Single wavelength, dual temporal offset module for AOSLO systems

Abstract

We designed a generalizable module enabling dual temporal and spatial offset imaging. Two optical channels are generated from a single input. At a single focus, these can be used for blood flow calculations and template-free imaging. The system is supported by custom APD detector systems. Each of the optical channels is imaged onto five detectors through a 1-to-5 fiber bundle to acquire images from one confocal and four offset apertures. The module can be added to existing AOSLO systems at pupil conjugate planes. Examples of customizing the module for the Indiana AOSLO and APAEROS systems are presented.

1. Introduction

Over the past three decades, adaptive optics (AO) retinal imaging has made major strides in terms of both the accessibility to image broad populations and the ability to measure multiple cellular and functional details of the retina [1–6]. Both AOSLO and AO OCT can now measure multiple cell types in the living human retina, as well as record photoceptors responses and make quantitative measurements of blood flow and immune cell activity [7–12].

For the AOSLO, these advances have been achieved through a series of technical advancements. Manipulation of scan patterns has allowed precise quantification of blood velocity in larger arteries and veins as well as the distribution of velocity within the blood vessels [13,14]. Manipulations of the aperture configuration such that the AOSLO can capture multiply scattered light images as well as singly scattered confocal light images [4–6,9,15]. The capture of images based on multiply scattered light allows the AOSLO to operate much like a dark field microscope, increasing sensitivity to small index of refraction differences that are present between otherwise transparent structure and their surrounds. Thus, it has become possible to measure the inner segments of photoreceptors, ganglion cell bodies, erythrocytes, leukocytes, hyalocytes, microglia, and the mural cells of the retina [7,8,10,13,16–21]. This versatility has been demonstrated through a series of aperture manipulation ranging from programmable offset apertures [15], to multi-aperture imaging [6] to opposed (or split) apertures and recently to systems that use four or more offset apertures to generate contrast [22].

The AOSLO, because of the relative ease of adding channels and detectors, has also been able to use temporal offset imaging [14]. In this imaging mode, two separate imaging beams are used. These beams are superimposed at pupil conjugate planes, so that the two beams use the same scanning and wavefront control elements but are slightly separated in the image plane. As a result, a retinal feature can be imaged twice within the same video scan, and by adjusting the spatial separation between the beams, the temporal separation can be changed, meaning that the timing can be tuned [14]. This configuration has proven able to precisely measure blood cell velocity in vessels between 6 and 30 µm in diameter [23,24]. A complementary feature of this dual temporal offset mode of imaging is that two different retinal locations are measured simultaneously. Thus, as a result multiple pairs of points within a single scan frame have a known spatial relationship, and consequently the AOSLO can be used for template free imaging [25], that is, it is possible to produce a retinal image undistorted by eye movements, which otherwise requires a much higher frame rate.

The current paper presents a platform that brings together most of these AOSLO capabilities in a compact and affordable solution. We present this as an optical module, that is an addition to a system that can provide these features to both existing and new AOSLO designs. We first describe the optics, which provide two imaging channels at the same wavelength, eliminating issues arising from transverse and axial chromatic that would otherwise require setting the offset focus for each channel and varying the spatial separation at different retinal locations. In this module one of the channels is fixed in location and the other can be easily steered, providing an adjustable temporal offset between the two channels. Each of the optical channels is sampled with a 1-to-5 fiber optic bundle [22] to provide a confocal and four offset apertures. The collection of information from all ten channels requires multiple detectors, and therefore, we include the optical and electronic design of these components using avalanche photodiodes (APDs) to reduce costs (see Supplement 1 ^{(1.4MB, pdf)} ). Next, we show how this module can be adapted to different systems. We concentrate on the Indiana AOSLO, but also show schematically how it can be added to an APAEROS BMC system, a commercial system that is present in the lab of our collaborators at Northwestern University. Finally, we present sample results from the module from both the Indiana AOSLO and APAEROS optical systems.

