Refractive adaptive optics scanning light ophthalmoscope with fast 2D MEMS scanner

Abstract

We describe a refractive adaptive optics scanning light ophthalmoscope designed for small animal imaging through a 1.8 mm diameter pupil. The optical setup, based on a search of achromatic doublets through multiple lens catalogs, consists of a sequence of modified afocal relays that deliver diffraction-limited imaging in pupil and retina conjugates. Real ray tracing is used to compare imaging performance when correcting large focus errors using a pupil conjugate wavefront corrector, a traditional Badal optometer, and a modified Badal optometer. Polarization control, focal length selection, and systematic lens tilting are explored for mitigating reflections with minimal imaging performance degradation. A 2-dimensional optical scanner with a 29.2 kHz resonant frequency around one axis and low dynamic surface distortion allows doubling the frame rate of prior instruments and simplifies the optical setup. Scanner orientation and trigger electrical signals are used to correct sinusoidal image warping and line sampling jitter. The instrument is demonstrated by imaging mice under 800 nm illumination with two reflectance detection modalities: confocal and quadrant non-confocal.

1. Introduction

Non-invasive optical retinal imaging of animals plays a central role in advancing the understanding [1–3] and treatment [4–8] of human diseases that lead to irreversible vision loss. The most widely used paradigms for animal retinal imaging in research settings are ex vivo microscopy and in vivo ophthalmoscopy. Microscopy provides superior axial and transverse resolution but requires animals to be sacrificed to harvest their retina. Conventional ophthalmoscopy provides inferior transverse resolution due to the small numerical aperture of conventional ophthalmoscopes, but allows repeat structural and functional imaging, thus reducing the number of animals needed per study. Imaging through larger numerical apertures with adaptive optics (AO) wavefront aberration correction can deliver ophthalmoscopes with microscopic resolution.

AO ophthalmoscopy in human subjects has been used to study numerous eye [9–15] and neurodegenerative conditions [16–18]; however, in animals the technique remains at the proof-of-principle stage, with demonstrations to date limited to non-human primates [19–24], pigs [25,26], ground squirrels [27,28], tree shrew [29], rats [30,31], and mice [32–43]. Mice are critically important to vision science and the development of novel therapies for blinding conditions, due to the ever-growing number of genetic tools to fluorescently label cellular and sub-cellular structures of interest [44,45]. Initial implementations of AO ophthalmoscopes for mice imaging have proven challenging to operate and have not yet delivered the full resolution potential when using the full numerical aperture of the mouse eye [33,36,39,42,43,46]. AO ophthalmoscopes built with tilted spherical mirrors [33,42] are appealing because metallic coatings do not introduce dispersion that would degrade coherent imaging [36,42] or multi-photon imaging [34,47–50], but have modest fields of view. Refractive AO ophthalmoscopes [39,46,51–56], on the other hand, require less frequent optical alignment and can facilitate reaching retinal locations of interest through larger fields of view [46], but their reflectance imaging can be degraded by undesired lens reflections.

Here, we describe a refractive AO scanning light ophthalmoscope (AOSLO) designed to evaluate a 2-dimensional (2D) optical scanner as an alternative to a pair of orthogonal one-dimensional (1D) scanners, and to revisit popular strategies for mitigating lens reflections in imaging detectors. This scanner reduces instrument complexity and wavefront aberrations, while increasing light throughput, lowering dynamic surface distortion [57], and doubling the resonant frequency of current 1D scanners [58]. The resulting higher frame rate could benefit functional retinal changes in response to light stimuli in calcium ion concentration [34,49,59,60], photoreceptor reflectivity [61,62], blood cell velocity [37,63–65], and others. After tuning the 2D scanning parameters to mitigate undesired oscillations about the non-resonant axis, we used scanner-orientation signals from the control electronics to correct sinusoidal distortion and sampling jitter along the resonant scanning direction, employing three distinct strategies [66,67]. Finally, testing the refractive AOSLO in mice illustrates the limitations of polarization control and lens tilting to mitigate lens reflections.

