the phantom adhered to a fused silica wedge.

Download Full Size | PDF

The model eye [Fig. 2(e)] was assembled entirely of off-the-shelf opto-mechanical components: adjustable iris, anti-reflection-coated achromatic doublet lens with 19 mm focal length, lens tube chamber filled with isopropyl alcohol (IPA), and a 5° fused silica wedge onto which the phantom was pressed. O-rings prevented leaks at the achromatic and fused silica mounts on the lens tube. IPA was chosen because it is non-corrosive and provides a better index match to the glass materials than water. The dispersion of IPA is similar to water dispersion, though the model eye was not specifically designed to incorporate chromatic effects. The fused silica was wedged to further minimize specular back-reflections. To image different areas of the phantom, the model eye was mounted to a goniometer stage (GN05, Thorlabs Inc., Newton, NJ) for vertical tilt and on a rotatable post for horizontal tilt.

We imaged the phantoms with two research-grade multimodal AO systems, one built at the U.S. Food and Drug Administration (FDA) [16] and the other built at the National Eye Institute (NEI) [17], both systems having SLO and OCT channels. The model eye was mounted into the patient head rest, and the front of the achromat was aligned with the final pupil conjugate plane of the system. Closing the iris helped confirm proper alignment. Once the AO correction loop was closed, the spherical power on the deformable mirror was adjusted to push the focus to the plane of the phantom structures. Figure 3 shows AO-SLO and en face AO-OCT images taken with a 1° scan size on both systems. The OCT B-scans (not shown) were dominated by a single mirror-like reflection attributable to the gold-coated ends of the structures. The en face OCT images are an average over a depth range encompassing this reflection.

 

Fig. 3. AO-SLO and en face AO-OCT images of the (a) 0.6° phantom and (b) 1.5°/3°/5° phantom. In (b), spectrum saturation caused dark circles in the center of some structures in the NEI AO-OCT linear image.

Download Full Size | PDF

The 3° and 5° eccentricity regions are well resolved by both AO channels, appearing more sparse in the phantom than the equivalent retinal structures naturally appear in AO images, since the biological cones are also surrounded by smaller rod photoreceptors and other extracellular structures. The tightly packed mosaics at 1.5° (resolved with both channels) and 0.6° (resolved with AO-SLO only) closely resemble what is seen in the human retina at the cone OS tip layer [16,18]. Compared to their linear counterparts, the logarithmic OCT images of the 1.5°, 3°, and 5° phantoms illustrate more clearly that the apparent diameter of the structures decreases with center-to-center spacing, even though the structures all have the same physical diameter. This phenomenon is a direct result of speckle, wherein observed speckle point size decreases as the density of scatterers increases [19]. Though both the OCT and SLO light sources are coherent enough to generate speckle, the phase sensitive detection of OCT is more sensitive to the effect. The predominance of speckle in OCT reveals itself most fully in the 0.6° images, both linear and log, in which the seemingly random speckle pattern overwhelms the visibility of individual structures. The enhanced scattering from the gold coating on the structures likely increased the visibility of OCT speckle, which was a tradeoff to ensure SLO visibility of the structures at all.

Applying the phantom as a tool to quantify AO system lateral resolution, we measured the structure sizes in the SLO and linear OCT images and compared them to theory. Figure 4 plots the observed structure size normalized by the diffraction-limited theoretical size. The observed size is the mean full-width at half-maximum (FWHM) of horizontal or vertical profiles through 15–20 structures in each image. The theoretical size is the FWHM of the numerical convolution of a uniform distribution with width equal to SEM structure size (Fig. 1) and a distribution corresponding to the lateral point spread function (PSF) based on system parameters (Table 1) and the model eye treated as an ideal thin lens. Confocal detection, which reduces the overall lateral PSF slightly, was assumed to be negligible and, therefore, ignored in the theoretical PSF estimation.

 

Fig. 4. Mean structure sizes from AO images normalized by the theoretical size. The error bars represent one standard deviation.

Download Full Size | PDF

Table 1. AO System Parameters and Theoretical PSF Widths for 19 mm Focal Length Model Eye

View Table

A value of 1 on the graph in Fig. 4 indicates that observed and theoretical sizes are equal. The SLO sizes are tightly clustered around 1 (range 0.97–1.04), with the largest values occurring under the most challenging resolution condition at 0.6°. The maximum standard deviation is 12% of the mean, resulting from residual astigmatism in the images, which causes horizontal and vertical profile measurements to deviate from one another. The OCT size shows a clear speckle-influenced trend with eccentricity, and observed size drops well below theoretical size (minimum 0.92 at 1.5°). OCT size measurements at 0.6° were not possible because of speckle.

Overall, the phantom images and measurements confirmed that these two multimodal AO systems perform quite similarly. Both systems achieve AO-SLO measurements within 4% of diffraction theory, but the speckle-influenced AO-OCT structure sizes are not well predicted by the basic theory. Interestingly, cone OSs visualized in vivo with AO appear to decrease significantly in diameter from the periphery to the fovea [12,18] as we saw in our phantom, but the OSs do not actually have such an anatomical dependence [15]. Instead, this visual decrease is likely due to the waveguiding influence from cone inner segments (ISs) [12,20] with diameters that shrink significantly with reducing eccentricity [15]. The speckle clearly reduces the OCT image quality of other retinal layers such as the ganglion cell layer, where individual cells have only recently been visualized in vivo after considerable image registration and averaging effort [21], but the photoreceptor mosaic is routinely resolved with AO without the need for speckle-reduction methods. Our phantom simplified version of the photoreceptor mosaic with enhanced scattering from gold and without the other retinal scattering structures creates conditions for periodic interference patterns, bearing resemblance to Vohnsen’s scattering-based model of light propagation in the OSs [22]. Our experimental results with these photoreceptor phantoms provide the basis for a new hypothesis worthy of further study: subcellular light scattering within/between cells and cone IS waveguiding—two important features missing from our phantoms—practically eliminate the appearance of speckle in AO imaging of the retinal photoreceptor mosaic.