2. Methods

2.1. Optical design assumptions

The goal of the optical design is to build a module that takes a collimated light input and generates two beams that are overlapped in the pupil conjugate plane but can be separated at the image plane using a single tilting plane mirror (placed in pupil conjugate plane). The module has unity magnification from the input pupil to the output pupil. In addition, the module takes light returning from the retina and efficiently focuses each scanning beam onto a fiber optic bundle which provides the confocal and offset apertures. The system design, except for the lenses which focus returning light to the fiber bundles, is reflective and thus will work for any reasonable wavelength supported by the mirror coatings chosen. The collimated input beam can range in size up to 12 mm to accommodate a range of designs of the rest of the AOSLO system. Larger pupil sizes may require a different fiber bundle due to the scaling of the Airy disc (AD) and an altered design for holding the beamsplitters to allow an unobstructed passage of light to and from the retina as is discussed below. The module is constructed from achromatic mirrors, except for the lenses that focus the beam onto the fiber bundle. That means it can be implemented in any AOSLO systems, but different lenses may be needed depending on the wavelength used in the optical system. The system is designed to fit upon an 18 × 24 × 0.5 inches breadboard for ease of introduction into an existing system.

2.2. Design implementation

2.2.1. Input optics

The input beam from the light source passes into the module (Fig. 1), and is divided into two beams by a 8:92 plate beamsplitter (BS1; reflectance:transmission, BSN11, Thorlabs). The first optical channel (OCH1) starts with the beam transmitted through BS1, and the second optical channel (OCH2) begins with the beam reflected from BS1. For OCH1, a pupil conjugate plane (PP1) is in the plane of a second 10:90 beamsplitter plate (BS2; reflectance:transmission, BSN11, Thorlabs). The reflected beam from BS2 is folded by a plane mirror (M1) to hit the curved mirror (SM2, R = 750 mm) at a horizontal angle of 4 degrees. The beam is focused by SM2 onto a spatial beamsplitter (D, PFD10-03-P01, Thorlabs) which is located at the focal point of SM2, which is a retinal conjugate plane (RP). OCH2 originates at the reflection from BS1 and is then reflected from a plane mirror (M3) which is located at a pupil conjugate plane (PP2). Next, the light is folded by a second plane mirror (M4) to hit the curved mirror (SM5, R = 750 mm) at a horizontal angle of 4 degrees. SM5 focuses the beam at RP, but with a slight angular displacement of PP1 relative to PP2, and this tilt causes it to bypass D. As a result, the two beams have pupil conjugate locations but the beams emerging from them have slightly different angles such that they have different retinal conjugate locations (imaging plane). Both channels are passed to a plane mirror (M6) and tilted horizontally (45°) and vertically (-8°) to hit a curved mirror (SM7, R = 750 mm). SM7 is tilted vertically 4 degrees, which serves to cancel the astigmatism generated by M1 and SM2. The vertical tilt on planar mirror M6, combined with adjusting the distance between M6 and SM7 allows control of the height difference between the plane of the module and the output beam, as discussed below. The module and the rest of AOSLO systems are connected at pupil conjugate plane (PP) with a plane mirror (M0).

Fig. 1.

A. A schematic of the module. S1 – light source with collimator (840 nm), λ/2 – half-wave plate, M, M1, M3, M4, M6, M8, M9, M0 – plane mirrors, BS1, BS2 – beamsplitters (10:90, R:T), SM2, SM5, SM7 – mirrors (R = 700 mm), L1, L2 –achromatic lenses (fl = 175 mm), D – spatial beamsplitter, fb1, fb2 – fiber bundles with their schematic arrange of the offset aperture locations, PP1, PP2, PP – pupil conjugate planes, RP – retinal conjugate plane. With purple color rays for the non-steerable optical channel 1 (OCH1), green color for steerable optical channel 2 (OCH2), and blue shows common path. A top view. B. A side view of the last part of the module’s schematic.

The spatial offset of the imaging beams can be altered at the pupil conjugate planes (PP1/BS2, PP2/M3) in both optical channels. However, the system was designed to easily change the angle of OCH2 by adjusting M3. Because M3 is present for both the imaging and return beams, altering it changes the angle at the retinal position of the beam, but not the angle of the return light relative to the detectors. That is, the beam can be readily steered using this mirror and we therefore refer to this optical channel as the steerable channel, and OCH1 as the non-steerable channel. If one wanted to move the retinal plane location of OCH1, it would require changing both the angle of BS2 and the location of the sampling fiber. The system allows simultaneous displacement of the beams in horizontal and vertical dimensions. Adjusting the image plane offset in a pupil conjugate plane assures lossless entry of the light into the eye.

