Two-lens telecentric model eyes for image distortion measurement in adaptive optics ophthalmoscopes.

Save

Here we present the design of telecentric model eyes for measuring image distortion in adaptive optics ophthalmoscopes, using pairs of achromatic or near-achromatic commercial off-the-shelf (COTS) and/or custom doublet lenses. The proposed model eyes can operate over visible or visible plus near infrared wavelength ranges, across a 15-diopter focus range, with an 8 mm entrance pupil diameter. COTS lens selection was implemented as a systematic evaluation of catalog lenses after optimization of axial distances.

Custom lens optimization was initially guided by wavefront aberration constraints derived from object-shift 3rd order aberration theory, before traditional variation of surfaces and distances to minimize a performance metric. When a wavefront corrector compensates for field-constant aberrations, the model eyes achieve nominal wavefront RMS below λ/20 across a ∼6.9° circular full field of view for COTS and hybrid lens pairs, limited by vignetting from a self-imposed 27.9 mm clear aperture. For custom lens pairs, the field of view is about 10.2°, limited by aberrations.

A tolerance study indicates that a COTS lens pair for 450-1100 nm can achieve as-built diffraction-limited wavefront error of 0.044 λ @450 nm at 95% production yield. Model eyes with this lens pair yielded telecentricity quantified as image size changed per diopter of 0.1%/D when assembled with a camera and 0.01%/D when assembled with a power meter. The model eye was tested with an optical setup that emulates a fundus camera with Badal focus correction.

Introduction

Optical retinal imaging using a portion of the pupil less than 2 mm in diameter is an essential part of current eye care. Imaging through larger portions of the pupil requires correction of ocular wavefront aberrations beyond defocus, which can be achieved with adaptive optics (AO) [1–6], but it is limited to small fields of view by anisoplanatism [7–12]. AO ophthalmoscopes enable visualization of microscopic structures, providing insights that can advance understanding, diagnosis and treatment of eye disease [13–16].

Despite more than two decades of technological innovation, AO ophthalmoscopy has yet to be adopted as a clinical tool. A practical step toward clinical translation is to ensure retinal images acquired with AO are free of distortion due to non-isotropic sampling, non-uniform scanning velocity [17], sampling jitter [17,18], imperfect scan orientation, and optical distortion [19]. These distortions can be estimated by imaging a periodic distortion target in the retina at the conjugate plane of a model eye [20,21,17].

We recently demonstrated one such telecentric model eye designed to facilitate measurement of distortion as a function of AO ophthalmoscope focus [22]. This model eye comprises an iris diaphragm, a commercial off-the-shelf (COTS) achromatic doublet, and a grid of dots on a glass substrate, delivering nominal diffraction limited imaging over a -5 to +5 diopter (D) focus range with a 2° circular full field of view (FFOV) for an 8 mm pupil. For this pupil diameter, the 2° field of view exceeds the human eye average isoplanatic patch [12] and is comparable to the field of view of most current AO ophthalmoscopes.

As techniques for diffraction-limited AO retinal imaging over larger fields [23–25] and instruments with steering capabilities [26–29] emerge, model eyes with larger field of views are needed. Future model eyes must also accommodate the broader focus ranges enabled by advances in deformable mirror technology [30], liquid lenses [31] and refractive optical designs [32,33]. To address these needs, we propose and demonstrate telecentric model eyes using pairs of COTS and/or custom lenses, chosen to yield larger fields of view and broader focus ranges than previous model eyes [22].

Optical specification

The desired optical specifications for the model eye are summarized in Table 1 and compared with those of a previous model eye [22], both using an 8 mm entrance pupil to serve most AO ophthalmoscopes to date. A focal length of approximately 100 mm enables emulation of a + 5 D hyperopic eye by shifting the distortion target toward the model eye lenses without collision, while keeping the total model-eye length below ∼300 mm when emulating a −10 D myopic eye by shifting the distortion target away from the lenses. The refractive-error range is chosen to be asymmetric, reflecting the higher prevalence of strong myopia relative to strong hyperopia [34–36].