2. Optical design

2.1. Specification

The AOSLO optical specifications, summarized in Table 1, include imaging through a 1.8 mm diameter aperture stop over a 2.5° square field of view. The optical design targeted diffraction-limited retinal imaging (Marechal’s criterion) over prescribed focus, source/detector vergence, and spectral ranges, without pre-compensation of species-specific ocular aberrations, such as spherical aberration or longitudinal chromatic aberration (LCA) [33,39,68]. The 50 diopter (D) focus range was chosen to allow traversing the mouse retina thickness (230 µm/(2.6 mm)² ∼ 34 D) and compensating the estimated 15 D myopic refractive error in mice [69]. The source (and detector) vergence range was selected as ±29 D because it generates an axial shift, which is equivalent to ±1.5 D in human eyes (assuming ∼2 D of LCA and ∼1 D retinal thickness) [70]. Wavefront aberrations were also minimized at all pupil planes to avoid degradation of wavefront sensing and correction. The optical setup was optimized for 800 nm light, as this is commonly used for reflectance imaging and registration of fluorescence images [30,33,71]. The instrument was designed using commercial off-the-shelf achromatic lenses with 400-1000 nm antireflection coating. Clearance between the eye and the lens closest to it was ≥ 85 mm to allow for the insertion of a tilted quarter waveplate and anesthesia breathing mask.

Table 1. Desired Refractive adaptive optics scanning light ophthalmoscope optical specifications.

Specifications	Values
Pupil diameter at the eye	1.8 mm
Diffraction-limited full field of view (FFOV)	2.5°
Pupil RMS spot size at 800 nm a	< 1 Airy disk diameter for all vergences & FFOVs
Retina wavefront RMS at 800 nm	≤ λ/14 for all vergences & FFOVs
Badal focus range	-10 D to +66 D
Source vergence range	-29 D to +29 D
Wavelength range	400–1000 nm
Eye clearance	≥ 85 mm

^a At the following conjugates: 2-dimensional optical scanner, deformable mirror and eye.

2.2. Two-dimensional optical scanner

Traditionally, raster scanning in AOSLOs is achieved using two orthogonal optical scanners, either with their surfaces close to the same single pupil conjugate, or precisely at two pupil conjugates, separated by an optical relay [33,72]. Two-dimensional scanners allow simplifying the optical setup by removing the need for one pupil relay, while also increasing its light throughput [73,74]. Here we evaluate a new type of micro-electro-mechanical system (MEMS) 2D scanner (S13989-01H by Hamamatsu, Shizuoka, Japan) with 1.2 mm diameter clear aperture, a 29.2 KHz resonant frequency around one axis, a non-resonant axis with a maximum frequency of 100 Hz, and almost negligible dynamic surface distortion [57].

2.3. Pupil relays

Like most AO ophthalmoscopes, the proposed instrument consists of a sequence of pupil plane relays, which here were chosen to be modified afocal telescopes [75]. Each relay consists of two achromatic doublets, to mitigate spherical aberration and LCA, while also facilitating reproducibility through their low cost and broad availability. The flatter outer lens surfaces were oriented towards the nearest pupil plane to prioritize pupil spherical aberration correction, because the deformable mirror can correct the spherical aberration at the final retina conjugate.

2.4. Complete optical setup

The portion of the optical setup evaluated with real ray tracing consists of three individually optimized pupil relays (Fig. 1), that take the 5.2 mm diameter circular entrance pupil onto a 1.0 mm diameter area at the 2D scanner, then to a 7.1 mm diameter circular area over the deformable mirror and, finally, onto a 1.8 mm diameter at the AOSLO exit pupil, which is where the entrance pupil of the eye is placed during imaging. The lenses for each relay were selected by evaluating all possible combinations of achromatic doublets from Edmund Optics (Barrington, New Jersey, USA) and Thorlabs (Newton, New Jersey, USA), that provided the desired pupil magnifications and kept the total relay length below one meter. Lens pairs that met these conditions were further evaluated for wavefront aberrations at the pupil and retina conjugate planes, while enforcing afocality for 800 nm light. For each relay, only the distance between the pupil planes and the lenses were varied to minimize wavefront aberrations while retaining pupil conjugation. That is, whenever the distance between the entrance pupil of a relay and the first lens is varied in the ray tracing software, the distance between the second lens and the exit pupil plane is also adjusted accordingly. Hereon, were refer to these pupil planes as modified afocal relays. When the distance between the lenses is varied in the final relay, we refer to this focus correction as “modified Badal optometer”, which unlike the traditional Badal correction, it does not preserve retinal and pupil magnification. The lenses that provided the best performance for the three relays are listed in Table 2.

Fig. 1.

