Wide-vergence, multi-spectral adaptive optics scanning laser ophthalmoscope with diffraction-limited illumination and collection

Visualizing and assessing the function of microscopic retinal structures in the human eye is a challenging task that has been greatly facilitated by ophthalmic adaptive optics (AO). Yet, as AO imaging systems advance in functionality by employing multiple spectral channels and larger vergence ranges, achieving optimal resolution and signal-to-noise ratios (SNR) becomes difficult and is often compromised. While current-generation AO retinal imaging systems have demonstrated excellent, near diffraction-limited imaging performance over wide vergence and spectral ranges, a full theoretical and experimental analysis of an AOSLO that includes both the light delivery and collection optics has not been done, and neither has the effects of extending wavefront correction from one wavelength to imaging performance in different spectral channels. Here, we report a methodology and system design for simultaneously achieving diffraction-limited performance in both the illumination and collection paths for a wide-vergence, multi-spectral AO scanning laser ophthalmoscope (SLO) over a 1.2 diopter vergence range while correcting the wavefront in a separate wavelength. To validate the design, an AOSLO was constructed to have three imaging channels spanning different wavelength ranges (543 ± 11 nm, 680 ± 11 nm, and 840 ± 6 nm, respectively) and one near-infrared wavefront sensing channel (940 ± 5 nm). The AOSLO optics and their alignment were determined via simulations in optical and optomechanical design software and then experimentally verified by measuring the AOSLO’s illumination and collection point spread functions (PSF) for each channel using a phase retrieval technique. The collection efficiency was then measured for each channel as a function of confocal pinhole size when imaging a model eye achieving near-theoretical performance. Imaging results from healthy human adult volunteers demonstrate the system’s ability to resolve the foveal cone mosaic in all three imaging channels despite a wide spectral separation between the wavefront sensing and imaging channels. OCIS codes (110.1080) Active or adaptive optics; (170.4460) Ophthalmic optics and devices; (170.4470) Ophthalmology


Introduction
The adaptive optics scanning laser ophthalmoscope (AOSLO) is an important imaging tool that can achieve in vivo, near diffraction-limited visualizations of microscopic structures in the retina by compensating for the monochromatic aberrations of the eye [1].
Increasingly, adaptive optics systems are employing multiple wavelength channels for a range of imaging and vision testing applications. Most systems use different wavelengths for wavefront sensing and imaging. Systems for AOSLO microperimetry and visual psychophysics employ NIR wavelengths for wavefront sensing and tracking and deliver AOcorrected flashes of visible light. Other multi-wavelength applications include fluorescence [2][3][4], hyperspectral imaging [5], retinal oximetry [6] and multi-modal imaging [7]. There are two important factors to consider when designing, building, and interpreting results from a high-fidelity multi-wavelength AOSLO system: the chromatic aberrations of the eye and the chromatic aberrations of the system.

Chromatic effects of the eye:
First and foremost, the chromatic dispersion of the eye needs to be considered. The chromatic difference in defocus of the eye (longitudinal chromatic aberration, or LCA) has been extensively studied and its behavior is quite predictable and similar between individuals [8], albeit with inter-individual differences depending on the eye's specific optical parameters (e.g. corneal curvature). The transverse chromatic aberration, or TCA, of the eye has also been studied [9,10] and recently objective techniques to measure it [11,12] and correct it [13,14] have been employed. The extent to which the high order aberrations of the eye change with wavelength is less studied. As a rule, there must be differences in high-order aberrations as the ray paths that different wavelengths of light take through the optical system to reach a focus on the retina are different. But, researchers generally agree that these differences are small and often fall within the range of measurement error of the system [15][16].
In the current study, our aim was to carefully explore the implications of using NIR wavelengths for wavefront sensing and visible light for imaging. In a diffraction-limited system, the size of the point spread function is proportional to wavelength, so the best possible outcome is that the changes in high-order aberrations between wavelengths are negligible, and the benefits of reduced diffraction would yield increasingly sharp images at progressively shorter wavelengths even if wavefront sensing was performed at long (NIR) wavelengths. In the worst possible outcome, the high order aberrations would change enough to offset the benefits of reduced diffraction when imaging at wavelengths shorter than the wavefront sensing wavelength.