From our previous work [22], shorter focal lengths are likely problematic due to aberration variation across the field of view, larger field-curvature changes with emulated refractive error (vergence), and reduced depth of focus. As with the previous model eye, we impose telecentricity in the distortion target space to preserve image size and minimize optical distortion as vergence changes. We target minimal longitudinal chromatic aberration (LCA) over 450–1100 nm, to span the majority of the ocular media transmission window [37], reducing LCA blur in AO ophthalmoscopes that use broadband illumination.

We target RMS wavefront error below λ/20 for nominal performance and below λ/14 for as-built performance, respectively. The diffraction-limited constraint according to Marechal’s criterion (Strehl ≥ 0.8) will be enforced across the wavelength range and evaluated at a single wavelength per test. Finally, we will only consider COTS lenses with outer diameters of 30 mm or less to ensure compatibility with widely used opto-mechanical rod-mounting systems.

No diameter restriction is imposed for custom lenses. To minimize illumination and resolution variation across the FFOV, we require zero vignetting.

Optical design

The optics forms considered in this work were restricted to pairs of achromatic or near-achromatic doublets, not necessarily identical, to mitigate chromatic aberration, cost, and complexity. Three lens-pair types were considered: two COTS doublets, two custom doublets, and hybrid (COTS and custom). Each pair was optimized to using broadband antireflection coatings in the visible, near infrared, and visible plus near infrared wavelength ranges.

The maximum field of view of a lens with a 27.9 mm clear aperture (inner diameter mounting ring) requiring no vignetting can be calculated using paraxial ray tracing. The clear aperture can be calculated as DA(1+Pf)+uFFOVf, with DA the aperture stop diameter (a focal length in front of the model eye optics), P the largest desired myopic power in diopters, f the lens focal length, and uFFOV the full field of view in radians. Therefore, a model eye with 100 mm focal length optics, 8 mm diameter aperture stop, maximum focus error of 10 D, and 27.9 mm clear aperture diameter, we get a maximum FFOV of 6.8°.

The selection of COTS lenses and the design of custom lenses were performed using the real ray tracing software CODE V (Synopsys, Sunnyvale, CA, USA), with the same processes implementable in Zemax (Ansys, Kirkland, WA, USA). Files for both software packages are provided as Code 1 (Ref. [38]). In these applications, vergence was introduced by setting the object distance to the entrance pupil (in meters) to be the negative of the inverse of the desired optical power (in diopters) and by optimizing the axial distances to achieve distortion-target conjugation.

The four distances used correspond to eyes requiring −10, −5, 0, and +5 D spectacle sphere correction and were set as zoom configurations.

AO ophthalmoscopes typically have one [1,39,2,40–42] or more [43–45,25] pupil-conjugate wavefront correctors. These can compensate for the field-constant aberrations of the model eye optics, thus relaxing the optical performance requirements. We simulated this wavefront correction as a linear combination of Zernike polynomial phase surfaces (fringe Zernike surfaces in CODE V, Zernike fringe sag surfaces in Zemax) up to the 4th radial order, fixing the coefficients for piston, tip, tilt, and defocus to zero, and varying the remaining coefficients to minimize wavefront RMS across wavelengths, vergences and FFOV.

COTS optics

The selection of COTS lenses differed from traditional optical design workflows, which vary surface curvatures, thicknesses, and spacings to meet performance targets. Instead, we tested pre-manufactured lenses drawn from existing lens catalogs (expanded by the manual addition of the latest lenses), restricting optimization to axial positions and aperture stop placement. This process was automated with macros in CODE V and applied to lenses by Thorlabs (Newton, NJ, USA) and Edmund Optics (Barrington, NJ, USA).