Sequence of optical afocal relays optimized for pupil (P) imaging at all conjugates and retinal imaging (infinite conjugate) between the first and last conjugate. Note that the distance between each achromatic doublet and the adjacent pupil plane is not the focal length of the lens (f_i), but rather the focal length plus or minus a distance Δ_i.The Δ' in the last telescope corresponds to the distance change from Badal optometer

Table 2. Optimal achromatic doublets for refractive AO ophthalmoscope, all by Edmund Optics (Barrington, New Jersey, USA).

Surface	Part #	Focal length at 587.6 nm (mm)	Distance to next surface (mm)
Entrance pupil			298.5
Lens 1	49366	250	298.3
Lens 2	49328	50	45.7
2D MEMS scanner			53.1
Lens 3	49329	60	459.2
Lens 4	49369	400	604.4
Deformable mirror			580.9
Lens 5	49369	400	500.6
Lens 6	49796	101.6

Source vergence was induced by changing the point source object distance and compensated by a paraxial lens (i.e., idealized lens free of aberrations) with varying focal length at the AOSLO exit pupil. Refractive error was induced with a paraxial lens at the AOSLO exit pupil and compensated by varying either the deformable mirror surface or the distance between the lenses in the final optical relay to operate the modified Badal. Due to the rotational symmetry of the optical system, the optimization and performance evaluation used only three points in the field of view along the same radius with normalized coordinates 0, $1/\sqrt{2}$ and 1. The deformable mirror surface was modeled as a linear combination of 9 Zernike fringe sag polynomials in the optical design software Zemax (Kirkland, Washington, USA). These Zernike coefficients, together with the distances between lenses and pupil planes, were varied during the optimization.

The numerical aperture of the pupil conjugates was set to of 0.01 radians in object space (i.e., light source), which corresponds to a 2.5° full field of view in the retina. The optimization simultaneously minimized wavefront root-mean-squared (RMS) at all pupil planes, and at three scan angles at each of the three retina conjugates for 800 nm light. The optical system was optimized for an eye with zero focus error, and then evaluated under various refractive errors corrected by either the deformable mirror or the modified Badal, with the retinal conjugate performance summarized in Fig. 2 below. The asymmetry in the refractive error range over which diffraction limited performance can be achieved shows that deformable mirror correction is better for hyperopic eyes, while the modified Badal is better for myopic eyes.

Fig. 2.

Adaptive optics ophthalmoscope retina spot diagrams for a 1.8 mm exit pupil, illuminated with 800 nm light, for three source vergences and refractive errors corrected with a pupil conjugate deformable mirror (top) or a Modified Badal optometer (bottom).

Spot diagrams in Fig. 3 show the imaging performance of pupil conjugate planes, with only the spots at the pupil of the eye changing with Badal correction. The increase in spot size with field coordinate at the MEMS scanner is due to the defocus that results from out-of-plane tilting of the scanner surface, which is most noticeable in this pupil conjugate due to the smaller pupil size and larger angle of incidence than at the deformable mirror. As with the retina conjugate, the pupil magnification, and thus the Airy disk size, change with Badal correction. It is worth noting that the retina and pupil magnification changes due to the modified Badal focus means that the retina sampling (i.e., pixel spacing), as well as the transverse and axial resolution will change proportionally.

Fig. 3.

Spot diagrams for adaptive optics ophthalmoscope pupil conjugates for 800 nm light and retina focus errors corrected with a pupil conjugate deformable mirror or a Modified Badal optometer.

3. Reflection mitigation

Current anti-reflection coatings for visible or near infrared wavelength ranges can achieve reflectivity as low as 0.25%. This reflectivity, however, is unacceptably high for reflectance retinal imaging due to the low retinal backscattering [76]. Hence the need for additional strategies to mitigate reflections from refractive surfaces common to the illumination and imaging optical paths of the ophthalmoscope and the eye, with the tear film providing the strongest reflection (∼14%) [77]. A reflection mitigation approach common in ophthalmic instruments, is the use of non-overlapping illumination and imaging aperture stops [78], which requires the alignment of the ophthalmoscope so that corneal reflections get diverted from the imaging stop. Unfortunately, this is not practical, or even possible, in AO ophthalmoscopes, as they use a large portion of the pupil of the eye for imaging, and sometimes, illumination. Hence, the need for alternative reflection mitigation methods, such as the ones discussed next.

