1. Introduction

Twenty-five years ago, the advent of adaptive optics (AO) ophthalmoscopy [1] marked a paradigm shift in our ability to visualize the retina non-invasively through the natural optics of the living eye. In a microscope, the higher the numerical aperture of the objective, the higher the resolution. In the eye, if we follow the same line of reasoning, the larger the pupil, the better should be the resolution of the retinal image. However, the presence of optical aberrations introduced by passage of light through the anterior segment optics of the eye causes the eye’s point spread function to expand with larger pupils and thus the resolution of the retinal image to decrease. AO technology enables measurement and correction of these optical aberrations in a closed loop, thereby condensing the eye’s point spread function and increasing the resolution of the retinal image. AO-equipped ophthalmoscopes can achieve a lateral resolution close to the theoretical diffraction limit even while imaging through the natural optics of the eye. Considering that fully dilated pupils reach ∼5-8 mm in diameter, the resolution of AO ophthalmoscopes thus can reach ∼2 µm [2]. Indeed, with the invention of AO ophthalmoscopy, investigators were suddenly able to visualize cellular sized structures within the living human retina, such as photoreceptors, nerve fibers and capillaries (Fig. 1), non-invasively. Arising from this capability, a new field of scientific study emerged; one centered on exploiting the advantages afforded by AO to investigate basic and clinical questions of the visual system both in health and disease.

 

Fig. 1. Adaptive optics ophthalmoscopy. (A) Schematic of aberration measurement and correction provided through an AO loop to enable aberration-corrected imaging of the living eye. (B) Individual AO images of the photoreceptor mosaic in a normal-sighted eye illustrating the advantage provided by AO (B1: AO off, B2: AO on). Figure courtesy of Stephen Burns.

Download Full Size | PDF

In the twenty-five years that AO technology has been applied to the eye, most (if not all) ophthalmic imaging techniques known to date have been combined with AO, at least in a research setting. Further, AO ophthalmoscopy has enabled visualization of most cell types in the retina. Despite this, AO ophthalmoscopy remains largely a research tool with widespread adoption of the technology into clinical practice remaining a future prospect. The following review examines clinically-related applications of AO ophthalmoscopy to date, with a specific focus on the ways in which the technology has enabled structural and functional assessments of the human visual system in health and disease.

The exponential increase over the past twenty-five years in research publications, investigative teams, and number of AO-assisted imaging devices, makes it impossible to cite every publication using AO in this limited review. We have chosen to focus on applications of AO ophthalmoscopy with clinical relevance, and thus primarily discuss results that have been attained in living humans. We limit our discussion of basic science investigations to those that enable clinical translation in the present or foreseeable future.

2. Historical overview of adaptive optics ophthalmic imaging systems

The first demonstration of AO imaging of the human retina was achieved using a flood illumination ophthalmoscope equipped with a Shack Hartmann sensor for wavefront analysis and a deformable mirror for wavefront correction in 1997 [1]. This adaptive optics flood illumination ophthalmoscope (AOFIO) revealed individual cone photoreceptors across the retina and opened up a wide range of exciting applications in ophthalmology and vision science leading to the first quantifications of cone density in retinal disease [3,4]. Though the methods of performing wavefront analysis and correction can vary, the most commonly encountered AO configuration in ophthalmoscopy today remains the use of a Shack Hartmann wavefront sensor and a deformable mirror [5,6]. By 2002, the desire to improve the optical sectioning and image contrast attained with AOFIO had motivated the development of the first confocal AO scanning laser ophthalmoscope (AOSLO) [7]. The advantages of confocal AOSLO image sequences were immediately apparent in their superior image quality and the new applications enabled by confocal AOSLO, for example the visualization of retinal blood flow and the retinal nerve fiber layer (RNFL), fueled a direction change in the AO research community towards scanning based devices. While all current AOSLO systems remain based on a similar design to the Roorda et al. 2002 original [7], the most commonly reproduced AOSLO design to date was developed by Dubra et al. [8].

AOSLO has seen many exciting developments over the past several years due to the versatility of combining confocal detection with non-confocal detection schemes. The first multi-modal application for AOSLO combined confocal and short-wavelength autofluorescence imaging to elucidate the human retinal pigment epithelium (RPE) mosaic [9]. Essential to this scheme was a dichroic beamsplitter placed just before the confocal detector in order to simultaneously direct fluorescently emitted light to a second detector. Following this application, numerous other multi-modal AOSLO detection schemes arose. In 2012 Chui et al. [10] demonstrated that multiply scattered light could be detected with AOSLO by offsetting the confocal pinhole to capture off-axis photons and that this technique improved visualization of retinal vasculature. Each of these two detection schemes inspired additional non-confocal and fluorescence AOSLO imaging modalities including: non-confocal or dark-field reflectance [11], non-confocal split-detection [12], multi-offset pinhole [13], non-confocal quadrant-detection [14], fluorescein angiography [15], indocyanine green (ICG) fluorescence [16], and near-infrared autofluorescence [17]. The plethora of imaging schemes in use today arises from the variety of cells made visible under the different detection configurations and the extensive clinical applications enabled by the technology.

From a historical perspective, the AO imaging field developed during the same decade as another novel retinal imaging method, optical coherence tomography (OCT) [18]. OCT is an interferometric method that provides a histology-like cross sectional image of the retina with micrometer scale optical sectioning of the retinal layers. The complementarity between the high axial resolution of OCT and the high lateral resolution of AO ophthalmoscopes led to the development of a combined AOOCT system [19], which became the first in vivo retinal imaging device capable of three-dimensional cellular resolution. AOOCT, while having drawbacks such as slow scanning times and high system complexity, has recently enabled applications for three-dimensional cellular imaging through the full retinal depth, including enabling visualization of individual stacked, transparent inner retinal neurons, such as the ganglion cells [20].

While the term ‘adaptive optics’ is sometimes used by ophthalmologists in clinical practice to refer to cellular resolution retinal imaging devices in general, AO really refers to the components of an optical setup that form an adaptive optics loop, i.e. the components responsible for wavefront measurement and correction in a closed loop. Thus, AO is a tool for correcting the eye’s optical aberrations that can be added to the other technical components of ophthalmic imaging systems. We highlighted the combination of AO with flood illumination, SLO, and OCT to give a historical overview of some of the major technical developments made by the AO ophthalmic imaging field, but also because clinicians routinely use fundus photography, SLO, and OCT imaging in their everyday clinical practices and AOFIO, AOSLO, and AOOCT are the AO imaging corollaries to these clinical devices. Like clinical devices, there are in actuality a number of technically unique AO imaging schemes involving different configurations of retinal illumination and detection. For example, a variety of scanning configurations exist, ranging from non-scanning, full-field illumination as is used in AOFIO or AO-full-field-OCT [21], to line scanning devices such as an AO-line-scan-OCT [19] or AO-line-scan-ophthalmoscope (AOLSO) [22], to point scanning systems including AOSLO. As well, both spectral domain [23] and swept source [24] implementations of AOOCT have been demonstrated, as has AOOCT angiography [25,26]. As with clinical imaging technology, the different implementations vary in the level of adoption of the technique. We stress however, that AO in theory can be combined with any retinal imaging technique. Further, we are unable to identify any presently available clinical imaging schemes to which AO has not been added.

Following the success of early AO device development, the first commercial AO system became available, based on AOFIO technology (rtx1, Imagine Eyes, France) [27]. The advantages and disadvantages of this device remain those inherent to AOFIO; the system enables cellular resolution of cone photoreceptors outside of the central 1° of the fovea in both health and disease over a 4° field with the speed and robustness demanded of a commercial device, but the optical sectioning and image contrast remains inferior to that of scanning based systems. Various commercial prototypes of AOSLO have been presented over the years, from Physical Sciences Inc. [28,29], Optos [30], Canon [31,32], to Imagine Eyes [33], with some of these combined into multi-modal devices. These devices benefit from the improved optical sectioning and image contrast afforded by scanning systems, along with the usability factors of a commercial device, but with the disadvantage of increased cost and complexity.

Irrespective of technical specifications for different devices, however, Imagine Eyes’ rtx1 AOFIO currently has attained a unique status in the AO field as the only AO imaging device to date that has gained regulatory approval in some countries. In 2012, the rtx1 attained conformité européenne (CE) approval in Europe followed by the equivalent in Japan, China, and Korea. As a result, clinicians in these regions do not need to seek research approval from an institutional review board to use the rtx1 device on their patients. The rtx1 device has not been approved by the United States (USA) Food and Drug Administration (FDA), however, and it therefore remains at present a research-only device in the USA.