As a first step, we narrowed the pool of achromatic lenses to those with an effective focal length (EFL) ≥ 120 mm and an outer diameter of approximately 30 mm and exported this subset to a CODE V buffer data file for ray-tracing evaluation. We then narrowed the possible achromatic lens pair combinations using the thick lens equation to retain only pairs capable of a combined 100 mm EFL and positive lens separation for all vergences. The remaining lens pairs were evaluated in all four orientation permutations.

After optimizing the axial distances among the aperture stop, lenses, and image planes, only those pairs that satisfied the RMS wavefront and telecentricity specifications across a 6° circular FFOV, the full range of vergences, and the wavelength band were retained.

Custom optics

Ideally, the model eye optics would achieve desensitization of field curvature and other aberrations with respect to retinal image conjugate shifts across the 15 D focus range. This aberration desensitization to conjugate-position changes has been extensively studied because of its broad implications for optical systems. Originally developed by Smith [46], aberration change equations have been revisited by Wynne [47], Hopkins [48], Sasián [49], and Mahajan [50].

These equations apply to zoom systems, where adjusting the spacing between optical elements changes magnification and field of view without altering the object and image distances [51–54], as well as optical systems with fixed components in which depth of field is increased to keep a range of conjugates in focus with acceptable image quality [55–58]. Here, we adopt Hopkins’ formulation because it enables the direct formulation of geometric constraints that are amenable to real ray tracing software implementation.

Hopkins showed that within the paraxial regime, the Seidel aberration coefficients Si∗ after axial conjugate shifting are related to the Si coefficients before the shift by the matrix equation, (1)(S1∗S2∗S3∗S4∗S5∗S6∗S7∗S8∗S9∗)=[14δE¯6(4δE¯)22(δE¯)24(δE¯)3(δE¯)4δE¯3(δE¯)23(δE¯)3013δE¯δE¯3(δE¯)2(δE¯)30δE¯2(δE¯)200102δE¯(δE¯)200δE¯00010000000001δE¯00000000100000000012δE¯(δE¯)200000001δE¯000000001](S1S2S3S4S5S6S7S8S9)Spherical aberrationComaAstigmatismPetzvalDistortionPupil sph. aberration(Pupil distortion − coma)(Pupil astig. −astig.)(Pupil coma−Distortion). with δE¯=(y∗−y)/y¯ being the object-shift factor (ratio of marginal ray height change to chief ray height) and H=Rpupil(HFOV) being the Lagrange invariant (entrance aperture radius Rpupil times half of the field of view, HFOV).

An important assumption in Hopkins’ derivation is that the Lagrange invariant remains constant during conjugate shifts. This means that the object size and image distance are changing to maintain the same angular field of view in object space and numerical aperture (NA) in the image space when the object distance is varied. Given that the Lagrange invariant constancy is met due to telecentricity, this formulation is well-suited for the design of the proposed model eyes.

Monochromatic optimization constraints

When using a single COTS achromatic doublet, field curvature and its variation with vergence due to astigmatism become the limiting aberrations for circular FFOVs greater than about 2°, and for strongly myopic eyes, with coma as the next limiting aberration. Therefore, we provide mathematical conditions for implementation as constraints in ray tracing software to minimize the change in these aberrations with vergence using the Hopkins object-shift equation.

The change in astigmatism due to conjugate shift is (2)S3∗−S3=[2S5+S9]δE¯+S6(δE¯)2, with S9=HΔ(u¯2) and Δ(u¯2) being the difference between the squares of chief ray angle in the image (ui―) and object (uo―) space, respectively. This formula holds both for a single surface and for groups of surfaces, because in the sum of surface contributions, intermediate terms cancel out, leaving only the first and last surface terms. Therefore, noting that the telecentricity condition means ui―=0 and, by definition, uo―=HFOV, we have that, (3)S3∗−S3=[2S5−Rpupil(HFOV)3]δE¯+S6(δE¯)2.