2.2.2. Light returning from the retina

The light reflected from the retina is imaged back through the AOSLO and module elements, folded by plane mirror (M8 for OCH1 and M9 for OCH2) and imaged onto a multimode fiber bundle (6323-01-00-00.915-A, Berkshire Photonics) by achromat lenses L1 and L2 (fl = 175 mm, LA1229-B-N-BK7, Thorlabs, for OCH1 and OCH2, respectively). Each of the fiber bundles includes a confocal and four offset apertures (see Fig. 1). The fiber bundle used has a central fiber with a core diameter of 105 µm, and four surrounding fibers with core diameters of 550 µm. The fiber sizes was calculated to produce a confocal aperture 1.7 Airy disc diameter and 12.8 Airy disc diameter for the offset apertures using 840 nm (the imaging wavelength in the Indiana AOSLO). For the 780 nm used in the APAEROS system, the Airy disc diameters are proportionally larger. We intentionally used slightly larger fiber diameter to maintain light efficiency for small pupil sizes that are common in older individuals.

An important system-dependent aspect of the design is the magnification of the fiber bundle (fb1, fb2), relative to both the retina and the spatial beamsplitter that combines the beams. This magnification represents a balance between the pupil size at the module and the fiber bundle chosen. For instance, for a fixed fiber bundle, the relative magnification from the fiber bundle, through the system, and to the retina can be changed by varying lenses L1 and L2. These, in turn, will set the size of the bundle as imaged at RP. This size is important because it controls the spatial separation required at the retinal plane. In general, the two imaging beams should be separated by about 15 Airy Disc diameters, since there is little multiply scattered light beyond this [26]. Closer spacing will produce “ghost” images where a displaced aperture for one of the optical channels collects some multiply scattered light from the other optical channel. With the current system this can be minimized by using a lateral displacement (in the fast scan direction), allowing a wide range of time delays between the two optical channels. Examples of these tradeoffs are discussed below under implementation examples.

2.2.3. Detectors, high voltage and the cooling rack mount system

While the fiber bundle outputs can be directed to any point detector, most current AOSLO’s use either photomultiplier tubes (PMTs) or avalanche photodiodes (APDs, Fig. 2(A), assembly presented in Code 1 [27]: Electronics specifications of APD, cooling systems, and high voltage supplies, Code 2 [28]: APD schematics, Code 3 [29]: APD and detector lenses schematics). Because the dual temporal offset system uses 10 optical channels, the cost of using PMT’s can become prohibitive. APDs can provide similar performance to PMTs except at very low light levels, such as may be encountered in fluorescence imaging [30]. For this reason, we used a high sensitivity, low noise APD (C30659-900-R8AH, Excelitas Technologies Corp.). These have a bandwidth of 50 MHz and a noise equivalent power of ∼15 fW/sqrt(Hz).

Fig. 2.

A: Cutaway drawing of the coupling from the fiber output to the detector (all elements, except the detector are from Thorlabs). The details of APD detectors are presented in Code 1 [27] and Code 2 [28]. B: ZEMAX geometric analysis of the image of a 0.5 mm onto the 0.8 mm APD detector. C: The ideal coupling efficiency, ignoring reflections of different diameter fibers with a numerical aperture (NA) of 0.22 onto the APD detector. Good performance as evaluated using ZEMAX geometric image analysis is achievable for sizes up to ∼ 0.75 mm, but a different lens set would be needed for larger fibers.

The gain of the APD is controlled by applying a variable bias voltage to the APD. While the response of the APD is linear, the gain curve itself is sensitive to temperature and thus a given bias voltage will not produce the same gain from day-to-day. Because it can be desirable to use an AOSLO for quantitative measurements, we stabilized the temperature of the APD as discussed by Elsner et al. [31]. We also provided a second stage of gain close to the APD, and thus we mounted the APD, a secondary amplifier (Texas Instruments OP847) and Peltier coolers (CP40136, CUI Inc.) in an Al box (Code 1 [27]. This box was connected to a rack-mounted controller that contains two subsystems – high-voltage supplies and temperature controllers. Each rack mounted controller can control up to 4 detectors.

The high-voltage supplies are provided using DC to DC converters (0.5US12-P0, Ultravolt, Advanced Energy). All four are programmed using USB through a USB to I2C controller (FT232H, Adafruit Industries), which in turn controls a DA controller (MAX 5815, 4 channel, 12-bit), which sets the voltage. An Analog Devices 79943 is also connected via the USB port, and this is used to monitor the outputs of the DC-to-DC converter. The temperature controllers are based on a custom circuit combined with a Red Lion PID temperature controller. Detailed descriptions of these components are provided in the Code 1 [27].