3.1. Lens tilting

Lenses are often tilted in research-grade vision science and retinal imaging instruments to divert undesired reflections away from the aperture stop. This tilting is done empirically, without quantification of the resulting degradation of imaging performance predicted by nodal aberration theory [79,80]. This tilting introduces, among other aberrations, field-constant and linear astigmatism, of which only the former can be corrected with a pupil conjugate wavefront corrector, such as a deformable mirror. Here, we propose tilting the first lens in a pupil relay as needed to avoid reflections that reach the imaging sensors, and then determine the second lens tilt angles necessary to compensate for the aberrations induced by the tilting of the first lens. We have previously shown that this approach can be used to mitigate constant astigmatism, linear astigmatism and/or other wavefront aberrations [81]. Only lenses in the pupil relay closest to the eye were tilted, as preliminary experiments showed that they introduce the strongest reflections, with the resulting wavefront RMS plots for retinal imaging shown in Fig. 4. The focus range in each scenario is limited by one or more of four possible factors, namely: the bumping of the relay lenses against each other; vignetting by the lens clear aperture; pupil diameter at the eye increasing beyond 2 mm to avoid vignetting in a typical mouse eye with dilated pupils; or wavefront aberrations not meeting Marechal’s criterion [82]. The plots for the traditional Badal pupil relay show that not tilting the lenses provides the largest focus range, both with the deformable mirror and Badal correction, with the range being limited by vignetting or the lens bumping against each other. Tilting the lens closest to eye by 4.5° to eliminate reflections, reduces the focus range by a factor of three, with aberrations being the limiting factor when using the deformable mirror or the Badal for refractive error correction. If in addition, the other lens is tilted by 0.29° around the same axis and 0.36° around the orthogonal axis, we see the diffraction-limited focus range increase by ∼6D, which is not enough to restore the larger range seen when neither lens is tilted. The modified Badal configuration with no lens tilting provides the lowest wavefront RMS at the retina, although it results in a narrower focus range than with the traditional Badal due to vignetting and/or the lenses bumping into each other. Also, the lens closest to the eye in the modified Badal only needs to be tilted by 3.5° to remove reflections, thus providing a larger focus range than the equivalent scenario with the traditional Badal. Finally, tilting the second lens by 0.09° degrees along the same axis and 0.35° degrees around the orthogonal axis reduces the wavefront RMS, further increasing the focus range to almost match the no tilt case. The corresponding pupil conjugate comparison shows a slightly worse performance for modified Badal in Fig. 5, although all configurations have RMS spot diameter under 1.0 airy disk unit.

Fig. 4.

AOSLO wavefront RMS for 800 nm light in retina conjugates as a function of focus correction using a deformable mirror or a traditional/modified Badal optometer. Wavefront RMS values correspond to the maximum circular field of view for each source vergence, with the values below the diffraction limit being limited by vignetting, lens collision and/or exit pupil diameter becoming larger than 2 mm in diameter.

Fig. 5.

AOSLO RMS spot size for 800 nm light when imaging the entrance pupil onto the exit pupil as a function of focus error correction using traditional and modified Badal optometer.

3.2. Lens focal length

Another reflection mitigation strategy is the choice of lens focal lengths. That is, when multiple lens pairs can deliver the same afocal relay magnification, but with different total optical path lengths, usually, only the minimization of wavefront aberrations and instrument size are used as selection criteria. Here, we also consider the amplitude of the reflections as a function of the lens focal length. To demonstrate this approach, we evaluated three afocal telescopes with 1:4 magnification using the illumination design software LightTools (Synopsys, Sunnyvale, CA, USA). The first lenses had focal lengths of 100, 250 and 400 mm, and the energy reflected by the second lens captured by the entrance pupil was 0.013, 0.0040, and 0.0028%, respectively, indicating that the reflection is approximately inversely proportional to the afocal relay length as seen in Fig. 6.

Fig. 6.

Irradiance/Illuminance (W/mm²) of reflections from the second lens (green beam) observed at the afocal relay stop aperture with non-sequential ray tracing. The irradiance reduces with longer focal length with an assumed 1W of incidence radiance flux (red beam).

3.3. Polarization control

An alternative strategy for mitigating reflections in refractive ophthalmoscopes is to use orthogonal polarizations for illumination and imaging, an approach common in ophthalmic wavefront sensing [52,83]. This approach, however, creates retinal image artifacts due to cornea form birefringence [84,85] and retina form birefringence [86–88] [84,89–91], that can complicate image interpretation. A common variation of this approach uses an additional quarter waveplate between the ophthalmoscope and the eye, with its refractive index axes at 45 degrees with respect to the impinging linear polarization and with its surface slightly tilted with respect to the illumination to divert reflections. In this way, lens reflections are mitigated, but those from the cornea or crystalline lens are not, thus resulting in imaging artifacts due to ocular birefringence.