If we now impose that astigmatism remains constant across conjugate shifts to keep field curvature constant across the model eye focus range, the following two conditions must be met, (4)S5=Rpupil(HFOV)3/2, and (5)S6=0.

This second condition means that pupil spherical aberration must be zero, which also makes image distortion invariant with conjugate shift.

If in addition, we want to minimize coma change with conjugate shifts (i.e., S2∗=S2), according to Eq. (1) we have to impose that, (6)[3S3+S4+S8]δE¯+[3S5+2S9](δE¯)2+S6(δE¯)3=0, with S8=HΔ(uu¯) and S9=HΔ(u¯2). As before, we can invoke telecentricity in image space (i.e., ui―=0).

If we then consider the minimization of coma variation around an emmetropic model eye (i.e., object at infinity, object space marginal ray angle, uo=0), we have that Δ(uu¯)=uiui―−uouo―=0. Taking this into account and applying the conditions for invariant distortion and astigmatism discussed above, we have that Eq. (6) becomes (7)(3S3+S4)δE¯−Rpupil(HFOV)32(δE¯)2=0

If the quadratic term in δE¯ can be considered negligible in comparison with the linear term (the largest value δE¯ could be is 0.23 within the desired focus range), then the condition for minimum coma variation with conjugate shift in the model eye around the emmetropic eye reduces to, (8)3S3+S4=0, where the quantity 3S3+S4 is known as tangential astigmatism.

In practice, it might not be possible to meet any of these conditions, let alone all of them simultaneously and while keeping S3 small. We can, however, incorporate these conditions on the ray tracing software merit function, to guide the optimization algorithm through the lens parameter search space.

Finally, Petzval curvature is defined as the following sum across lens surfaces, (9)S4=−∑l⁡Hl2clΔ(1/nl) where cl is the radius of curvature of surface l and nl is the refractive index of the material after surface l. In our case, because the eye is telecentric, the Lagrange invariant H does not change with vergence, and neither do the lens curvatures or refractive indices. Therefore, Petzval curvature does not change with image conjugate.

Polychromatic optimization constraint

Achromatic doublets do not completely cancel LCA, and therefore, the distance between the lenses and the aperture stop of the model eye should be adjusted to achieve telecentricity at each wavelength of interest. We do, however, aspire to minimize the variation of LCA and transverse chromatic aberration (TCA) with vergence by using the Hopkins object-shift transfer matrix for LCA and TCA [48], (10)(C1∗C2∗C3∗C4∗)=[12δE¯(δE¯)2−δE¯01δE¯000100001](C1C2C3C4)LCATCAPupil LCA(TCA−Pupil TCA).

Hence, the variation of LCA with vergence is given by (11)C1∗−C1=(2C2−HΔ(δn/n))δE¯+C3(δE¯)2, which can only be zero in the absence of pupil LCA (i.e., C3=0), and when TCA in the image plane meets the condition (12)C2=RpupilHFOV2∑l⁡Δ(nl,long−nl,shortnl).

Since the object and image space are in air, the summation of the term Δ(nlong−nshortn) across all surfaces of the system is zero, this means that we need to impose the absence of TCA in the image plane (i.e., C2=0). On the other hand, the variation of TCA only depends on the pupil LCA. Therefore, if we seek to deliver a model eye with no image LCA or TCA variation with vergence, then the optics should be free of pupil LCA and image TCA.

Real ray tracing optimization

The custom lens design used the software’s default RMS wavefront merit function, modified as follows. First, we strictly enforced the effective focal length to be in the 90-110 mm range by using the EFL operator in CODE V (EFFL in Zemax) and zero chief ray angle on the image plane to achieve telecentricity by using the AOI operator in CODE V (RAID in Zemax). Then, we imposed the following constraints derived from the object-shift conditions as equally weighted targeted constraints, which can be seen in the optimizers and merit functions in Code 1 (Ref. [38]): 1.Distortion at field edge to be (HFOV)2/4, as per Eq.