2.2.4. Matching fiber output to APD detectors

While APDs are less expensive and more robust than PMTs, they have the disadvantage of having a small active area. Thus, it is necessary to carefully couple the fiber output to the detector (Fig. 2(A), also Code 3 [29]). Our fibers have an NA of 0.22 and a core diameter of 105 and 550 µm. The APD active area is 0.8 mm in diameter. Thus, in principle, imaging 1:1 is possible. However, because we wanted a small footprint, a relatively low cost per detector, and the ability to work with a range of fibers diameters, a diffraction limited design was not used. Instead, we slightly minify the tip of the fiber onto the APD active using a plano-convex lens (fl = 50 mm, LA1131-B, Thorlabs) to collimate the light from the fiber and an achromatic doublet (fl = 40 mm, AC254-040-B, Thorlabs) to focus the beam onto the detector. The lenses were mounted into a lens tube assembly, which is screwed into the face of the detector assembly. (Figure 2(A), and Code 3 [29]). The tube screw-in mount can focus the fiber tip onto the APD. The positioning of the fiber output is done with the coarse plate positioner (SPT1N, Thorlabs).The final design allows fiber diameters, assuming the same NA of up to 0.75 mm (Fig. 2(C)). A translation stage (MT3, Thorlabs) was used to finely position the fiber tip at the focus of lenses L1 and L2 since these positions are critical in obtaining a high quality confocal image and properly sampling the tails of the point spread function with the four offset fiber apertures.

2.3. Adapting the generic design to real systems

2.3.1. General considerations

In general, the module assumes that light is introduced into the AOSLO system as a collimated beam at a pupil conjugate location. This is a common approach in AOSLO systems. As described above, the module splits the input beam into two spatially separated beams and brings them together again at a pupil conjugate plane. This suffices for the main imaging beam, however, there are variations between systems that need to be accommodated. Four main factors need to be considered when matching this design to a real system – the system layout (height above the optical table), the beacon and wavefront sensor configuration, the system pupil size and the impact on fiber bundle choice, and the fast scan direction.

2.3.2. System layout

The height above the optical table differs between the two real optical systems we use as examples (the Indiana AOSLO and the APAEROS system). In both, the output beam is parallel to the table surface. We can precisely control the height above the table by a combination of varying the pedestal height of the final concave mirror (SM7) and varying the distance between the turn mirror (M6) and SM7 as long as we maintain the tilt of SM7 to cancel the astigmatism created at the other concave mirrors. For systems that do not introduce the beam parallel to the optical table, a mirror at the exit pupil can rotate the optical system to match (not shown).

In order to ensure both optical channels came to the same focus, we measured the wavefronts of each channel. A separate Shack-Hartman sensor (WFS30-7AR, Thorlabs) was placed at the pupil conjugate plane (M0). The wavefront aberration, especially defocus were minimized and equalized by adjusting the distance between SM2 (for OCH1) or SM5 (for OCH2) to SM7.

2.3.3. Beacon and wavefront sensor configuration

The wavefront sensor and associated beacon are not part of the module but it is important to plan for these. While in principle a complex detection scheme for the wavefront sensor beam could allow using the imaging as the beacon, it would require the displacement calculation to match the spatial separation of the two spots on the retina. Rather we use a separate light source at a different wavelength for a wavefront sensor beacon.

The location to insert this beacon is also important. If the beacon were introduced at the entrance pupil for the dual temporal offset module, the beacon would generate two spots, one for each of the spatially offset beams. This can be simply avoided by introducing the beacon either later in the system or by applying a bandpass filter in one of the arms of the module. If the wavefront sensor were placed such that it could sense the aberrations of the entire optical system except lens L1 and L2, the dynamic range of the wavefront sensor would be impacted. That is – the sensor would be sampling light that returned from the retina through two halves of the retinal plane, due to the presence of the spatial beamsplitter (D in Fig. 1). While this would be fine when aberrations were low, it could be difficult and require more complex processing if there were considerable aberrations. Such as when starting an imaging session with myopic or hyperopic defocus. If the wavefront sensor were placed in just one of the sub-channels (for instance by using a dichroic beamsplitter in place of mirror M1 (Fig. 1) and putting the sensor at the pupil conjugate), the field of view of each lenslet would be divided by the spatial beamsplitter, with the resulting limit to dynamic range for the sensor. While these issues could be handled in software, it was deemed, given the very high optical quality achievable with the optical design (Fig. 2), to have the module outside the common path AO control loop. If this is undesirable, then it is possible to execute a static correction for any non-common path errors that results [32] but to date this has not been required – as shown in the results.