3. Applications of adaptive optics for imaging retinal structure

Historical and technical details aside, the value of adding AO technology to ophthalmic imaging devices is seen most readily in scientific applications. Ophthalmoscopy has long been a standard part of the ophthalmic exam as it enables visualization of retinal, neural, and even systemic disease. Full understanding of disease pathogenesis, progress and treatment, however necessitates investigations into the structural and functional pathology at the cellular level as well as a detailed understanding of how pathology differs from the normal visual system. AO ophthalmoscopy provides this capability, and as a result of the past quarter-century’s technical developments, the normative structure of most retinal cell types has been visualized in vivo using at least one of the numerous AO ophthalmoscopy detection schemes developed to date. In addition, AO imaging has been used in studies of most major blinding diseases including AMD, diabetic retinopathy, glaucoma, and numerous inherited retinal degenerations, as well as in studies of systemic disease such as hypertension. We emphasize that in comparison to imaging control volunteers, imaging patients with disease is more difficult due to the frequent presence of additional challenges such as optical opacities, large eye movements, dry eyes, small pupils, etc. The progression from imaging in laboratory settings to imaging naïve controls and then patients in the clinic is an important process for advancing clinical applications of AO ophthalmoscopy. Below, we highlight in more detail AO imaging findings from the major retinal cell types both in healthy and diseased eyes.

3.1 Photoreceptors

Vision begins when photoreceptors absorb light incident on the retina. Photopigments, contained within the cone and rod photoreceptor outer segments, isomerize when light is absorbed, thereby initiating phototransduction and starting a process by which electrical signals are passed from the cones and rod to the inner retinal neurons and ultimately to the brain. Photoreceptors thus provide a critical first step in vision, with loss of cones and/or rods resulting in visual impairment.

The cone photoreceptor mosaic was the first cell class elucidated with AO imaging [1]. Cones have waveguiding properties [34] which provide advantages for maximizing photoisomerization by guiding incoming light through the inner segment to the outer segment. These waveguiding properties work in reverse as well and the small backreflections from the cone inner-segment/outer-segment (IS/OS) junction and cone outer segment tip [35] are waveguided back through the photoreceptor inner segment and exit through the pupil of the eye [36]. The resulting Gaussian shaped waveguided reflectance profile of individual healthy cones when combined with the high resolution capabilities of AO technology thus gave natural contrast between neighboring cells, resulting in the cone photoreceptor mosaic being visualized in AOFIO and AOSLO imaging modalities with relative ease (Fig. 2) [1,7]. Rod photoreceptors, more difficult to image because their small size approaches the diffraction limit, have also been visualized (Fig. 2) [37].

 

Fig. 2. The normal photoreceptor mosaic imaged with AO. (A) AOFIO image of the normal cone mosaic at 1° superior to fixation. (B) AOSLO images of the photoreceptor mosaic in the parafovea using confocal imaging at 1° and 10° temporal to fixation using confocal and non-confocal split-detection. Individual cone and rod photoreceptors can be identified in the images (blue dots: cones, orange dots: rods). Confocal and non-confocal split-detection images show outer segment waveguiding and inner segments correspond one-to-one. (C) AOOCT enables visualization of the parafoveal cone mosaic in three dimensions [92]. Aligned B-scans (left) can be segmented to visualize the inner segment/outer segment (IS/OS) junction and cone outer segment tips (COST) en face (C-scans, middle). B-scans at the level of the photoreceptors (right) show each cone contains a reflection at the IS/OS junction and COST. Panel (C) courtesy of Ravi Jonnal.

Download Full Size | PDF

Visualization of reflectance from waveguiding cones in the living eye made possible quantifications of photoreceptor morphology (Fig. 2) which had previously only been possible with histology [38]. Individual cell identifications and segmentations have led to a number of metrics describing the cone mosaic phenotype including cone density, spacing, size, number and distance of neighbors, orientation, regularity and reflectivity [39–43]. As with histology, normal cone densities were found to decrease exponentially with eccentricity from the fovea [44–50], show meridional effects, with cone densities along the horizontal meridian exceeding densities measured along the vertical meridian at equivalent eccentricities [45,46,48,50], and exhibit a three-fold range in inter-subject variability for peak cone density at the fovea [40,51]. Cone density has also been shown to vary with axial length and thus refractive correction, with myopes generally exhibiting lower linear cone densities (cones/mm²) but higher angular densities (cones/deg²) [44,51,52]. In contrast to histology, in vivo imaging enables longitudinal studies of the same cells in the same participants; AO ophthalmoscopy has shown cone densities to remain stable over short periods of time [53], though cross-sectional studies have shown a decrease in cone density with age [45,46,54]. It remains to be determined if, when, and at what rate cone density decreases in aging.

In addition to the waveguided reflectance of photoreceptors, non-confocal split-detection imaging with AOSLO also has elucidated the photoreceptor mosaic. Multiply scattered light from the photoreceptor inner segment combined with directional detection, yields images where the photoreceptor mosaic appears as a bubble wrap-like structure where each photoreceptor consists of a dark- and bright-half (Fig. 2) [12]. In normal healthy retina, the confocal and non-confocal split-detection images of the cone mosaic have one-to-one correspondence with each cone inner segment in the non-confocal split-detection image having a corresponding waveguided reflectance in the confocal image. While both techniques have enabled quantification of the mosaic parameters, non-confocal split-detection images of the cone mosaic have proven superior for repeatable cone identifications [55–57].

Numerous blinding conditions manifest as a result of structural and functional degeneration of photoreceptors. Understanding photoreceptor degeneration has therefore been a primary focus of clinical applications utilizing AO ophthalmoscopy because of the ability to visualize the photoreceptors, the availability of quantitative metrics to describe the mosaic phenotype, and the involvement of the photoreceptor mosaic in retinal disease. Indeed, photoreceptor structure has been characterized in each of the major blinding diseases (AMD [58], diabetic retinopathy [59–61], and glaucoma [62,63]) as well as in a plethora of the 340 inherited retinal diseases [64] currently identified [65], including (but not limited to) retinitis pigmentosa [66–71], cone-rod dystrophy [3,72], Stargardt’s macular dystrophy [73–75], choroideremia [76–78], Usher’s syndrome [79,80], and achromatopsia [81–83].

Structural imaging both in cross-sectional and longitudinal studies has revealed decreased cone density (or alternatively increased cone spacing) generally as a consequence of disease. In this regard, there are commonalities in quantified photoreceptor metrics among retinal diseases. However, the mechanisms of photoreceptor degeneration can vary, resulting in disease specific mosaic phenotypes. Waveguided reflectance, thought to arise from an intact outer segment [35], may be mottled, abnormal, or absent in disease, to varying degrees (Fig. 3). For example, there is a large body of literature in achromatopsia [81–89] to show that the rod photoreceptors maintain normal waveguiding properties while remnant cone photoreceptors, present in severely reduced numbers, do not waveguide and instead appear as dark holes in a confocal AOSLO image (Fig. 3(A), Visualization 1). Co-located non-confocal split-detection AOSLO images however, shows these ‘holes’ actually contain remnant cone inner segments [12]. In addition to achromatopsia, this finding has also been shown in fundus albipunctatus and oligocone trichromacy [90,91].

 

Fig. 3. Confocal and non-confocal split-detection AOSLO images (1 and 2) of the foveal region in inherited retinal diseases: (A) achromatopsia, (B) choroideremia, and (C) GUCA1A-mediated cone-rod dystrophy. Yellow asterisks mark the fovea. (3 and 4) Confocal and non-confocal split-detection images, respectively of the region within the white square in (1 and 2). For achromatopsia (A), cone density is reduced, and the cones visible in the non-confocal split detection image do not exhibit waveguided reflectance in the confocal image (blue arrows). Red arrows point to rods which maintain waveguided reflectance. For choroideremia (B), the cone mosaic is relatively intact within the central region out to the atrophic border (yellow arrow). Orange asterisk: an outer retinal tubulation. Blue and red arrows point to clumps of hypo- and hyper-reflective cones, respectively. Yellow arrow marks the sharp border of atrophy. For GUCA1A-mediated cone-rod dystrophy (C), cone density is reduced, only a subset of the cones observed in non-confocal split-detection exhibit waveguided reflectance on confocal. Red arrows point to cone locations that exhibit waveguided reflectance, blue arrows point to cones with abnormal, reduced waveguiding.

Download Full Size | PDF

As a second example, images in patients with choroideremia have shown areas of both normal and reduced cone densities both in confocal and non-confocal split-detection imaging [76–78], with most cones exhibiting waveguiding properties, though some are dim and mottled with patches of both hypo- and hyper-reflectivity [77]. Further, AO imaging reveals distinct borders between the relatively intact central retina and the atrophic mid-periphery in choroideremia (Fig. 3(B), Visualization 2), verifying the RPE plays a critical role in the degenerative pathway for this disease [77]. Abnormally reflective, waveguiding cones have also been visualized in retinitis pigmentosa and Stargardt’s macular dystrophy [66,73, 74 ].