Chromatic effects of the AOSLO system:
To facilitate an exploration of the implications of high-order changes in chromatic aberrations in the eye for multi-wavelength AOSLO systems, it is paramount to minimize or fully characterize the chromatic effects of the AOSLO itself. The bulk of this manuscript describes this effort. Current-generation AOSLOs are typically able to achieve high resolution over wide vergence ranges (~3 diopters). Optimal performance has been achieved by designing the relay telescopes in the AOSLO with either a non-planar design using off-axis spherical mirrors [5,17], a planar design using a combination of off-axis toroidal and spherical mirrors [18], or an on-axis, lens-based design using polarized light and polarization gating to minimize back-reflections from refractive surfaces [19]. However, to our knowledge, a similar consideration of vergence effects in the collection path of an AOSLO has not yet been described, despite the use of off-axis optical elements such as plate/wedge beamsplitters and dichroic mirrors in the collection path, which induce vergence-dependent aberrations. Here, we present a multi-spectral AOSLO design with diffraction-limited performance in both the illumination and collection paths over a 1.2 diopter vergence range. The optical design utilized all commercially available off-the-shelf optics thereby improving the cost-efficiency of our system compared to prior AOSLOs using custom-sized spherical mirrors [5] or custom toroidal mirrors [18]. To facilitate the optical alignment and construction of this AOSLO design, we developed a detailed optomechanical model and constructed a laser-cut polycarbonate stencil, which indicated the placement of all optomechanics onto an optical table. After the AOSLO was constructed and aligned, we experimentally validated the system's resolution and collection efficiency using a phase retrieval technique and a model eye setup.
Finally, we present imaging results at the fovea in two healthy adult volunteers and quantitatively compare resolution across imaging channels using a Fourier analysis. For wavefront sensing at 940 nm, the benefits of reduced diffraction extends to images taken with shorter wavelengths. In the two subjects that were imaged using optimal confocal pinholes [20], the foveal cone mosaic was well resolved at all wavelengths (840, 680 and 543 nm). The benefits of reduced diffraction with shorter wavelengths was readily visible but images at 680 nm and 543 nm were similar in quality indicating that the effects of higher order aberrations of the eye may begin to play a small role for shorter wavelengths.

System overview and optical design
The schematic and optical design of the multi-spectral AOSLO system is shown in Figure 1 with a specification of components in Table 1. Light from a supercontinuum light source was separated into three imaging channels spanning different wavelength ranges (543 ± 11 nm, 680 ± 11 nm, and 840 ± 6 nm, respectively) and one near-infrared wavefront sensing channel (940 ± 5 nm). All spectral channels were aligned to be collinear using pairs of mirrors for each color channel and 3 dichroic mirrors. A 10:90 (R:T) wedge plate beam splitter was placed after the dichroic mirrors to separate the illumination and collection paths and was followed by four sets of spherical mirror-based telescopes arranged in a non-planar off-axis manner that relayed an image of the system's entrance pupil onto the resonant scanner, galvanometer scanner, deformable mirror and, finally, the pupil of the eye. The positions and angles of all optics after the beam splitter in the illumination path (see Table 2) were optimized to minimize aberrations, mainly astigmatism, using optical design software (Zemax, LLC, Kirkland, WA) and the technique described by Dubra et al. [5]. Diffractionlimited illumination spots were achieved for a 1° field of view (FOV) across all AOSLO spectral channels, as indicated by the spot diagrams and Strehl ratios at the bottom of Figure  1.
To minimize aberrations in the collection path, we optimized the optics following the wedge plate beamsplitter in transmission (including the wedge plate beamsplitter itself) and fixed all optical components that had an effect on the illumination path. From our initial analysis, we determined that the wedge plate beamsplitter had the largest contribution to the collection path aberrations. To minimize aberrations induced by this beamsplitter, we optimized for the wedge angle and incident beam size (within optomechanical constraints and beamsplitter availability) to produce a minimum spot size across all spectral channels while allowing a sufficient wedge angle to reject the ghost artifact from the back surface of the beamsplitter (see analysis in . Based on the simulation results, we chose to use a 0.5° wedge plate beamsplitter (BSX10, Thorlabs Inc, Newton, NJ) positioned along the optical path such that the incident illumination/collection beam size for all spectral channels was minimized given optomechanical constraints.  After minimizing the aberrations caused by the wedge plate beamsplitter, the resolution of the AOSLO's collection path was shown to be diffraction-limited (Strehl ratios > 0.8) for all imaging channels and for the WFS channel (see spot diagrams of collection path in Figure 1).