2.3.4. Pupil size

The design pupil size interacts with the fiber bundle sizes chosen as well as the lenses which focus the light returning from the retina onto the fiber bundle face. For instance, if a 12 mm input pupil is chosen, and the lenses are 175 mm focal lengths, then a 100 µm diameter fiber chosen to be the confocal aperture would subtend approximately 3 Airy Disc diameters – which would be considered too large for most implementations. This can be altered by either choosing a different fiber bundle or changing the focal length of the lenses (or both). As shown above the proposed design of the coupling of the fibers to individual APDs can efficiently handle fiber sizes up to about 750 µm (Fig. 2(C)).

2.3.5. Fast scan direction

The fast scan direction is important in terms of setting the temporal offset between the two imaging beams. In general, for AOSLO measurements, the temporal offset will be on the order of 1 to 5 msec, and this will mean that the temporal offset should be varied by adjustments of PP2 in the slow scan direction to control the temporal offset. For very fast events, this may not be the chosen case. In either case, this can be addressed by controlling the orientation of the spatial beamsplitter (here a D-shape mirror) since use of an XY variable mount for PP2 allows adjustment in either dimension.

2.4. Implementation 1: Indiana University AOSLO

2.4.1. System layout

The Indiana AOSLO is a diffraction-limited system based on a woofer-tweeter design (Fig. 3) [14]. It was comprised of a back end using a traditional 4f configuration (mirrors m1-m13) and a front end using a modified 2f design which allowed steering over ∼30 deg without a change in fixation [22]. The original system used two imaging beams at different wavelengths to provide the temporal offset (780 nm and 840 nm). One of the imaging beams (840 nm) was also used as the beacon for wavefront sensing.

Fig. 3.

A schematic of the Indiana AOSLO with module. Components of the original system are indicated with small letters (sm1) and red rays, whereas components that were changed or added are indicated in caps (M1) and blue rays. S1, S2 – light source with collimator (840 nm, 780 nm, respectively), λ/2 – half-wave plate, M, M1, M3, M4, M6, M8, M9, M0, m0, m, m4, m9, m13, m17 – plane mirrors, BS1, BS2 – beamsplitters (10:90, R:T), d – dichroic beamsplitter, p – pellicle beamsplitter, SM2, SM5, SM7 – mirrors (R = 700 mm), sm1, sm12 – mirrors (R = 1000 mm), sm2 – mirror (R = 750 mm), dm3, dm11 – deformable mirrors, sm5, sm8 – mirrors (R = 500 mm), sm6, sm14 – mirrors (R = 400 mm), m7 – fast scanner, sm10 – mirror (R = 2000mm), m15 – horizontal scanner, m18 – vertical scanner, fm16, fm19 – field mirrors (R = 304.8 mm), L1, L2 –achromatic lenses (fl = 175 mm), D – spatial beamsplitter, WFS – wavefront sensor, PP – pupil conjugate plane, RP – retinal conjugate plane. A top view.

To implement the new module (see Code 4 [33]: mechanical 3D design of module for the Indiana AOSLO), we wanted to separate the beacon from the imaging. We employed an 840 ± 10 nm SLD (S1, 20 mW, cBLMD-series Compact Broadband Light Source Modules, Superlum, Ireland), which is combined with a collimator (LB10-820S-F2 Silicon Lightwave Technology, Inc., USA) and a half-wave plate which rotates the azimuthal angle of the polarized light coming from the light source.

The height of all the elements located on the breadboards, except SM7, was set to 3.5 inches above the breadboard and SM7’s height was 2.5 inches resulting in a height above the optical table of 3 inches The distance between M6 to SM7 was set to achieve the 1 inch change in height. Given the breadboard thickness (0.5 inch), the light exiting the module was 3 inches, matching the beam height of the existing input pupil (M0).

The light exiting the module has a pupil conjugate plane at one focal length from mirror SM7. We placed a flat mirror (M0) at this location to allow precise alignment to the existing optics without moving the pupil location. The light was then relayed in a 4f set of relays, with the pupil planes occupied by the high frequency deformable mirror (DM; dm3, 4.4 mm aperture, 140 actuators, ±4µm stroke, Multi-DM, Boston Micromachines Corporation, Cambridge, Massachusetts, USA), the fast resonant scanner (m7, SC-30, EOPC with custom mirror), and the second high stroke, low frequency DM (dm11, Mirao TM52-e, Imagine Eyes, Orsay, France). The front end of the Indiana system includes two large spherical mirrors (diameter 227 mm, fl = 152.4 mm) operating close to a 2f configuration to minimize system aberrations as the imaging beam is moved by a set of galvanometers (m15, m18) located at pupil conjugate locations.