Finally, as a third example, GUCA1A-mediated cone-rod dystrophy exhibits yet another AO phenotype (Fig. 3(C), Visualization 3) with a subset of the parafoveal cones showing waveguiding properties and a subset of cones showing reduced waveguiding properties similar to cones in achromatopsia [72,93]. Non-confocal split-detection imaging in this case revealed a complete mosaic. Longitudinal imaging in cases such as these may provide insight into whether the non-waveguiding cones are more susceptible to disease progression than their waveguiding counterparts.

The inherited retinal degenerations highlighted here were chosen for their different photoreceptor phenotypes visible on AO imaging. The photoreceptors in numerous other retinal diseases have been characterized with AO imaging, and indeed to date there are 340 distinct inherited retinal degenerations that can be identified/distinguished by their genetic causes [65]. It remains to be determined how many different multi-modal AO phenotypes will be associated with these different genotypes and their associated conditions. Multi-modal findings however, provide precise and clinically relevant information on the health or degenerative state of photoreceptors. For example, as shown in Fig. 3, inner segments viewed in non-confocal split-detection mode can often remain visible after degeneration of the outer segment, which renders the cone invisible or abnormally waveguiding in confocal mode. These cones with intact inner segments are targets for novel vision restoration techniques such as gene and cell therapies. As such, understanding photoreceptor phenotypes within the natural history and progression of retinal disease is sure to remain a priority for present and future clinical studies.

Inherited retinal degenerations as a whole represent a major cause of blindness, though individually they are considered orphan diseases affecting only a small proportion of the population. Age-related macular degeneration, on the other hand, affects more than 10 million Americans [94] and 67 million Europeans [95]. Though the disease causing mechanisms of AMD are not localized to the photoreceptors, the vision loss associated with AMD becomes apparent when photoreceptors become compromised in the disease process. As a result, there are numerous studies which have investigated the photoreceptor phenotype in AMD using AO ophthalmoscopy. Cross-sectional studies which have attempted cone spacing and density measurements in AMD have shown mixed results with some studies yielding cone spacing/density measurements at normal levels [54,96] and others showing a reduction in cone density [97,98] that correlates with disease severity [99,100]. Several studies in AMD have shown photoreceptor density and reflectivity is altered at retinal locations overlaying drusen [97,100], subretinal drusenoid deposits [58,101–104] (Fig. 4(A)) and outer retinal tubulations [105].

 

Fig. 4. (A) The cone mosaic surrounding a subretinal drusenoid deposit in a 73-year-old male with non-neovascular AMD [101]. Using confocal AOSLO, waveguiding cones are visualized around the subretinal drusenoid deposit, the edge of which appears as a dark ring. Panel (A) courtesy of Yuhua Zhang. (B) Drusen, observed as hyper-reflective rings, in a 65-year-old patient with intermediate AMD revealed by gaze dependent AOFIO imaging [109].

Download Full Size | PDF

The appearance of drusen below the retina remains a classic phenotype of AMD [106,107]. Therefore, the detection of drusen remains an important strategy for identification of patients with early AMD. Up to now, color fundus photography remains the clinical standard for drusen detection, with OCT providing additional valuable information about drusen volume. Use of AO modalities should allow resolution of smaller drusen than with these conventional clinical modalities. However, drusen are generally revealed with low contrast in en face modalities including AOFIO and AOSLO [97,98,100,108]. An image acquisition and processing method was recently developed to enhance the contrast of drusen from AOFIO images acquired at different gaze positions (Fig. 4(B)) [109]. Using this method, more drusen were detected including those of smaller dimensions than those captured with the clinical gold standard technique, giving this approach potential to detect the development of small drusen in early-stage AMD. In addition to photoreceptor and drusen imaging in AMD, AO ophthalmoscopy has shown detailed borders of geographic atrophy with progression monitored via timelapse imaging, and hyperreflective pigment clumps [104,108,110,111].

3.2 Retinal pigment epithelium

The RPE serves numerous functions in the retina, including, but not limited to, acting as the blood-retinal barrier, providing metabolic support to the photoreceptors, and taking an active role in the rejuvenation of photopigment through the visual cycle [112]. As a consequence, disease processes that interrupt the normal function of the RPE often result in photoreceptor degeneration and subsequent irreversible vision loss. Thus, non-invasive visualization of the RPE in health and disease has been a long-standing interest for clinicians and scientists.

Imaging of the RPE with AO ophthalmoscopy however proved more difficult than imaging the photoreceptors, in part because the RPE lies more posterior to the photoreceptors, and the waveguiding properties and absorbance of the overlaying photoreceptors block reflectance originating from the RPE. Thus, to image the RPE, investigators have employed alternative imaging strategies to provide contrast and thereby detect the RPE layer.

In contrast to the photoreceptors, the first published report of the human RPE using AO ophthalmoscopy came from cases of diseased retina [113]. Patients with cone-rod dystrophy were imaged with an AOSLO, and in areas of the retina where the photoreceptor mosaic was not present, investigators were able to detect a honeycomb pattern of cells in the confocal reflectance image. Support for this reflectance pattern being representative of the RPE came from quantitative measures of cell spacing and comparison with clinical imaging modalities including fundus photography and microperimetry. While exciting, this reflectance imaging strategy was limited in utility by the requirement that the photoreceptors must no longer be present in an intact mosaic in order to observe the RPE (Fig. 5).

 

Fig. 5. Multi-modal AOSLO imaging in retinitis pigmentosa reveals the transition zone between the intact photoreceptor mosaic and loss of the photoreceptors. Blue arrows show the cone mosaic in both confocal and non-confocal split-detection AOSLO images. White and yellow arrows point to locations where the cone mosaic is no longer intact, and the RPE becomes visible both in confocal and dark-field imaging modalities.

Download Full Size | PDF

The first images of the human RPE mosaic underlying an intact photoreceptor mosaic came by incorporating short wave autofluorescence imaging techniques into the AOSLO system [9] (Fig. 6(A)). As a byproduct of the visual cycle and phagocytosis, both normal functions of the RPE, RPE cells normally accumulate lipofuscin, a substance which has intrinsic fluorescence properties with a peak in fluorescence excitation and emission within the visible wavelength spectrum [114]. Autofluorescence using an infrared excitation wavelength was also shown to elucidate the RPE mosaic by detecting the intrinsic fluorescence from melanin granules [17,115] (Fig. 6(B)). In both short wave and infrared autofluorescence AO images, the RPE cells appeared as donut-type structures, presumably because the RPE cell nucleus does not contain either lipofuscin or melanin and therefore it does not fluoresce. This method for imaging the RPE provided the advantage that the RPE could be visualized non-invasively beneath an intact photoreceptor mosaic. The main disadvantage however, was that the fluorescent signal was very weak in comparison to signal from reflectance imaging, and therefore the autofluorescence image required long detection times to acquire a high number of frames, along with multi-modal registration techniques to produce an average image with sufficient signal [9]. Regardless, these methods have proven useful for investigating the RPE mosaic structure in normal retina [9,17,116,117], as well as in patients with AMD [115,118,119], Stargardt’s [120], and other diseases [121] (Fig. 6(I)). Fluorescence lifetime imaging has also been applied to investigate the human RPE mosaic [122].

 

Fig. 6. RPE imaging in health and disease. Healthy RPE cells can be visualized using (A) short-wavelength autofluorescence AOSLO [9], (B) near-infrared autofluorescence AOSLO [17], (C) dark-field imaging AOSLO [17], (D) AOOCT [127], (E) ICG fluorescence AOSLO [16], and (F) transscleral AOFIO imaging [128]. In all modalities, RPE cells appear as a in a honeycomb pattern of tightly packed cells. (G) Enlarged RPE cells with hypofluorescent nuclei are visualized with ICG fluorescence in Bietti Crystalline Dystrophy [129]. (H) Transscleral AOFIO near the fovea reveals pigmentary clumps in the RPE layer of a 78-year-old female patient with geographic atrophy with foveal sparing. (I) Foveal mosaic of enlarged RPE cells visualized using near infra-red autofluorescence in a case of radiation retinopathy that caused loss of photoreceptors at the same location [115]. Panels (B), (C), (D), (E), and (G) courtesy of Johnny Tam. Panel (H) courtesy of Kiyoko Gocho.

Download Full Size | PDF

Alternate methods of providing contrast for visualization of the RPE mosaic have also been employed including non-confocal dark-field imaging [11] (Fig. 6(C)), and AOOCT [123] (Fig. 6(D)). Non-confocal dark-field imaging reveals the RPE structure through multiply-scattered light [11,16,124]. The technique however, may be best for revealing the foveal RPE especially when subject pigment is lighter, as in Caucasian subjects [11]. AOOCT imaging of the RPE mosaic is made possible by organelle motility over time changing the noise pattern observed from the RPE layer [123]. Subcellular registration techniques in three dimensions combined with averaging of volumes over time was critical for success and recent AOOCT studies have reduced the averaging required to observe the RPE [125–127].