4. Complete instrument

The AOSLO, depicted in Fig. 7 below, was built using a non-contact optical approach for axial placement of optical elements [92]. The instrument incorporated four light sources: a 635 nm laser diode (S1FC635PM; Thorlabs, Newton, New Jersey, USA) used for optical alignment, an 800 nm central wavelength superluminescent diode (SLD; Superlum Diodes Ltd., Carrigtwohill, Co. Cork, Ireland) with an optical bandwidth (FWHM) of 9 nm for reflectance retinal imaging, and two 850 nm central wavelength SLDs (Superlum Diodes Ltd.) with 10 nm optical bandwidth for creating a wavefront sensing beacon at the retina and to calibrate the wavefront sensor and AO control. The 800 nm SLD delivered an average power of 570 µW at the pupil of the eye, while the 850 nm wavefront sensing beacon delivered ∼40 µW. Both light sources were ON only during the imaging portion of each resonant scan cycle and OFF otherwise. The three light sources were made coaxial by ensuring that their beams were centered when seeing by two cameras focused at different conjugates (near- and far-field), when a temporary fold mirror was inserted in the system.

Fig. 7.

Layout of refractive AOSLO with a fast 2D MEMS scanner, a Shack-Hartmann wavefront sensor and a deformable mirror, all conjugated with the pupil of eye. The separation between the two lenses closest to the eye can be varied to focus on the retina as needed (Badal optometer). The pupil image on the top right shows iris diaphragms of the three light sources in focus, showing that they are in optical conjugates. The bottom left picture shows the edges of the optical scanner and deformable mirror in focus, showing that they are in optical conjugates.

The adaptive optics system consisted of a continuous sheet protected silver coated deformable mirror with 97 actuators and a 7.5 mm clear aperture (DM97-08 by Alpao SAS, Montbonnot, France) and a custom Shack-Hartmann wavefront sensor (SHWS) built with a lenslet array with 7.8 mm focal length and 300 µm pitch (Adaptive Optics Associates, Cambridge, Massachusetts, USA) and Aqua CCD camera from Qimaging (Surrey, British Columbia, Canada) focused to account for focal shift [93].

The 800 nm confocal reflectance images were captured with a photomultiplier tube (PMT; H7422-50 by Hamamatsu Corp., Shizuoka-Ken, Japan) after spatial filtering with an ∼0.8 Airy disk diameter circular pinhole to improve transverse resolution [94]. Non-confocal 800 nm multiple-scattered light was relayed onto a bundle of 4 multimode optical fibers with 400 µm cores (BFL44HS01, by Thorlabs) each of which delivered light to a H7422-50 PMT to capture non-confocal quadrant images (labelled North, South, East and West) [95]. The fiber bundle end was moved axially empirically to maximize quadrant split-detection image contrast [96].

5. Methods

5.1. Animal preparation

C57BL/6J mice (Jackson Laboratories, Bar Harbor, Maine) were anesthetized by xylazine and ketamine based on their body weight (0.01 mg xylazine/g + 0.08 mg ketamine/g). Eye pupils were dilated and cycloplegia induced with one drop of 1% tropicamide (Somerset Pharma LLC, NJ). A + 10 Diopter rigid PMMA clear contact lens with 1.6 mm base curvature and 3 mm diameter (Advanced Vision Technologies, Lakewood, Colorado) was used for partial compensation of refractive error and corneal hydration. Ophthalmic ointment was used for hydration of the non-imaging eye and for brushing the whiskers away from the cornea (GenTeal Tears Ointment (Alcon, Fort Worth, TX) or Neomycin and Polymyxin B Sulfates and Bacitracin Zinc Ophthalmic Ointment, USP (Bausch + Lomb, Bridgewater, NJ). All experimental procedures were performed in compliance with animal protocols approved by the Institutional Animal Care and Use Committee at Stanford University.

5.2. Data acquisition

Images from the confocal PMT, non-confocal quadrant PMTs and the MEMS scanner orientation signal were captured simultaneously using a 16-channel digitizer operating at 40 MHz sampling rate (ATS9416 by Alazar Tech. Inc., Pointe-Claire, QC, Canada). The scanner non-resonant axis was driven by a 50 Hz asymmetric sawtooth. Each dataset consisted of sequences of 150 images (384 rows, 409 columns). This sampling corresponds to an estimated four samples per Airy disk diameter for this wavelength, assuming a 1.8 mm pupil diameter.