2.4.2. Beacon and wavefront sensor configuration

The wavefront sensing channel (Fig. 3, green beam path) was added to the imaging channel at a dichroic mirror (d, Di02-R830-25-D, Semrock Inc., USA). The wavefront sensing channel combined a 780 nm beacon light source (S2), beamsplitter (p) and a Shack-Hartman wavefront sensor (WFS). Between the dichroic mirror and the WFS, we located a pellicle beamsplitter (p, 10:90, reflectance:trasmission, BSN11, Thorlabs) to introduce the beacon (10 mW, SLD-mCS, Superlum, Ireland). The WFS was based on a 12-bit sensor camera (Uniq Vision 1820, Camera Link) with lenslet array with approximately 450 samples within a nominal pupil of 6.4 mm.

2.4.2. Pupil size

The Indiana system has an entrance pupil diameter of 6.4 mm. The configuration of the focusing lenses and fiber optic channels are as described above.

2.4.3. Fast scan

Due to the rotation of the scan in the Indiana system, the fast scan direction in the module is vertical. Thus, for the spatial beamsplitter we used a spatial beamsplitter with the edge oriented horizontally. The oscillation frequency is 15.1 kHz.

2.5. Implementation 2: APAEROS BMC system

2.5.1. System layout

For the APAEROS system (Fig. 4, Code 5 [34]: mechanical 3D design of module for the APAEROS system) the primary difference for implementation is that 1) the wavelength for imaging is 780 nm and the wavefront sensing is at 840 nm; 2) the height above the optical table is 4.3 inches, and the beacon, imaging beam and an alignment laser are all introduced at a 10% reflective mirror in a common path.

Fig. 4.

A schematic of the relation of the module to the original BMC APAEROS system. Beamsplitters (BS) already present in the APAEROS are used to optically couple the new module. S1, S2 – light sources with collimator (780 nm and 840 nm, respectively), BS1, BS2 – beamsplitters (10:90, R:T), L, L1, L2 – lenses, WFS – wavefront sensor, M, M1, M3-M6 – plane mirrors, SM2, SM5, SM7 – concave mirrors, D – spatial beamsplitter, fb1, fb2 – fiber bundles, PP1, PP2 – pupil conjugate planes, RP – retinal conjugate plane, red square represent the optics of the APAEROS system. A top view.

We changed the height above the breadboard to 2.5 inches. We altered the distance between M6 and SM7, as well as the vertical tilt direction of M6 such that the exiting beam was at the same height above the table was 4.3 inches.

2.5.2. Beacon and wavefront sensor configuration

The APAEROS system uses separate imaging and beacon light sources. Thus, we used the original sources and redirected the imaging beam (blue line in Fig. 4), which was originally combined with the beacon beam (green line in Fig. 4) and reflected off a 10:90 beamsplitter (reflection:transmission). The imaging beam was directed to the input pupil of the module. Light returning from the retina is transmitted through the beamsplitter to a dichroic filter, which diverts the 780 nm imaging beam to a detector subsystem. We can utilize the same dichroic filter to introduce light into the system and to separate the return light into the existing wavefront sensor pathway and the module.

2.5.3. Pupil size

The APAEROS system has an entrance pupil diameter of 7 mm. The configuration of the focusing lenses and fiber optic channels are as described in the Matching Fiber output to APD detectors in the Design Implementation section.

2.5.4. Fast scan

There is no rotation of the beam in the APAEROS system, and as a result the spatial beamsplitter edge needs to be oriented vertically.

2.6. Subjects and image acquisition

We tested 5 eyes from 5 subjects with the age varying from 32-63. They were healthy subjects as well as the diabetic patients without diabetic retinopathy. For each of them, we recorded 100-frames videos of cones and retinal vessels in at least 5 locations.

The images were post-processed to eliminate image distortion from the slow scanner, and the eye movements [35]. Next, for each acquired frame, the split images were calculated following the formulas (1-3)/(1 + 3), (2-4)/(2 + 4), (1 + 2-3-4)/(1 + 2 + 3 + 4), and (-1 + 2 + 3-4)/(1 + 2 + 3 + 4), where 1 – image from top left offset detector, 2 – image from top right offset detector, 3 – image from bottom right offset detector, 4 – image from bottom left offset detector. For the purpose of this article, the split images were merged by using the maximum values from split images at each pixel.