The RPE has also been imaged with fluorescence techniques following injection of indocyanine green (ICG) dye [16]. Historically, ICG has been used clinically to visualize the choroidal vasculature as the infrared excitation wavelengths employed allow deeper penetration into the retinal tissue with less absorbance by ocular pigments [130]. When AO imaging was combined with ICG fluorescence techniques, investigators found that ICG could stain some of the RPE cells [16]. This partial and variable staining is beneficial in that it provides contrast between neighboring cells such that the mosaic is revealed. However, cells that do not take up the ICG lack inherent contrast, so it can be difficult to identify all cells in the mosaic with this technique. In addition, as is the case with other fluorescence imaging techniques, the signal originating from ICG fluorescence requires substantial averaging. Despite these challenges, ICG imaging has been used to investigate the RPE mosaic in healthy controls [16,127] and in individuals affected by disease (Fig. 6) [127,129,131,132].

Finally, borrowing from phase imaging concepts in microscopy, Laforest et al. [133] devised a method to add an additional illumination source to an AOFIO device in order to illuminate the retina transsclerally. In recent years, two distinct transscleral imaging configurations have emerged [128,133] in which the eye is illuminated from one or both sides of the sclera and the image is generated through a combination of two or several sequentially or simultaneously acquired images from different illumination angles. Clinical application of this technique is showing promise for revealing the RPE mosaic in healthy eyes and pigment distribution in diseased eyes (Fig. 6).

A persistent feature of all of the AO RPE imaging techniques to date, and perhaps the reason the techniques for RPE imaging are so numerous, is that they all have drawbacks or are technically challenging. Broader dissemination of RPE imaging technology will be needed to determine the extent to which AO imaging can be used to assess RPE morphology both in health and disease.

3.3 Retinal ganglion cells

Inner retinal neurons, which are responsible for relaying visual signals from the photoreceptors to the brain, lie on top of, or more anterior to, the photoreceptors. A consequence of this retinal anatomy is that the inner retinal neurons are necessarily transparent so that light can pass through them to arrive at the photoreceptor layer. The retinal ganglion cells (RGCs), located in the inner retina, are the last retinal cells that transfer the visual signal out of the eye through the optic nerve to the brain. RGCs are affected in diseases such as glaucoma where increased intraocular pressure leads to their death. In up and coming optogenetic therapies for retinal degenerations, the RGCs are being rendered light sensitive via viral delivery of opsins in order to restore light sensitivity to retinal cells once photoreceptors and intervening inner retinal neurons have expired [134]. Imaging of individual RGCs in vivo therefore may prove crucial for clinical applications aiming to understand diseases directly affecting the inner retina, such as glaucoma, or to investigate the safety and efficacy of experimental optogenetic therapeutics.

Imaging RGC somas has proven particularly challenging due to their transparency and their multi-layered arrangement outside of the fovea. Thus, applications which seek to image the RGCs (and other transparent inner retinal cells) require a different mechanism to provide contrast between cells than the standard reflectance detection used for photoreceptors. Two main approaches have demonstrated RGC imaging in the living human retina in recent years.

First, Rossi et al. [13] used a non-confocal AOSLO approach to increase contrast on RGC features. Rather than a split-detection style configuration, they implemented a multi-offset approach, averaging differences of many image pairs located at varying positions from 8 to 20 Airy Disk diameters from the confocal aperture, acquired sequentially. In this way they were able to show a contiguous RGC mosaic with phase contrast. A recent paper [135] describes in vivo human RGC imaging with an improved multi-offset design. However, the technique remains an AOSLO based method and it is thus limited in its optical sectioning, meaning that the 3D organization of RGCs in retinal areas where they are arranged in multiple layers cannot be fully resolved in depth.

Second, an AOOCT was used to enhance contrast on transparent retinal structures by taking advantage of the fact that the positions of subcellular organelles in cell cytoplasm fluctuate over time [136]. At any given moment compared to the next, the organelles are located at very slightly differing positions. These submicrometric fluctuations can be picked up as tiny phase changes in the AOOCT signal in otherwise static images. If a sufficient number of images are acquired and averaged over a period of time, these minute phase differences will begin to build up a signal that enhances contrast on cell cytoplasm. This gives rise to an image of the ganglion cell bodies (Fig. 7). The definition and contrast of the cells is high and homogeneous across the field. The 3D micrometric resolution of AOOCT means that the full 3D organization of the RGC mosaic is revealed. Early clinical results are very promising, revealing new basic knowledge about RGC death in glaucoma [137].

 

Fig. 7. Retinal ganglion cells resolved with AOOCT in a 54-year-old healthy control subject (left) compared to a 51-year-old glaucoma subject (right) at 12° temporal retina. Note the lower density and enlarged cells in the glaucomatous eye [137]. Figure courtesy of Zhuolin Liu.

Download Full Size | PDF

3.4 Retinal nerve fiber layer and lamina cribrosa of the optic nerve

In addition to the cellular mosaics described above, AO ophthalmoscopy can detect microscopic changes in the RNFL and lamina cribrosa in the optic nerve head earlier than clinical imaging devices, potentially leading to new insights into mechanisms of optic nerve disease such as glaucoma. In vivo imaging of these structures in healthy and glaucomatous human eyes using AO ophthalmoscopy has been reported by several groups [138–152].

The RNFL is composed of nonmyelinated ganglion cell axons that form the optic nerve and it is considered the most clinically accessible site for detecting glaucomatous axon damage due to elevated intraocular pressure. Visualization of individual nerve fiber bundles has been achieved using AO ophthalmoscopy [140,150,153–155]. AO imaging also has revealed subtle glaucomatous changes in nerve fiber bundles and their reflectivity that are difficult to detect with standard clinical OCT and fundus photography [146,148,151,152]. Of note, this technique has identified reduced RNFL reflectance intensity as a feature that precedes RNFL thinning and visual field defects in glaucoma [148,156]. Figure 8 shows an example of peripapillary retinal nerve fiber bundle defect progression over time in a patient with open angle glaucoma imaged using confocal AOSLO [156].

 

Fig. 8. Progression of peripapillary retinal nerve fiber bundle defect over 17.3 months in a 46-year-old patient with glaucomatous damage at the inferior region imaged using confocal AOSLO [156]. (A) Montage of the peripapillary AOSLO images obtained at 17.3 months superimposed upon the fundus photograph of the right eye. The regions within the green and red rectangles obtained at the baseline and 17.3 months are magnified in B1 and C1 and B2 and C2, respectively. (B1 and B2) Superior region shows relatively healthy retinal nerve fiber bundles and stable surface reflectively over time. (C1 and C2) Inferior region shows progression of a focal retinal nerve fiber bundle defect over 17.3 months. Defect widths are shown in the inferior region at both visits.

Download Full Size | PDF

The lamina cribrosa of the optic nerve is composed of a mesh-like network of collagenous beams, through which nerve fibers exit the eye posteriorly. It is hypothesized that mechanical deformation of the microarchitecture of lamina cribrosa produced by intraocular pressure elevation contribute to early optic nerve damage, making it a structure of interest for glaucoma detection and monitoring progression. AOOCT has allowed for direct visualization of lamina cribrosa at different axial depths, revealing the microscopic details of its intricate network of collagenous beams [142,144]. AOSLO has also demonstrated the morphological changes of enlarged pore area and increased pore axis ratio associated with elevated intraocular pressure in glaucomatous eyes [141], thus providing opportunities to improve our understanding of glaucoma and other optic neuropathies.

3.5 Retinal and choroidal vasculature

The high metabolic demands of the retina render it particularly vulnerable to vascular changes that impair its oxygen supply, nutrient delivery, and metabolic waste removal process. In vivo microscopic visualization of the retinal vasculature, directly and non-invasively, offers a unique opportunity to study its physiology and pathological changes in aging, diabetes, and hypertension. Retinal capillaries with diameter smaller than 15 µm are generally not visible in conventional color fundus pictures due to their poor optical contrast. While the use of intravenous injection of sodium fluorescein as an exogenous contrast agent can improve visualization of smaller blood vessels, it is an invasive procedure which limits the frequency of clinical use. AO ophthalmoscopy reveals the microvasculature in living human retina non-invasively (Fig. 9; Visualization 4), making it an appealing tool for detection and monitoring subclinical changes at the capillary level [10,25,153,157–165]. Its non-invasive nature is especially helpful for longitudinal tracking of retinal vasculopathies in humans [166–168], providing new insights into the dynamics of disease progression and assessment of treatment response over time. AO imaging studies of foveal avascular zone (FAZ) geometry [157–159,169,170], vascular perfusion density [157,169,171,172], capillary tortuosity [173], wall-to-lumen ratio [174–179], and microaneurysm morphology [180–185] of the human retinal microvasculature, have provided a framework for development of qualitative and quantitative biomarkers. These biomarkers offer the opportunity for earlier identification of subclinical microvascular changes, enabling more sensitive detection of pre-clinical disease progression and new strategies for treatment monitoring.