5.3. Image processing

The retinal raw images were first corrected for distortion due to: periodic pixel sampling and the resonant scanner sinusoidal angular velocity variation [66,67]; imperfect synchronization between the resonant scanner and pixel sampling [66,67]; and optical distortion [97]. In addition, a full frame registration was applied to correct the misalignment due to eye or animal movement due to breathing, and averaged over 75 frames to improve the signal-to-noise ratio [71,98]. Dark field, horizontal split and vertical split images were generated by combining registered and averaged quadrant detection channels [95].

6. Experiments

6.1. MEMS resonant scanning axis

Resonant optical scanners oscillate with sinusoidal angular velocity variation. Often, their synchronization with image digitizers is only implemented with pixel precision, which introduces sub-pixel line jitter. We evaluated a previously demonstrated technique for measuring and correcting these problems [66,67], as well as two new alternatives for when the scanner orientation analog signal is too noisy or not available. In all three approaches, depicted in Fig. 8, the sinusoidally warped images with horizontal line jitter are captured with a digitizer simultaneously with the scanner analog orientation signal and/or the corresponding digital synchronization signal (typically generated by the scanner electronics by passing the analog orientation signal through a zero-crossing circuit).

Fig. 8.

Strategies for estimating desinusoid and line sampling jitter parameters using analog and/or digital signals from a MEMS resonant scanner. The exclusive use of the scanner analog orientation (strategy #1) results in sine period estimation with 5% error and poor jitter estimation. The use of the analog signal to estimate the sine amplitude, period and average phase and a synthetic sine wave triggered by the scanner digital trigger signal (strategy #2) results in the same period error but superior jitter correction. Finally, the use of the scanner low-passed filtered digital signal to estimate sine period and phase (strategy #3) allows estimating the sine period with 0.1% error and similar jitter correction to strategy #2.

The first strategy was an attempt at replicating previously used methodology [66,67] based on the use of scanner analog orientation signal. Despite this signal being “rectified” and distorted by the scanner electronics (see top plot of Fig. 8), we were able to select slightly less than half of the sine cycle (highlighted by the green background) to estimate the sine amplitude (5% error), period and phase by averaging across multiple image lines, to correct the sinusoidal image warping. The analog signal within this fraction of the sine cycle, however, is too noisy to allow sub-pixel jitter correction, as the panels on the right side of Fig. 8 show, where the corrected vertical line looks even more jagged after jitter correction.

A second hybrid strategy was pursued by using the analog signal to estimate the sine amplitude, period and average phase as before, and the scanner native digital signal to trigger an analog sinusoidal signal generated by a function generator. This synthetic analog signal is then fitted to estimate and correct for the line-to-line phase changes (jitter). The corresponding corrected vertical strip images show how the jitter correction is beneficial, albeit not perfect.

Finally, we created a synthetic analog scanner orientation signal by low pass filtering the digital trigger signal, which is then fitted like in the previous strategy to estimate the sine period, average phase and jitter, but not scan amplitude, obtaining a 0.1% period estimation error.

The axis of optical scanners are known to wobble or precess [99,100], and here we aimed to compare the impact of this wobble on confocal imaging when using the MEMS scanner and that of a traditional galvanometric scanner (SC-30, Electro-optical Products Corp., Escondido, California, USA). This was achieved by using a sub-Airy disk confocal detector to capture images through almost the entire sine cycle after aligning the detector to maximize the brightness of the first half of the image. In the absence of axis wobble, the portions of the image corresponding to each semi-cycle would be identical mirrored versions of each other. In practice, however, the wobble manifests as an asymmetry in the brightness between the left and right sides of the image, as shown in Fig. 9. This asymmetry is ∼30% in the traditional scanner and less than 1% in the MEMS scanner, indicating that the MEMS scanner precess less.

Fig. 9.

Confocal reflectance images of paper (different magnification, fields of view and field curvature) using almost the entire resonant scanning cycle to illustrate the scanner precession, which reduces the intensity captured during the second half of the cycle.