(4), implemented using the operator DIY in CODE V (DIST in Zemax).2.Zero pupil spherical aberration, as per Eq. (5), defined as [49], (13)S6=−∑p⁡[np(up―+y¯R)]2y¯Δ(u¯n)=03.Zero tangential astigmatism, as per Eq. (8), implemented in CODE V using the operator TAS (3×ASTI + FCUR in Zemax).4.Zero pupil longitudinal chromatic aberration, as per Eq.

(10), defined as [49], (14)C3=∑p⁡[n(u¯+y¯R)]y¯Δ(nlong−nshortn)=0. where nlong and nshort are the refractive indices of the material after the surface at the longest and shortest wavelength in the desired spectral bandwidth.5.Zero transverse chromatic aberration, as per Eq. (12), implemented using the operator LAT in CODE V (LACL in Zemax).

Following merit-function customization, the design steps were carried out as follows: •Use best performing COTS design as initial system configured with 6° FFOV, and minimize merit function by varying radii of curvature, air gaps, and glasses across wavelengths, refractive errors and field points.•Flip lenses to evaluate all orientation permutations and retain configuration with lowest merit function value.•Replace fictitious glasses with real glasses using either the glass fitting software functionality or by consulting glass catalogs, and then, repeat optimization.•Remove weighted constraints from merit function and insert a Zernike polynomial phase surface in the model eye aperture stop, to emulate the field-constant wavefront correction of a continuous-sheet deformable mirror, and re-optimize air gaps, leaving lens surface curvatures and glasses unchanged.•Repeat previous step, gradually increasing the FOV until the design reaches the nominal ∼1/20 wave wavefront RMS threshold, increasing lens thicknesses when necessary, to prevent excessively thin lens edges that are difficult to manufacture.

Hybrid optics

The custom lens in the hybrid lens pairs was designed by applying the same workflow used for the custom lens pair, while leaving the COTS lens unchanged.

Final designs

The selection and optimization of lenses yielded the designs shown in Fig. 1. COTS, hybrid and custom designs are all compact with lens separation being ∼0.1 mm.

Also, whenever using at least one COTS lens, the best performing lens pairs always have crown glasses facing each other, while the custom lens pairs perform best when flint glasses face each other.

Available COTS achromatic lenses with near-infrared anti-reflection coating (600-1050 nm) were found to perform worse than those with the visible coating (425-675 nm) and broadband coating (400-1000 nm), and thus, were discarded. Figure 1 shows that the COTS lens pairs achieve the target performance over a 6.9° FFOV when used with the wavefront corrector.

The hybrid lens pairs (specifications in Table 2) can deliver the desired performance when used with a wavefront corrector, but over a slightly larger FFOV than the COTS-only pairs, owing to the additional degrees of freedom provided by the custom lens surfaces and diameter. The custom doublet in these hybrid pairs is achromatic and corrected for spherical aberration and coma, just as the COTS lens in the pair.

Finally, the custom doublet pairs (specifications in Table 2), optimized with an unlimited clear aperture, meet specifications over the largest fields of view, as expected given the greater number of adjustable lens parameters. Unlike the COTS and hybrid designs, these custom doublet pairs balance chromatic aberrations together, rather than individually.

COTS optics

Given that both COTS lens pairs achieve comparable performance in the visible range, we use the 450-1100 nm pair (Edmund # 49376 and 49378), oriented as per Fig. 1 to characterize its as-built performance and optical distortion, prior to comparing assembly strategies and testing in an optical setup that emulates an ophthalmoscope with Badal focus correction.

As-built performance simulation

To evaluate the real-world performance of the broadband COTS lens pair, we simulated random manufacturing and assembly errors within vendor-specified tolerances (Table 3), and plotted the cumulative performance distributions (Fig. 2) with and without a wavefront corrector at 450 nm, with performance at longer wavelengths being superior. In the simulations without a wavefront corrector, image defocus was the sole compensator.