 

Fig. 9. Imaging of the normal foveal capillary network and FAZ. (A) Confocal AOSLO imaging shows the microvascular structure at the fovea. (B) Confocal fluorescein angiography perfusion map obtained using oral fluorescein. (C) Motion contrast perfusion map generated based on the blood cell flow induced reflectivity variation from registered non-confocal offset pinhole videos [159]. (A-C) show the foveal capillary network of the same subject. (D) AOOCT-A image taken by the multi-modal multiscale system created in the European MERLIN project. Panel (D) Courtesy of Kiyoko Gocho.

Download Full Size | PDF

The capillaries surround the FAZ represent the smallest vessels in the eye and multiple AO imaging techniques, such as confocal [157,158,166,169,186,187], fluorescein angiography [15,188], and non-confocal/split-detection [159,170,189], have been employed to visualize this vasculature region (Fig. 9). Several morphometrics have been used to quantify FAZ geometry in normal eyes and those with vasculopathies, including FAZ area, equivalent diameter, perimeter, and acircularity index [157,173]. AO ophthalmoscopy studies have demonstrated the wide range of FAZ size and shape variation in healthy controls, which frequently overlaps with pathologic conditions [157,158,170,189]. Combining fluorescein angiography with AO imaging has allowed evaluation of the complexity of the anastomotic foveal microvascular network in healthy and vasculopathic eyes [15,171]. Using this approach, subtle subclinical reduction in parafoveal microvascular density was detected in the ‘uninvolved’ fellow eyes of non-ischemic central retinal vein occlusion [172]. Similarly, increases in parafoveal capillary diameters and tortuosity have been revealed in diabetic eyes without clinically detectable retinopathy using confocal and non-confocal AOSLO [173,190].

Both AOFIO and AOSLO have been used to measure blood vessel lumen diameter and wall thickness in healthy controls and patients with hypertension and diabetes. Wall-to-lumen thickness ratios have been shown to increase significantly with age [175,191], as well as in patients with hypertension [174–177] and diabetes [190,192]. Figure 10 shows an example of wall-to-lumen ratio measurement of a retinal arteriole in a healthy subject imaged using offset pinhole AOSLO, in which the vessel wall, lumen, and mural cells that compose the arteriole walls are clearly visible [177,193].

 

Fig. 10. An arteriole located in the optic disc of a 26-year-old healthy subject imaged using non-confocal offset pinhole AOSLO. The outer diameter and inner diameter are indicated by the red and cyan line, respectively. Wall to lumen ratio is computed as the difference between the outer and inner diameters divided by the inner diameter [177]. Yellow arrows indicate examples of mural cells along the blood vessel walls [193].

Download Full Size | PDF

Microaneurysms are one of the earliest vascular abnormalities that can be detected with standard clinical techniques. However, the mechanism for the formation and resolution of individual microaneurysms remains a mystery. Using AO ophthalmoscopy, features of retinal microaneurysms have allowed them to be classified into different morphologic types [180,181,183,185,194]. Their association with neighboring disorganized retinal changes in diabetic eyes [182] supports the idea of a complex neurovascular dynamic orchestrating these changes. In addition to microaneurysms, AO imaging has provided visualization of other hallmark features including vessel loops and hairpins, capillary non-perfusion, capillary dilation, and neovascularization which may serve as additional qualitative metrics for early identification of microvascular disease (Fig. 11).

 

Fig. 11. Imaging of parafoveal microvasculature in human retina using non-confocal offset pinhole AOSLO. (A) Parafoveal capillary network in a 25-year-old healthy subject. (B) A 43-year-old diabetic patient with focal retinal microvascular changes such as capillary wall thickening (red arrow), capillary dilation (yellow arrows), and microaneurysm (cyan arrow).

Download Full Size | PDF

In contrast to the retinal vasculature, choriocapillaris is highly fenestrated with fine pore-like structures located immediately posterior to the Bruch’s membrane. Disrupted choriocapillaris in AMD is well documented in histopathologic studies [195–197]. Clinical imaging of choriocapillaris has been most challenging due to its dense capillary network and the overlaying melanin rich RPE which blocks the imaging light from reaching the target structure. The use of AOOCT-A [26,198] and AOSLO ICG angiography [199], has allowed in vivo imaging of choriocapillaris structure and perfusion in human retina. These imaging approaches hold great promise for tracking choriocapillaris changes for early diagnosis of disease such as neovascular AMD.

3.6 Immune Cells

In vivo imaging of retinal immune cells over time offers the potential to better understand the pathological significance of their activity, as reflected in their morphology (size, shape, density, and distribution geometry) and behavior (motility and chemotactic activity) in autoimmune and inflammatory diseases. Recent advances in label-free transparent tissue imaging have enabled direct visualization of immune cells using AO ophthalmoscopy in the living human retina [20,135,136,200–203]. Prior studies have demonstrated the use of AOOCT to observe the distribution and dynamics of macrophages on the inner limiting membrane (ILM) surface (presumably hyalocytes) [136,200–202]. Using this imaging technique, ILM macrophage density and motility were characterized in healthy controls and patients with glaucomatous damage. Ramified ILM macrophages appeared to be evenly distributed across the retina except for the central macula where cells were absent or sparse in healthy controls [201]. In glaucomatous eyes with hemi-field defects, however, cell distribution showed regional differences with higher cell densities over damaged regions than over adjacent healthy regions. This study also tracked cell motility over time; while cell soma tended to remain tethered to a general location, cell process motility was sporadic with high variability in velocity over ∼20-25 min of imaging in controls. A follow up study characterized hyalocyte morphological changes and displacement on the ILM surface over 1-2 hour periods using non-confocal quadrant-detection AOSLO [203]. In this study, different cell morphologies such as spindle-like, star-like, rod-like, and amoeboid shapes were identified in healthy controls. Various degrees of cell shape-shifting were characteristic of cell soma and process configuration over time (Fig. 12, Visualization 5).

 

Fig. 12. Hyalocytes imaging in a 32-year-old healthy subject using non-confocal quadrant-detection AOSLO. (A and B) Arrows indicate two ramified hyalocytes imaged at two time points with their somas and processes varying in shape and orientation noticeably. Time of acquisition in the lower-left corner denotes the hrs:mins:secs. (C) Red-green overlay shows hyalocytes at two time points. Yellow indicates stationary structures that appeared in both time points due to the combined contributions of red and green. The entire image sequences of these cells over 2 hours are shown in Visualization 5.

Download Full Size | PDF

4. Applications of adaptive optics for investigating retinal function

AO ophthalmoscopy has enabled numerous applications for investigating retinal structure. However, the observation of cellular retinal structure, intact or otherwise, does not necessarily imply cellular function. Thus, techniques to evaluate visual system function are required in addition to the high resolution structural imaging afforded by AO ophthalmoscopy. This need was understood early in the development of AO imaging techniques, and the past twenty-five years have seen a number of methods employed for studying the function of the visual system, and in particular, how it correlates with visual system structure.

Different types of functional measurements abound, but all measures can be categorized into two groups: psychophysical function and physiological function. Psychophysical measurements of function test visual perception and require a subjective response to a visual stimulus. On the other hand, physiological measurements of function test the body’s natural response, and are objective, as subjects cannot consciously alter the measurement. Examples of standard psychophysical measurements of visual system function include visual acuity, visual fields, and retinal sensitivity, while examples of physiological function include electroretinography (ERG), and pupilometry. These latter examples of physiological function really represent biomarkers of function; for example, having a normal ERG does not necessarily mean someone has normal vision, but a reduction in the ERG is known to be correlated with decreased visual system function. Thus, physiological measurements of function ultimately must be correlated with psychophysical measurements in order to serve as true functional measures.

Numerous AO studies have attempted to investigate the function of observed retinal structures, both in terms of physiological function and psychophysical function. Below we highlight several areas where applications of AO technology have revealed basic, translational, and clinical understandings of visual system function. We limit our discussion to applications which involve imaging, as AO vision simulators and studies that use AO to manipulate aberrations are covered in other reviews [204].