6.2. MEMS non-resonant axis

The rotation of the 2D scanner mirror around its non-resonant axis exhibits a “ringing” or oscillation that is typical of under-damped oscillators. This ringing is noticeable in AOSLO images after the rapid orientation change that takes place when the scanner returns the beam from the bottom of the field of view to its top, as seen in the top left panel of Fig. 10 with a 50 frames per second (fps) image acquisition rate. This is a known problem, described in the scanner manual, together with instructions on how to select the movement parameters in order to mitigate this problem. These instructions dictate that the sawtooth waveform rise time or fall time must be set to a multiple of the scanner resonant frequency’s period (∼570 Hz, 1.75 ms period), while the other must not be so (see top right panel of Fig. 10).

Fig. 10.

Images of Ronchi rulings with a 2-dimensional MEMS optical scanner affected by undesired oscillations (ringing) along the non-resonant axis. The deterministic ringing component, most visible on the top left panel, can be mitigated by appropriate selection of the rise and fall times of the sawtooth analog signal driving the scanner orientation (top right panel). Despite this minimization, residual ringing can be observed on the bottom panels, with both a deterministic component, which can be seen in the average of 500 images (left), and non-deterministic component shown by the non-zero values of the standard deviation of the same 500 images (right). Note that both the average and standard deviation images are displayed on the same gray scale value, illustrating that the non-deterministic ringing is non-negligible.

Notably, even after mitigating the non-resonant axis ringing and correcting sinusoidal image distortion and sampling jitter, we can see that when averaging 500 images (Fig. 10 bottom left panel) residual vertical distortion in images of a different Ronchi ruling remain. This indicates the presence of deterministic (i.e. repeatable) residual ringing, which should be correctable if properly measured. Unfortunately, when calculating the standard deviation of the same set of 500 images, we see a substantial variation of the Ronchi ruling lines, indicating the presence of a strong non-deterministic component. This non-deterministic component cannot be corrected in post-processing, even if recording the scanner orientation signal provided by the manufacturer, because it is extremely noisy, weak and coupled with the analog signal from the resonant scanning axis.

6.3. Reflections evaluation

Despite the use of lenses with anti-reflection coatings, long focal length and confocal (i.e., spatially filtered) detection, lens reflections remain substantial, as illustrated in Fig. 11. These reflections can be thought of as an additive background that could be subtracted from the raw images, provided the pixel values are not saturated and that corresponding images without an eye are captured. Unfortunately, this is not practical because as the sequence of images show, reflections in the AOSLO change with deformable mirror focus, and thus, one would have to capture background images for all deformable mirror foci and detector gain values.

Fig. 11.

AOSLO confocal reflectance images of a mouse eye, showing dramatic changes in reflections from lens #6 due to deformable mirror focus change.

6.4. Polarization control

Images sequences were collected in a C57BL6 mouse with a linear polarizer in the portion of the illumination path that does not overlap with the imaging path, and a second polarizer in the portion of the imaging path that does not overlap with the illumination with orthogonal polarization transmission. In addition, images were collected with a quarter waveplate between the AO ophthalmoscope and the eye. The averages of these image sequences can be seen in Fig. 12, with each row corresponding to images captured simultaneously. The top two rows correspond to the same raw images, registered using all the image pixels first (top row), and then, by excluding saturated pixels, which correspond to reflections (2^nd row). The latter produced sharper images, with the small nerve fiber layer bright dots being easier to resolve throughout the confocal image. The use of crossed polarizers results in a reflection-free but grainier image despite the averaging, due to the lower signal detected by the confocal PMT, perhaps, due to corneal birefringence. Regardless of the lower SNR, the elimination of the reflections improves the image registration, improving the visibility of the nerve fiber bundles and vessel walls in the dark-field, and non-confocal split-detection images. Finally, the use of crossed polarizers with a quarter waveplate shows the highest SNR and no obvious reflections, which translates into images revealing the finest structural detail in all modalities. This configuration, however, resulted in substantial corneal reflections that affected the wavefront sensor, which had to be mitigated by using annular wavefront sensing illumination. Unfortunately, both this approach and the use of a narrow wavefront sensing beacon require the lateral translation/adjustment of the wavefront beacon for every retinal location, as the beacon rays impinge on the cornea with varying angles of incidence. This makes the operation of refractive AO ophthalmoscopes more complex and has led to the development of wavefront sensorless AO ophthalmoscopes [35,39,72]. These instruments have demonstrated transverse and axial resolution comparable to those of AO ophthalmoscopes with wavefront sensors, but at the expense of lower temporal bandwidth and often also sacrificing the correction of wavefront aberrations with high spatial frequency.

Fig. 12.