3. Results

The Airy disc radius (AD) of this system for an emmetropic eye is 2.1 µm (calculated for 840 nm). When used with the deformable mirrors flat, and a myopic model eye that matches the myopic shift we design into the system, the point spread function (PSF) over more than 6-degree region is diffraction limited. As projected onto the retina, the offset fibers each sample ∼10AD in diameter across. Given the above, the entire fiber bundles subtends ∼40 µm at the retina (Fig. 5(A)).

Fig. 5.

A: A schematic of fiber ends (blue) as imaged onto the retina. B: PSF (Zemax) of the relay system, circle is AD. C: Retinal PSF imaged back through the AOSLO onto the detector planes. Scales are in micrometers.

The array of offset apertures provide nearly complete coverage of retinal scattering since they sample out to 11 AD. The minimum distance of the spatially offset between the two optical channels with all apertures unobscured is ∼80 µm laterally. Retinal PSF projected on the detector planes for a myopic eye showed that samples cover the entire fiber bundle’s extent (Fig. 5(C)).

The tilt of the pupil conjugate plane mirror (M3, Fig. 1) generates beam displacement in the retinal conjugate plane, however, there is no displacement at the pupil. On the path back from the eye, the beam traverses M3 again, nullifying the change in angle. As a result changing the displacement between OCH1 and OCH2 does not require realigning the fiber end for each retinal separation. This was confirmed experimentally by imaging a target in a model eye and continuously changing the angle of M3. The example of two different OCH2 offset in relation to OCH1 were presented in Fig. 6.

Fig. 6.

The confocal image was maintained over all displacements which did not redirect the steerable beam in a way that caused it to miss the spatial beamsplitter. Example of single frame images with different temporal offsets, collected within a single imaging session. Here, we showed an image for OCH1, which is fixed with images from OCH2 while setting different angles on mirror M3. Images are from the Indiana AOSLO. A: Reference image from OCH1, B: image from OCH2 displaced vertically by 36 µm in relation to OCH1, C: image displaced 65 µm vertically in relation to OCH1. The redline shows the location of a descending capillary (small round black feature in images).

Imaging performance was excellent with good quality images available from all 10 detectors (Fig. 7(A)). These data could then be combined to provide multiple comparisons based on the cellular constant (Fig. 7(B)). Data have been collected from both the Indiana AOSLO (Fig. 7) and the APAEROS (Fig. 8) systems. The images shown were acquired at ∼1 µm/pixel and the temporal displacement between the non-steerable channel (OCH1, top row images) and the steerable channel (OCH2, middle row) was 124 pixels horizontal (the fast scan direction) and 54 pixels vertically. With a fast scan frequency of 15.1 kHz, the vertical displacement shown represents an approximately 3.58 ms “delay” between when a given feature was imaged with OCH1 and OCH2.

Fig. 7.

Example of averaged AOSLO images from the Indiana AOSLO acquired with the dual temporal offset module. Pictograms show the offset aperture or confocal apertures that were utilized for each subimage. A: Images formed from individual apertures. First row – images from superior left aperture, second row – images from superior right aperture, third row – images from inferior right aperture, fourth row – images from inferior right aperture, and fifth row – images from confocal aperture. The first column is for OCH1, the second for OCH2. B: computed images based on combining data from A. Rows 1-4 – split images, row 5 – combined images computed by merging information from all four offset detectors in each channel using a simple maximum for the split images at each location. Columns as in A are for each of the two optical channels. The contrast was adjusted for display.

Fig. 8.

Images acquired with the APAEROS system from two offset channels (A, B) and the confocal channel (C). Also shown is the calculated contrast image based on the difference between the two offset images (D). Arrows: diagonal (black), horizontal (red), and vertical (green) arrows point to the changing contrast of the vessel walls and RBC’s based on the offset direction and feature orientation (A and D) and the use of the differential contrast to better combine information that is orientation independent (C). The contrast was adjusted for display.

The contrast of the retinal feature is highly dependent on the orientation of the offset aperture (Fig. 7(A) rows 1-4, and Fig. 8(A), (B)). To expose these differences more clearly, we decided to place only two offset channels and the calculated split image from the APAEROS system (Fig. 8). What is clearly visible in one channel is not in the other ones (arrows in Fig. 8 point out examples of these details). Combining images from four offset aperture detectors enhances the visibility of retinal details, such as retinal vasculature, more than the computed split images (Fig. 7 merged, Fig. 8(D)).