When a wavefront corrector was included, both image defocus and the previously discussed Zernike coefficients were used. The plots show that, in the absence of a wavefront corrector, performance degrades as the field of view increases and the simulated refractive error increases (i.e., worse for more myopic eyes). The correction of field-constant aberrations changes both trends, so the model eye performs best at mid-field, with nearly identical performance in the central and outer regions of the field of view and substantially reduced sensitivity to focus error.

Importantly, the wavefront corrector shifts the yield curves toward RMS values below the specified target of λ/14 at 450 nm, with performance at all other wavelengths being better. Plots can be generated from the ray-tracing scripts and macros provided in Code 1 (Ref. [38]).

Optical distortion

Real ray tracing shows that the distortion of the 450-1100 nm COTS optics, in a coordinate system centered on the optical axis, is well described by the rotationally symmetric third order polynomial used in our previous work [22], (15)Δx≈Ax(x2+y2)Δy≈Ay(x2+y2), where A=6.12×10−6P−8.00×10−5 varies with model eye lens, the refractive error P is in units of diopters (negative P values correspond to myopic eyes), (x,y) are the field of view coordinates and (Δx,Δy) represent distortion, both in units of degrees. This polynomial, which corresponds to barrel image distortion, brings points across the field of view closer to its center, as shown in Fig. 3 below.

Assembly

The model eye consists of an iris diaphragm as the aperture stop, two COTS achromatic doublets housed in a Thorlabs 30 mm lens tube, and a distortion-target grid formed by a square array of circular black dots 125 ± 2 μm in diameter with 250 ± 1 μm center-to-center spacing on a glass substrate (Thorlabs part # R2L2S3P2). The two doublets were mounted in the lens tube with a nominal on-axis separation of 0.30 mm, provided by a retaining ring. This retaining ring also functions as the system aperture stop, with a diameter of 27.9 mm.

Two alternative methods, depicted in Fig. 4 below, were used to reach telecentricity by adjusting the distance between the iris diaphragm and the lens tube with the lenses: •Bring the iris diaphragm into focus when imaged through the model eye lens pair with an auxiliary camera focused at infinity [22,59].•Make the power transmitted through the iris diaphragm insensitive to on-axis shifts of a point source (e.g., the tip of a single-mode optical fiber) on the opposite side of the lens pair.

The accuracy of the first approach depends on the depth of focus of the auxiliary camera and the subjective judgement of best focus (perhaps objective, if using a sharpness metric). These methods were compared by measuring the change in distortion target magnification over the axial range over which the distortion target appears in focus when closing the iris diaphragm down to 1 mm. The results, plotted in Fig.

4, indicate that, in this case, the power meter approach is superior to the use of an auxiliary camera with subjective best focus judgement.

Demonstration

The model eye was assembled and tested with a refractive optical system akin to an ophthalmoscope without adaptive optics but with Badal focus correction (see Fig. 5). This instrument was used to capture images of the distortion target at various vergences.

The optical setup was assembled using an opto-mechanical rod system to remove the need for mechanical degrees of freedom to align the model eye in relation to the ophthalmoscope [22]. The emulated ophthalmoscope consists of a monochromatic camera BFS-U3-122S6M-C by Teledyne FLIR (Willsonville, OR, USA) and four COTS achromatic doublets. The lenses for the ophthalmoscope were selected based on the availability of lenses in our lab that could deliver the desired FFOV and spatial sampling, with minimal field curvature and vignetting.

The lens pair closest to the ophthalmoscope’s exit pupil is identical to that of the model eye itself, and it is followed by a 1.5 D negative singlet placed in a retina conjugate plane as a coarse field-flattener. This singlet is decentered by ∼1 mm because its diameter is larger than the spacing between the rods in the optomechanical mounting system, and therefore, the lens was resting on two adjacent rods. The intermediate retina plane was relayed to the camera sensor by an afocal telescope formed by a pair of 100 and 50 mm focal length achromatic doublets (AC254-100-AB-ML and AC254-050-B-ML, by Thorlabs).