Registered averages of reflectance and multiple-scattered AOSLO images of a mouse eye, captured using 800 nm light. Each row corresponds to a different lens reflection mitigation strategy, with the images on them captured simultaneously and, thus, showing the same retinal layer. The red borders indicate the imaging channel used for image registration.

6.5. Focusing through the mouse retina

With the long focal length lenses described earlier, cross-polarizers and a tilted quarter wave plate to mitigate the reflections in the system, we captured the images shown in Fig. 13 below. All retinal layers were imaged with a contact lens without power. This is important, as some studies have reported using contact lenses with power, implicitly suggesting that the power needs to be adjusted to optimally image a particular retinal layer of interest [40], while our images indicate otherwise.

Fig. 13.

Registered averages of reflectance and multiple-scattered AOSLO images of a mouse retina, captured using 800 nm light. Each row corresponds to simultaneously captured images of the nerve fiber layer (NFL), the inner plexiform layer (IPL), the outer nuclear layer (ONL), and the photoreceptor layer. The red borders indicate the imaging channel used for image registration. Short movies of blood flow in capillaries and vessels can be seen in Visualization 1 ^{(60.4MB, avi)} and Visualization 2 ^{(66.6MB, avi)} .

7. Summary

A refractive adaptive optics scanning light ophthalmoscope designed for small animal imaging through a 1.8 mm diameter pupil was described. The selection and placement of achromatic lenses, achieved through the evaluation of catalog lenses and the use of modified afocal relays, resulted in an optical setup with nominal diffraction limited imaging performance over multiple source vergences and large refractive error ranges in both pupil and retina conjugates.

Real ray tracing was used to compare wavefront aberration correction of large focus errors using a deformable mirror, a traditional Badal optometer, and a modified Badal optometer. These simulations show the previously reported asymmetry in the limits of the refractive error correction range over which diffraction limited imaging can be achieved [70,101,102]. The fact that the deformable correction is better for hyperopic eyes, while the Badal is better for myopic eyes, indicates how the AOSLOs should be operated.

Three widely used strategies for lens reflection mitigation were revisited, with imaging through crossed polarizers with a quarter waveplate showing the best cancellation of lens reflections in the imaging channels, but with strong corneal reflections in the wavefront sensing channel, degrading the AO wavefront correction. The use of crossed polarizers without the waveplate showed the effective mitigation of lens reflections in the imaging and wavefront sensing channels, but at the expense of a major signal-to-noise ratio reduction in the imaging channels.

Non-sequential ray tracing showed how the use of lenses with longer focal lengths can reduce lens reflections manyfold, but not as much as polarization control or lens-tilting. Finally, we used the ray tracing software to systematically find lens tilt angles that both vignette lens reflections while minimizing wavefront aberrations in both pupil and retina conjugates, improving on the common practice of tilting lenses empirically until reflections are vignetted but without considering the impact on image quality. As proposed, tilting lenses in a traditional Badal configuration substantially reduces the focus range over which diffraction limited performance can be achieved, but if the changes in retinal magnification and resolution from a modified Badal focus correction is acceptable, then this configuration almost completely restores the diffraction limited focus range achievable without lens tilt.

A new 2D optical scanner with 29.2 kHz resonant frequency and negligible dynamic optical distortion was evaluated and three strategies for estimating and correcting sinusoidal image warping and line sampling jitter were demonstrated. Unfortunately, when operated the non-resonant axis at 50 Hz, the scanner surface exhibits both deterministic and non-deterministic oscillations that severely limit the scanner applicability to high resolution imaging in general, but more specifically, to ophthalmic imaging.

Finally, non-confocal quadrant detection and confocal reflectance imaging of retinal structures in healthy mice illustrated the impact of the reflection mitigation approaches, which are applicable to all scanning and non-scanning ophthalmoscopes, as well as other reflectance-based imaging modalities such as optical coherence tomography. Other technical advances such as the modified refractive afocal pupil relays, the modified Badal optometer, together with some understanding of the wavefront RMS and vignetting compromises when using Badal or deformable mirror focus correction, are also applicable to other AO ophthalmoscopes.

Funding

Research to Prevent Blindness 10.13039/100001818 ( Departmental Award); National Eye Institute 10.13039/100000053 ( P30EY026877, R01EY031360, R01EY032147, R01EY032669); Advanced Research Projects Agency for Health 10.13039/100023015 ( ARPA-H).

Disclosures

The authors declare no relevant conflict of interest.

Data availability

Data underlying the results presented in this paper is not publicly available at this time but may be obtained from the authors upon reasonable request.