4. Discussion and conclusions

The new module successfully allows current AOSLO systems to be upgraded to a full implementation of the dual temporal offset approach, with each channel providing quad offset capabilities. Although the presented module inserts a number of elements that are not in the common AO control loop, we have observed no decrement in performance. This is in part because we designed the system to work with all optical elements having a relatively low NA and compensated the tilts in the horizontal direction with identical tilts in the vertical directions. This configuration compensates off-axis astigmatism and achieves diffraction limited performance. While, in principle, the module could be decreased in size by using more compact mounts and shorter focal lengths, care would need to be taken to maintain the image quality. This is especially true for the lenses which focus light onto the fiber bundles, since light displacements will induce astigmatism in the focus onto the fiber bundle and these displacements become more critical with shorter lenses.

An advantage of the system is that detection apertures can be easily changed by swapping fiber bundles, enabling multi-aperture configurations with more than four offset fibers or higher resolution using fiber bundles with multiple sub-Airy Disc apertures [36]. Since the current system is intended for measuring vascular structure as well as blood flow, both temporal offset channels have full quad detection capabilities. For applications where only rapid, template-free imaging is required, or a need to measure fast changing events, the number of channels can be simply adjusted by changing the number of detectors and the fiber used for each offset channel. Another advantage of this module is that one can easily change the spatial separation between both optical channels simply by changing the angle of the M3 with the mount knobs as shown in Fig. 6. One can produce a wide range of offsets from 0 to more than 150 lines of temporal difference between channels but there is an important tradeoff. In the fast scan direction, the offset controls how many “samples” one has to perform the correction for eye movements. Our value of about 1/10 of the frame means that there are matching locations between the two imaging channels only in the overlap region between the two channels. A large offset will reduce the retinal range for template free imaging. This design is potentially generalizable to allow different focus planes for the two channels, although this has not been implemented since it requires moving more than one optical element to maintain all pupils conjugate.

There are a few limitations to the proposed system. First, the system is designed for two channels with the same wavelength. This eliminates the need for chromatic aberration compensation. The use of different wavelengths as we implemented in the past [14,23,37], causes two imaging locations to vary in spatial offset at different retinal location due to the transverse chromatic aberration of the eye. The use of two channels with the same wavelength also has the potential disadvantage of generating a fixed relationship between the relative intensities of the two beams, since OCH1 passes through only one beamsplitter, providing 10% of the light to the channel, and OCH2 passes through two, providing 9% of the light. If this is a major limitation for some implementations then the introduction of a half-wave plate between BS1 and BS2 would allow tuning the relative reflectance, however it will also change the amount of light detected for the confocal channel, since that light is highly polarization preserved [38]. As mentioned when discussing the adaptiation of this module to different systems, the use of a different imaging wavelength may require changing the fiber bundle and/or the lenses that focus the light onto the fiber bundle input in order to match the change in numerical aperture that occurs with changes in wavelength (or pupil diameter). Second, while the angular offset of the two beams can be easily modified without the need to change the position of the fiber bundle the minimum temporal offset between channels is fixed to about 100 pixels in the fast scan direction and 0 pixels in the slow scan direction. This results because of the need to keep retinal scattered light from one channel from leaking into the offset apertures of the second channel. This limitation means that the highest frequency of sampling would correspond to about 350 kHz based on the 100 pixel separation and a 35 MHz pixel clock. For smaller separations, a different aperture configuration could be used, for instance the use of only two confocal apertures. There is also a slight cross-talk between channels caused by reflections at the beamsplitters. For OCH2, approximately 90% of the light returning from the retina passes through BS towards the detectors. But approximately 10% of the light is also reflected back towards BS1. At BS1, most of this light is transmitted, but about 10% of that reflected light from OCH2 is reflected by BS1 and is directed to the fiber optics bundle of OCH1. That is, about 1% of the imaging light returning from OCH2 is actually detected by OCH1. By the optical specifications this ghost image has roughly 0.008 times the intensity of the OCH1 image. We verified that it could be detected by blocking OCH1 and increasing the OCH1 detector gain, confirming it is less than 1%. More detailed measurements were not conducted, for this “ghost” image, since in order to detect it at our maximum imaging beam required a very high gain and the images were very noisy.

Funding

National Institutes of Health 10.13039/100000002 ( 1OT2OD038128-01); National Eye Institute 10.13039/100000053 ( R01 EY024315).

Disclosures

The authors declare no conflicts of interest.

Data availability

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

Supplemental document

See Supplement 1 ^{(1.4MB, pdf)} for supporting content.