4.1 Adaptive optics microperimetry

Microperimetry is a technique which involves the combination of retinal imaging and stimulus presentation for assessing retinal sensitivity. The method presents calibrated stimuli in an adaptive fashion to precise locations in the retina. Typically, retinal tracking is employed to ensure stimuli are targeted to the same retinal position in sequential trials. Study participants are asked to respond to the presence of a stimulus for each trial, generally in a yes/no fashion, and stimuli irradiance is adjusted such that the stimuli irradiance at the threshold of vision is identified for each location tested. The concept of adaptive optics-guided (AO-guided) microperimetry, is the same, but it operates at the higher-resolution levels afforded by AO [205]. Indeed, AO-guided microperimetry provides two distinct advantages over clinical microperimetry: 1) the stimulus is delivered with aberration corrected optics and thus can be confined to a size equal to the point-spread function of the diffraction-limited eye; and 2) the high resolution of the tracking image provides increased accuracy for identifying where the stimuli landed on the retina. Indeed, AOSLO has yielded the fastest eye-tracking technology to date with eye-tracking rates of close to 1 kHz [206–208]. With these two advantages, AO-guided microperimetry has enabled testing the sensitivity of spots as small as single cone photoreceptors [209]. This may well represent the ultimate gold standard for assessing retinal sensitivity, for in the extreme example, the technology allows identification of the threshold level of vision for each individual cone cell. Admittedly, the technique is laborious both for the operator and the participant, as numerous trials are required to determine the sensitivity threshold for each location (or cell) tested. Thus, practically speaking, AO-guided microperimetry is a technique which is best used in vision science and clinical studies with well-defined experimental questions that can be answered with relatively few study locations in a low number of study participants. Despite this limitation, AO-guided microperimetry has found a place in correlating cellular structure with functional phenotypes in a number of clinical studies to date [71,210,211].

Initial experiments using AO-guided microperimetry investigated basic neuroscience questions about the visual system, for example over what extent does the visual system respond to single-cone-sized stimuli? Investigators found that subjects perceived single-cone-sized stimuli in foveal and parafoveal retinal regions, but single-cone-sized stimuli beyond approximately four degrees eccentric to the fovea in the visual field did not result in a percept regardless of stimulus contrast [209]. This finding provides critical information for what retinal regions should be targeted in single-cell AO-guided microperimetry studies. In addition, sensitivity thresholds for dark or weakly waveguiding cones in normal retina were found to be similar to thresholds measured in normally waveguiding cones, showing that cone reflectance is independent of cone sensitivity [212,213]. It was shown that complete spatial summation in the fovea, otherwise known as Ricco’s area, encompassed approximately two dozen cones [214] and that intra-subject sensitivity thresholds were lower for regions of higher cone density within the foveal retinal region [49].

There are still only a handful of studies that have used AO-guided microperimetry to assess retinal sensitivity in retinal disease and abnormal structural conditions. These studies have primarily been used to correlate structure and function within and adjacent to lesion areas, and have provided insight to disease conditions. The first such study was carried out in three patients with macular telangiectasia type 2 [210]. Here, participants showed measureable retinal sensitivities indicating retained function even within lesion areas that did not show normal waveguiding photoreceptors and which also lacked IS/OS and COST reflections on OCT. This was a surprising finding, as prior to that study, it was accepted that retinal regions lacking photoreceptor layer structure on OCT and waveguiding photoreceptors on AO imaging did not contain functioning cones. It was found however, that the areas with retained function despite the lack of visible photoreceptors, also exhibited an intact external limiting membrane (ELM) along with a hyporeflective gap in the underlying IS/OS junction layer on OCT imaging. Thus, AO-guided microperimetry revealed a new structural phenotype, the presence of which investigators could use to predict locations with retained function.

AO-guided microperimetry has also been used to investigate retinal sensitivity in patients with choroideremia [211]. AO structural imaging in choroideremia has revealed sharp transitions between retained islands of retina and areas of degeneration with loss of the photoreceptor mosaic [77,78]. AO-guided microperimetry confirmed a tight correspondence between structure and function; participants exhibited a visual percept when cones within the retained island were stimulated and did not experience a percept when retinal locations in the degenerated regions were stimulated even with the maximum contrast stimuli afforded by the AO-guided microperimetry system [211]. The border cones showed a small transition zone in retinal function, as with the structural findings. Another hallmark of choroideremia is the structural presence of numerous outer retinal tubulations (ORTs), where the ELM dives toward the choroid and wraps under itself forming tubes on OCT that contain degenerating outer retinal cells [215,216]. ORTs are present in locations with end stage disease, and en face appear as finger-like projections originating from the retained retinal island. AO-guided microperimetry in locations of ORTs also revealed a lack of a visible percept, implying that the remnant photoreceptors present in the ORTs were non-functional (Fig. 13) [211]. The tight correlation between structure and function revealed with AO-guided microperimetry in the case of choroideremia implies that structural biomarkers will be sufficient biomarkers for assessing disease progression and response to therapy in this disease.

 

Fig. 13. AO-guided microperimetry in conjunction with confocal (A) and non-confocal split-detection (B) AOSLO imaging in choroideremia reveals a sharp transition in retinal sensitivity at the atrophic border and along an outer retinal tabulation [211]. Circled dots show the test-locations and are scaled to the size of the stimulus on the retina. Figure courtesy of William S. Tuten.

Download Full Size | PDF

Finally, a third study investigated the cone sensitivity to cone density ratio using AO-guided microperimetry and AO imaging in patients with RHO and RPGR mutations [71]. Cones in RPGR-mediated retinal disease were found to be more functionally compromised than cones in RHO-mediated disease; the sensitivity to density ratio was reduced in RPGR disease and normal in RHO disease. This suggests that RPGR cones exhibit functional loss prior to structural loss, and implies the availability of a treatment window in which cone function can be restored prior to structural degeneration.

The power of all three AO-guided microperimetry studies in patient populations described here, arise from the high resolution structure-function correlations. In all cases, the functional assessments led to better understanding of the structural phenotype. Future studies with AO-guided microperimetry in additional disease conditions will aid in the interpretation of disease-specific structural phenotypes and lead to identifying the most appropriate outcome measures for assessing retinal disease and its treatment.

4.2 Densitometry

The earliest AO measurements of photoreceptor physiological function were reported soon after the initial application of AO to the eye. By employing carefully calibrated stimuli to selectively bleach L-, M-, and/or S-cone photopigment, investigators were able to measure a change in the reflectance of individual photoreceptors in response to the bleaching stimuli, and thereby could identify the cone subtypes within the trichromatic cone mosaic (Fig. 14) [217]. Of interest was the high variability in the L, M cone ratio in the parafoveal photoreceptor mosaic between individuals with normal color vision [218–220]. Psychophysical experiments to investigate color perception to small single cone spots soon proceeded, such that vision scientists now understand that color perception for small single-cone sized spots depends both on the cone subtype stimulated as well as the cone subtypes of the immediately surrounding cones [221–224]. The concept of photopigment bleaching has also been used to investigate photoreceptor reflectance, for example, it was shown that rod photoreceptor reflectance increased following light adaptation in Oguchi disease [225], in contrast to cone reflectance which did not show a trend in any direction.

 

Fig. 14. AO densitometry reveals the three cone classes in the normal retina [220]. (A) Shows a histogram of the number of cones versus the change in intensity between fully bleached and dark adapted photopigment. S-cones show little change in intensity following a full bleach, while L- and M- cones show a larger change. Blue and yellow lines depict Gaussian fits to the histogram and denote the S- and L-/M- cones respectively. (B) Histogram of the number of cones versus the angular coordinate of the change in intensity for a selective L-cone bleach verses a selective M-cone bleach. Red and green lines show Gaussian fits to the histogram and denote the L- and M- cones respectively. (C) Psuedo-colored cone mosaic depicting the L-, M-, and S- cones in a normal trichromat in red, green and blue, respectively. Figure courtesy of Ramkumar Sabesan.

Download Full Size | PDF

4.3 Optoretinography

Recent advances in physiological function measurements include the emerging field of optoretinography [24,226–234]. The term ‘optoretinogram’, coined by Mulligan and Macleod [235], refers to an optical signal that is measured in response to a visual stimulus. Its clinical corollary is the well-known electroretinogram (ERG), which measures an electrical signal in response to calibrated visual stimuli. Optoretinogram signals have also been called intrinsic signals or scintillation [226,227,229]. In essence, the method involves imaging the retina using an infrared illumination source, directing a calibrated visual stimulus to the photoreceptors, and measuring a change in the infrared image following the stimulus onset in comparison to the pre-stimulus image.