The entrance pupil of the model eye was placed to coincide with the exit pupil of the ophthalmoscope. The distortion target was back-illuminated with 680 nm light scattered from paper illuminated by a single-mode optical fiber located approximately 50 cm from the paper to achieve relatively uniform illumination across the distortion-target field of view. Images of the target were captured with the iris diaphragm set to 6 mm to mitigate aberrations not corrected due to the absence of a wavefront corrector.

The target was positioned at four distances from the model eye lenses (see Fig. 3 diagrams), with the camera sensor brought into focus through Badal correction, that is, by adjusting the spacing between the ophthalmoscope lens pair (near the iris) and the field flattening singlet.

The images of the dot grid (Fig. 6) were processed with the algorithms described previously [22]. The estimated distortion maps for four model eye foci are shown on the leftmost panels of Fig.

7, with the origin of the ideal grid defined at the top left corner of each panel. The middle plots in this figure show the residual distortion after correcting tip, tilt, and shear, as well as the scaling difference between the x- and y-axis. The distortion changes with focus, and the distortion field node shifts towards the right side of the field.

This shift arises because of field-flattening lens decentering (∼1-2 mm). Finally, the plots on the right show the residual distortion after correction of using a 3rd order polynomial fitting, which might be due to mid-spatial frequency surface errors, errors due to image pixelation, or others. Fitting errors for the grid dots and foci when estimating the distortion using 3rd and 5th order polynomials in Fig.

8, show that for this ophthalmoscope, a 5th order polynomial reduces fitting errors by approximately 20-25% relative to the 3rd order polynomial.

The magnification changes with vergence shown in Fig. 9 are an order of magnitude larger than those measured in the model eye alone (0.2 vs. 0.01% per D), illustrating the utility of the model eye to measure optical distortion and telecentricity [60].

In addition, the near rotational symmetry of the optical setup should produce negligible shearing distortion and no shearing changes with vergence. The data, however, indicates a maximum shear change of 0.075°, which for an image with 1,000 pixels across the FFOV, would correspond to ∼0.1 pixel between the first and last image row or column, which is within measurement error.

Conclusions

Telecentric model eyes were designed and demonstrated for measuring image distortion in AO ophthalmoscopes, using combinations of COTS and custom achromatic or near-achromatic lens pairs. Model eyes spanning the visible and near-infrared spectral ranges were designed to achieve nominal and as-built diffraction-limited performance across a 15 D focus range through an 8 mm pupil when a wavefront corrector is employed.

The COTS lens selection consisted of a systematic evaluation of achromats in commercial catalogs, followed by optimization of axial spacings. The custom lens design optimization was initialized with the best COTS lenses, after which the surfaces, thicknesses, and glass types were allowed to vary, first guided by analytically derived constraints, and then using an ideal wavefront corrector. Both approaches converged to a final configuration in which the lenses are separated by approximately 0.1 mm.

The model eyes achieved nominal wavefront RMS below λ/20 across a circular full field of view of 6.9 and 7.0° for COTS and hybrid pairs, respectively, both limited by the optics maximum 30 mm outer diameter and no vignetting. Removing these restrictions allows fully custom lens pairs to deliver 10.2°–10.6° fields of view, with performance limited by residual wavefront aberrations.

A tolerancing study shows that with an idealized wavefront corrector, the ideal broadband COTS lens pair for 450–1100 nm delivers as-built diffraction-limited performance across the entire field of view when lenses are within the manufacturer’s stated tolerances. Two practical assembly methods for achieving telecentricity were demonstrated, using a camera and a power meter. Finally, testing with an optical setup emulating a fundus camera with Badal focus correction shows that the model eyes can be used successfully over the desired field of view and focus range.

Supplemental information