Alignment procedure for the AOLSO light delivery and collection
In order to take full advantage of AO correction, the optical path for each imaging channel must be coincident with that of the wavefront sensing spectral channel. Spectral channels were coaligned by making image and pupil planes coincident across all channels. To allow for the alignment of an image plane without misaligning the corresponding pupil plane, we placed the initial collimating lens for each spectral channel on an XY translation stage with a separate iris placed in the front focal plane. With this design, the image plane for any spectral channel could be fine-tuned by shifting the collimator lateral position without affecting pupil position.
To bring all spectral channels into focus at the same plane as the retina, the AOSLO requires optical pre-compensation of the eye's longitudinal chromatic aberration (LCA). We implemented compensation by adjusting the input vergences of each spectral channel. In this AOSLO design, we chose the 680 nm spectral channel as the reference channel with 0 D vergence making this path collimated prior to the wedge plate beam splitter and at the eye. With 680 nm as a reference, the LCA of the eye was calculated to have a vergence difference of -0.607 D at 543 nm, 0 D at 680 nm, +0.392 D at 840 nm, and +0.5516 D at 940 nm [8,10]. This vergence shift was implemented for each spectral channel by shifting the axial position of the channel's input fiber such that the image plane for a model eye with a focal length of 100 mm would be 106.46 mm at 543 nm, 100 mm at 680 nm, 96.22 mm at 840 nm, and 94.77 mm at 940 nm. To aid in alignment, a camera was placed at the calculated image plane of the model eye for each spectral channel and the fiber was precisely positioned to minimize the spot size seen on the camera.

Wavefront sensorless adaptive optics to correct static system aberrations
To improve the imaging performance of the AOSLO system and make it robust to minor system aberrations, we utilized wavefront sensorless adaptive optics to determine the optimal wavefront shape that maximizes detected light. Our implementation of sensorless adaptive optics is similar to that described by Hofer et al. [21] and applies random perturbations to the deformable mirror shape to optimize for the AOSLO image's mean intensity rather than the wavefront sensor readings. By using a small confocal pinhole (with a sub-Airy disk diameter), aberrated and out-of-focus light is rejected by the confocal pinhole and the wavefront shape that optimizes for mean intensity also corresponds to the wavefront with minimal aberrations. Once the deformable mirror shape is optimized based on the detected image's mean intensity, the residual wavefront reading on the WFS is recorded and is assigned as the target shape for subsequent WFS-based AO correction.

Image processing and human subject protocol
The system components and hardware customized for this system are listed in Table 1. The custom image acquisition software and AO software are same as that of our previous AOSLO system and have been previously described in [22,23]. The image registration techniques used for stabilizing eye motion was a strip-based cross correlation method previously described in [24,25].
The University of California Berkeley Institutional Review Board approved this research, and subjects signed an informed consent form before participation. All experimental procedures adhered to the tenets of the Declaration of Helsinki. Mydriasis and cycloplegia were achieved with 1% tropicamide and 2.5% phenylephrine ophthalmic solutions before each experimental session. Subjects bit into a dental impression mount affixed to an XYZ translation stage to align and stabilize the eye and head position. Both subjects (20112L and 20076R) were healthy young adult volunteers. Structural imaging was performed on both subjects using the 543 nm, 680 nm, and 840 nm imaging channels and the 940 nm WFS channel for AO correction.