To date, most optoretinogram studies have investigated signals from normal cone photoreceptors. En face, the infrared reflectance of cone photoreceptors increases or decreases, or scintillates, in response to a visible stimulus. This optoretinogram signal has been measured over a population of cones where the measure was demonstrated to have an action spectrum which followed the photopic luminosity function, or the visual sensitivity of the eye [229]. We note that in contrast to densitometry which requires imaging with light in the visible spectrum to capture changes in the photopigment density between light and dark adaptation, optoretinography signals are measured with infrared illumination following the presentation of a visible stimulus, and the reflectance changes observed in the infrared are therefore different from the reflectance changes observed in the visible spectrum using densitometry methods. The earliest proposed mechanism for optoretinography was that the signal arose from interference between the multiple back reflections throughout the cone photoreceptor, primarily between the IS/OS junction and the COST layers [226]. The interference mechanism explained the findings that cone infrared reflectance could both increase and decrease in response to the same visible stimulus, and that the magnitude of the changes increased when using coherent imaging sources as opposed to incoherent or partially coherent imaging sources.

More recently, phase imaging systems, including AOOCT, have confirmed that the optoretinogram signal arises from a change in the optical path length (OPL) of the photoreceptor following phototranduction caused by the visible stimulus [232]; specifically, this OPL change is between the IS/OS junction and the COST [233]. Changes in the optical path length can be caused by a change in the physical distance between the two layers, a change in the refractive index of the outer segment, or both. The change in the OPL (ΔOPL) shows two characteristic components, first a fast ΔOPL contraction, and second a longer ΔOPL elongation [232]. The elongation ΔOPL has been shown to increase with increasing irradiance up to a saturation level [24]. It can be used to classify the L-, M- and S-cones (Fig. 15) [231,236] and is thought to arise from a physical lengthening of the outer segment [232].

 

Fig. 15. AO optoretinography reveals cone function and enables classification of L-, M-, and S-cones [231,234]. (A) Cone mosaic revealed with line-scan AOOCT. (B) Pseudo-colored cone mosaic depicting the L-, M-, and S-cones in red, green, and blue, respectively. Cone type was assigned based on the magnitude of ΔOPL following a 660 nm visible stimulus. (C) ΔOPL versus time after the stimulus onset. L-cones exhibit the strongest ΔOPL following the stimulus. (D) Histogram showing the number of cones versus the ΔOPL for the 660 nm stimulus. Blue, green, and red lines show three Gaussian fits to the data, and represent the cones that constitute the S-, M- and L-cones respectively. Figure courtesy of Ramkumar Sabesan.

Download Full Size | PDF

To date, optoretinography has only been explored in retinal disease in one published study [237]. Cones in retinitis pigmentosa patients were shown to have decreased optoretinogram signals within the transition zone of retinal degeneration, with only a handful of cones retaining ΔOPL magnitudes capable of determining their L-, M-, and S- cone phenotypes in the most advanced case, despite a high number of cells remaining structurally present in the AOOCT en face image. Of interest for future studies to explore is whether the cones that maintain a higher ΔOPL magnitude also exhibit higher sensitivity. Here, AO-guided microperimetry combined with AO optoretinography may play a pivotal role in establishing the needed link between the physiological optoretinogram signal and psychophysical perception to enable optoretinography to serve as a biomarker of visual system function. Additional studies correlating changes in optoretinogram signals with disease states, progression, and prognostic or treatment outcomes along with a well characterized database of optoretinography norms will be needed for optoretinography to gain traction in a clinical environment.

Optoretinography ΔOPL signals have also been explored in inner retinal neurons [238] and the ganglion cells downstream of the photoreceptors. Pfaffle et al. [239] showed that OCT (without AO) could measure ΔOPL signals in both the ganglion cell layer and photoreceptor cell layer simultaneously, that the pattern of the signals corresponded with the stimulus shape and the expected retinal displacement of the receptive fields for the group of activated ganglion cells, and that the ΔOPL of the ganglion cell layer was an order of magnitude smaller than the ΔOPL of the photoreceptor layer. Though this study lacked the cellular resolution afforded by AO-equipped devices, it provides strong evidence that optoretinography signals may be measurable across multiple cell types in the retina simultaneously. This will surely continue to be an area of interest in future studies as investigators work toward establishing optoretinography throughout the full retina.

4.4 Angiography

Although ultra-wide-field fundus photography and intravenous fluorescein angiography have come to be regarded as the clinical gold standards for characterizing retinal microvascular pathology, these methodologies are unable to visualize hemodynamics due to their limited spatial and temporal resolution. Major advances in AO imaging, including image acquisition strategies and offline image processing algorithms, have enabled the visualization of vessel wall components and moving intra-vascular blood cells in a dynamic fashion across a range of vessel diameters without the need of exogenous contrast agents.

Blood cell velocities in parafoveal capillaries have been determined using confocal AOSLO by tracking the shadows of individual leukocytes (light regions) [240–242] or aggregated erythrocytes (dark regions, immediately behind the light regions) [243] cast upon the underlying photoreceptor layer. While blood cell velocity of ∼1.3-1.5 mm/s was reported in healthy controls [244,245], no statically significant difference was found in diabetic patients without retinopathy when compared to controls [173,245]. One major drawback of these techniques is that measurements are limited to capillaries with “leukocyte-preferred paths” adjacent to the FAZ only. For larger blood vessels, a novel confocal AOSLO method which switches from raster-scanning to line-scanning mode by pausing the slow scan on the target blood vessel momentarily, is able to obtain a high-speed profile of erythrocyte flow across the vessel lumen, yielding precise velocity measurements in vessels larger than 30 µm in diameter [246,247]. Using this approach, retinal blood velocity and flow were found to be significantly higher in diabetic patients without retinopathy but lower in mild non-proliferative diabetic retinopathy as compared to non-diabetic controls [248]. These findings suggest that the increase in blood flow in early stage of diabetes could lead to vascular endothelial injury, resulting in capillary dropout and decrease in blood flow in non-proliferative diabetic retinopathy. A more recent approach using high-speed AOLSO revealed erythrocyte velocities of 0.7 to 2.6 mm/s in retinal capillaries [249]. Direct visualization of erythrocytes using a high-speed AOFIO [250,251] or dual-beam non-confocal AOSLO with a small temporal offset between channels [252–254] has also allowed velocity assessments across the retinal capillary network. All these approaches have demonstrated a large spatiotemporal variability in erythrocyte velocity (0.3 to 4.0 mm/s) with pronounced cardiac-dependent pulsatile flow velocity at the capillary level [249,251,252,254,255], possibly due to physiological modulation of blood flow within the capillary networks (Fig. 16).

 

Fig. 16. Imaging parafoveal capillary flow in a subject with type I diabetes [256]. (A) Shows an OCT-A image acquired using a commercial device (Heidelberg Spectralis). Foveal avascular zone is located at the lower right. (B) Motion contrast map acquired with an AOFIO at 400 fps with 593 nm imaging beam superimposed upon the corresponding region of the background OCT-A image. Image sequence of the red box region is shown in Visualization 6. (C) Velocity mapping using pixel intensity cross-correlation, in the same region as (B). Velocity color map ranges from 0 to 4.5-mm/s. Figure courtesy of Phillip Bedggood and Andrew Metha.

Download Full Size | PDF

Neurovascular coupling describes the mechanism by which local perfusion is modified in response to changes in adjacent neural activity. Alterations in neurovascular coupling have been observed in a wide variety of neurodegenerative and vascular diseases such as Alzheimer’s disease and diabetes [257,258]. Direct assessment of neurovascular coupling can be achieved by measuring retinal microvascular response to flickering stimuli while imaging living human retinas with AO imaging. This approach has demonstrated local retinal blood flow modulation in response to dynamic visual stimuli, indicating the presence of hemodynamic regulation due to the changes of metabolic demands in healthy human retina [253,259,260]. Future studies of patients with neurovascular diseases such as Alzheimer’s disease or diabetes should provide fundamental insights into the nature of impairments in neurovascular coupling in these conditions.

5. Clinical applications of adaptive optics ophthalmoscopy to evaluate therapeutic response

The Food and Drug Administration’s approval of voretigene neparvovec-rzyl (Luxturna) [261] marks a new era in the fight to cure blindness, in which disease treatments are targeted at the cellular level. Indeed, numerous therapeutic approaches for blinding disease are under development including gene and small molecule therapies [262], stem cell transplantation [263], and optogenetic therapies [264]. These therapeutic approaches attempt to reverse vision loss by restoring function to populations of individual cells that either remain structurally present or that are structurally implanted/grown. Experimental therapies must be tested prior to their acceptance, and the economics of such clinical trials put a high value on outcome measures that can demonstrate improvements (or lack thereof) in response to therapy as quickly as possible after intervention. Such outcome measures will be best formulated in the context of the most precise possible characterizations of disease progression. Standard techniques to assess visual system health, such as visual acuity or retinal sensitivity, integrate signals across hundreds of photoreceptors. As already discussed, AO ophthalmoscopy has enabled cellular-level visualization of retinal structure and function, and as such has high potential to show the cellular level changes of interest both in the natural progression of disease and in the application of therapeutics which alter cellular degeneration. Further, AO ophthalmoscopy may identify those patients who have the potential to most benefit from treatment by identifying patients and retinal locations which retain retinal structure but exhibit dysfunction. Clinical validation of the cellular-scale biomarkers made accessible thanks to AO technology thus could accelerate the development and adoption of new therapies and ultimately save sight.

Despite these advantages, there remains only a handful of publications evaluating experimental therapeutic approaches using AO ophthalmoscopy. The first followed three patients who were part of a larger study group all treated with a ciliary neuro-trophic factor (CNTF) implant [265]. The investigators measured cone densities in the implant-treated eyes compared to sham-treated eyes and untreated normal eyes. Cone densities in the sham treated eyes decreased faster than cone densities in the eyes with the CNTF implant in the group of three patients. The results from the larger study group however, did not reveal a therapeutic benefit for the implant-treated eyes in terms of measured visual acuity or visual field sensitivity [266].

A second peer-reviewed publication applying AO ophthalmoscopy to evaluate an experimental therapy followed nine choroideremia patients who were treated with a gene therapy delivered via a subretinal injection to detach the neural retina from the RPE [267]. In this study, investigators found that cone density remained stable between baseline and one-month post-injection time points, implying that there was not widespread toxicity to the study agent and that the subretinal injection procedure did not cause wide-spread cone loss despite the forceful detachment of the neural retina (Fig. 17).

 

Fig. 17. Cone inner segment mosaic visible using non-confocal split-detection AOSLO imaging before (top) and after (bottom) subretinal injection of AAV2.hCHM gene therapy in a patient with choroideremia [267]. Global alignment of the montages shows the same retinal features (yellow arrows) before and after application of the experimental therapeutic. Regions of interest (colored boxes) show magnified images of the photoreceptor mosaic; cone densities between baseline and one-month post injection timepoints were not significantly different. The data provide evidence supporting the safety of subretinal injections of AAV2.hCHM. Yellow asterisk denotes the location of the fovea.

Download Full Size | PDF

Finally, two studies document an AOFIO image from case studies of AMD patients treated with transplanted stem cell-derived RPE. The first claims photoreceptors were visible in a small region over the transplanted RPE patch in two cases [268]; the second claims RPE cells were visible at the upper margin of the RPE graft sheet in one patient [269].

In addition to assessing the safety and efficacy of experimental therapeutics such as those described above, AO ophthalmoscopy can be used to better understand already accepted and approved treatments for disease. For example, photoreceptor structure in RPE65-mediated disease following subretinal administration of Luxturna has been examined using AOFIO [270]. As a second example, the improved visualization of individual erythrocytes using non-confocal AOSLO imaging has allowed for the study and characterization of erythrocyte-mediated vaso-occlusive mechanisms and treatment evaluation in sickle cell disease. In sickle cell disease, the mutated hemoglobin is prone to precipitating out of solution, forming long, rigid intra-cellular polymers, causing erythrocyte cell wall damage and the characteristic sickle shape of erythrocytes. The result of ongoing erythrocyte damage is hemolytic anemia and recurrent small vessel occlusion. Using non-confocal quadrant-detection AOSLO, a recent study revealed different mechanisms of erythrocyte-mediated vaso-occlusion in sickle cell disease, including erythrocyte rouleau occlusion, sickled erythrocyte occlusion, and erythrocyte aggregation and thrombus formation at capillary bifurcations and within capillary segments [271]. In this study, AO imaging was also able to document resolution of thrombi and various perfusion changes after the initiation of oral hydroxyurea therapy (Fig. 18, Visualization 7).

 

Fig. 18. (A) Right eye fundus photo of a sickle cell disease patient, a treatment-naïve 31-year-old female with HbSS genotype and non-proliferative sickle cell retinopathy. White box indicates a region imaged at two visits using non-confocal quadrant-detection AOSLO [271]. (B) At the baseline visit, single frame image revealed two thrombi of blood cells within the same capillary (cyan arrow). A sludged erythrocyte (white arrow) and a fully perfused capillary (yellow arrow) were also visualized. (C) At the second visit, 2 months following initiation of oral hydroxyurea treatment, single frame image showed resolution of the thrombi (cyan arrow), and restoration of normal flow through the previously non-perfused capillary (white arrow). Interestingly, non-perfusion of a previously perfused capillary segment (yellow arrow) was also evident two months after treatment. Perfused blood vessels are tinted in red based on the corresponding motion contrast perfusion maps. See also Visualization 7.

Download Full Size | PDF

These studies highlight the potential for AO ophthalmoscopy to provide cellular level assessment of interventions beyond those that are clinically available. In the case of the CNTF implant, the retention of cone structure shows a positive result for the overall health of the visual system [272]; despite the lack of functional benefit, at a minimum retention of cone structure may enable future therapies to restore function to those cells even if the CNTF implant did not. In the case of gene therapy, AO imaging provided evidence in support of the safety of the subretinal injection surgical procedure, a finding that will be more broadly applicable to trials beyond choroideremia. In the case of sickle cell disease, AO imaging verified a response to treatment through the resolution of thrombi and improved perfusion. Future studies will almost certainly increasingly use AO ophthalmoscopy to assess therapeutic response both in multi-center clinical trials for experimental therapies and in clinics using accepted treatments.

6. Challenges for adaptive optics ophthalmoscopy in clinical use

Despite the numerous clinically-relevant applications described in this review, AO ophthalmoscopy has yet to alter standard clinical care. Ophthalmic imaging is used primarily to direct patient care, such as identifying when or how to intervene in a disease process. Acquisition of an AO image has yet to alter this decision making process, as for instance a fluorescein angiography image can direct the application of anti-VEGF treatment for AMD, or an OCT image can indicate surgical intervention for a macular hole. Applications which alter decisions regarding patient care are more likely to attain widespread adoption. Once AO ophthalmoscopy can provide clinicians with information that necessitates intervention, even one as ‘simple’ as determining the rate at which a patient should return to the clinic for follow-up monitoring, we predict the demand for AO devices will rapidly increase as patients and clinicians alike will insist on receiving/providing the best clinical care available. Indeed, we view the current lack of a care-altering application at the present time as the biggest hurdle AO ophthalmoscopy must yet overcome.

This begs the question: what will be the AO imaging application that alters clinical care? Photoreceptor imaging? RGC imaging? Vascular biomarkers? Functional responses? Something else? In the story of OCT, it was the ability to visualize macular holes needing surgical intervention as well as the development of a new treatment for wet AMD that launched OCT technology into standard care practices. Once there, ophthalmologists started exploring the use of OCT imaging in other conditions, to the extent that now OCT is a standard part of retinal specialists’ exams. In the case of AO, if treatment were again to be the driver of clinical adoption, this drive may come from innovations in optogenetics, small molecule, or gene and cell therapies. These therapeutic approaches, in particular, operate at the cellular level, and thus may benefit from AO ophthalmoscopy’s ability to follow cellular scale changes, which in theory should lead to detectable biomarkers at earlier stages than any overall improvement in visual performance. Developers of these therapies have a strong incentive to validate short-term biomarkers of therapeutic efficacy in order to accelerate adoption of their treatments. Indeed, the number of publications which apply AO ophthalmoscopy to investigate disease treatment (experimental or approved) has doubled since March 2022 [267,270,271] from those available earlier [265,268,269]. We predict this trend will continue as investigators increasingly examine disease treatment at the cellular level and we look forward to the discovery which will drive AO imaging technology into widespread clinical use.

If one were to ask a group of scientists what are the remaining challenges associated with AO imaging, one might get a number of unique answers such as problems associated with: large data storage needs, timely image processing and analysis, scanning distortions, or patient conditions (ocular opacities, dry eye, small pupils, large eye movements). Ask a group of clinicians and one might get a different set of responses such as: image acquisition time, operator training, system maintenance, image interpretation or a lack of reading centers. Ask an administrator, and one may hear about challenges surrounding: high device costs, a lack of regulatory approvals, or no available procedure reimbursement. Certainly, these are real challenges given the current state of the AO imaging field. We stress however, that these are all solvable problems; given sufficient attention and financial resources, we have no doubt that multidisciplinary teams of engineers, scientists, and clinicians from academia, corporations, and government bodies will be able to find needed solutions. Such resources will almost assuredly be given once the application(s) to alter clinical care is identified. In this regard, the increasing use AO imaging in clinical trials and studies of disease interventions will help to speed discovery.

7. Conclusion

In summary, AO ophthalmoscopy has enabled multiple avenues for investigating visual system structure and function in health and disease over the past twenty-five years. Innovative imaging systems and analysis combined with clinically-relevant applications have yielded insights to mechanisms of retinal disease and its treatment which were previously not possible. Further, the technology itself has had time to mature, and is increasingly available to both scientists and clinicians. The future of AO ophthalmoscopy is bright, and we look forward to the next quarter-century of discovery its application will bring.