Results
After constructing and aligning the multi-spectral AOSLO system, the optical resolution for each spectral channel was quantified using a phase retrieval technique and a through-focus stack of intensity images [26,27]. The phase retrieval algorithm utilized through-focus intensity images to iteratively solve for the complex PSF using the Gerchberg-saxton algorithm [28]. A Fourier transform of the complex PSF was then used to reconstruct the wavefront and determine optical quality metrics like Strehl Ratio, which we used to evaluate resolution for each spectral channel (see Table 3). The optical design and system alignment were further validated after measuring the system's collection efficiency per spectral channel as a function of confocal pinhole size (see Figure 7). In this experiment, the sample was a model eye with paper as the retina and the illumination profile at the pupil plane was a top hat with a circular aperture. In the absence of scattering effects, the spatial profile of the light incident on the confocal pinhole is expected to follow that of the double-pass point spread function's encircled energy [29], which follows the black curve on Figure 7. However, scattering within paper is expected to further broaden the light distribution at the confocal pinhole with its own point spread function, which has been shown to approximately follow a Lorentzian function [30] (see dashed lines in Figure 7). To measure the paper's scattering point spread function, a separate simplified setup with the same illumination profile and model eye from the AOSLO was used to measure the collection efficiency per imaging channel vs confocal pinhole size (see dense dotted lines in Figure 7). Comparing the encircled energy for the multi-spectral AOSLO with the measured encircled energy of the simplified setup, we see that our results match quite closely indicating the multi-spectral AOSLO's near-theoretical collection efficiency across all imaging channels. As reference, we also plot related work from Sredar et al. [31] and an encircled energy curve resulting from the convolution between the theoretical double-pass point spread function (without scattering) and a 2 µm FWHM Lorentzian for each imaging channel. This latter plot matches closely to the encircled energy measurements of our simplified setup indicating that the paper of our model eye has a PSF that approximately equals to a 2 µm FWHM Lorentzian function.  The practical implications are that one can achieve foveal cone resolution in both 543 nm and 680 nm spectral channel when the correction is informed by the measurements of the eye's ocular aberrations at 940 nm. This study shows the practical limits of the different spectral channels and that both 680 nm and 543 nm imaging can be used for structural foveal imaging, psychophysical experiments in the fovea [11] and measuring and correcting for TCA [13,14].

Conclusions
We have demonstrated a multi-spectral AOSLO design with diffraction-limited illumination and collection to achieve high-resolution, high-throughput retinal imaging. After constructing and validating the AOSLO performance, images were acquired at the foveal center from two healthy subjects to demonstrate the system's capability to visualize foveal photoreceptors in all imaging channels with wavefront correction based on a separate spectral channel. The use of this methodology and system design may provide increased collection efficiencies in other SLO or AOSLO designs that employ large vergence ranges over multiple spectral channels.

Funding
National Institutes of Health (National Eye Institute) (T32EY007043, R01EY023591, U01EY025501, P30-EY003176), Alcon Research Investigator Award, Minnie Flaura Turner Memorial Fund for Impaired Vision Research, Soroptimist International Founders Region Fellowship Disclosures AR: Patents: USPTO #7,118,216, "Method and apparatus for using AO in a scanning laser ophthalmoscope" and USPTO #6,890,076, "Method and apparatus for using AO in a scanning laser ophthalmoscope". These patents are assigned to both the University of Rochester and the University of Houston and are currently licensed to Boston Micromachines Corporation in Cambridge, Massachusetts. Both AR and the company may benefit financially from the publication of this research. Financial Interest: C.Light Technologies. Both AR and the company may benefit financially based on the publication of this research.