In vivo imaging of human retinal pigment epithelial (RPE) cells has been demonstrated through multiple adaptive optics (AO)-based modalities. However, whether consistent and complete information regarding the cellular structure of the RPE mosaic is obtained across these modalities remains uncertain due to limited comparisons performed in the same eye. Here, an imaging platform combining multimodal AO-scanning light ophthalmoscopy (AO-SLO) with AO-optical coherence tomography (AO-OCT) is developed to make a side-by-side comparison of the same RPE cells imaged across four modalities: AO-darkfield, AO-enhanced indocyanine green (AO-ICG), AO-infrared autofluorescence (AO-IRAF), and AO-OCT. Co-registered images were acquired in five subjects, including one patient with choroideremia. Multimodal imaging provided multiple perspectives of the RPE mosaic that were used to explore variations in RPE cell contrast in a subject-, location-, and even cell-dependent manner. Estimated cell-to-cell spacing and density were found to be consistent both across modalities and with normative data. Multimodal images from a patient with choroideremia illustrate the benefit of using multiple modalities to infer the cellular structure of the RPE mosaic in an affected eye, in which disruptions to the RPE mosaic may locally alter the signal strength, visibility of individual RPE cells, or even source of contrast in unpredictable ways.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Cellular level assessment of the retinal pigment epithelial (RPE) mosaic has provided critical insight into the role of these specialized cells in both normal vision and disease [1,2]. Histological study has permitted the construction of high-resolution maps of parameters thought to be relevant for susceptibility to eye disease, such as cell area, cell density, and pigmentation [3–6], and has advanced the understanding of the onset and progression of disease [7–9]. Most clinical approaches for imaging the RPE mosaic, such as fundus autofluorescence or infrared autofluorescence, reveal tissue level information but to date, cellular assessment of the human RPE mosaic remains challenging in most clinical settings without specialized technology such as adaptive optics (AO).
AO is a technology that can be combined with ophthalmic imaging instruments to achieve cellular-resolution imaging of the human retina by correcting for monochromatic ocular aberrations . To date, AO has been demonstrated in combination with both scanning light ophthalmoscopy-based systems (adaptive optics-scanning laser/light ophthalmoscopy, AO-SLO ) and optical coherence tomography (adaptive optics-optical coherence tomography, AO-OCT ) to achieve cellular resolution. While initial applications of AO in ophthalmology were focused on photoreceptor imaging, the capabilities of these technologies have subsequently been extended to image other structures [13,14], including the RPE mosaic. AO-SLO-based methods for imaging the RPE mosaic include AO-darkfield imaging based on non-confocal detection of scattered light , late-phase AO-enhanced indocyanine green (AO-ICG) imaging based on the fluorescence of indocyanine green (ICG) dye that is heterogeneously taken up by RPE cells following intravenous injection [16–18], and AO-near-infrared autofluorescence (AO-IRAF) imaging [19–21] based on the endogenous fluorescence of melanin [18,22]. AO-OCT imaging of the RPE mosaic has also been demonstrated based on time-gating of backscattered light from the RPE cell layer combined with volume averaging to mitigate speckle by exploiting organelle motility [23,24].
While in vivo human RPE cell imaging has been successfully demonstrated with each of these modalities, each technique has its own unique limitations which can hinder image interpretation. Additionally, each modality has its own inherent strengths and weaknesses based on interrelated factors that include imaging speed, signal-to-noise ratio (SNR), and specificity of contrast to RPE cells (discussed throughout this manuscript and summarized in Table 1). Together, these factors lead to variability in visualizing the RPE mosaic across an image that is further confounded by inter-subject variability that may also be dependent on imaging modality. In addition, there are local variations according to the retinal location in which images are acquired, and often, even across neighboring cells within a single image, adding to the complexity of consistently discerning the cellular structure of the RPE mosaic. In diseased eyes, when the RPE mosaic is disrupted, acquired images may appear quite different compared to images from healthy subjects and interpretation of RPE cell structure may be subjective and difficult to validate, thereby motivating the need for side-by-side comparison of RPE images within the same eye. Here, we investigate how the combination of multiple imaging modalities based on recent implementations [15,18,19,23] in an integrated imaging device is beneficial in achieving a consistent interpretation of the structure of the RPE mosaic, building upon prior studies that have integrated multiple imaging modalities [25–28].
2.1. Multimodal AO-SLO and AO-OCT system
A custom multimodal AO-SLO  was modified to incorporate spectral-domain (SD) AO-OCT (Fig. 1), which was operated independently of AO-SLO imaging. The AO-SLO used a 790 nm superluminescent diode (SLD) (S-790-G-I-15-M, Superlum, Carrigtwohill, Co. Cork, Ireland), and an 880 nm SLD (SLD-mCS-341-HP1-SM-880, Superlum, Carrigtwohill, Co. Cork, Ireland) for imaging and wavefront sensing, respectively. A 1080 nm SLD (EXS 210007-01, Exalos, Schlieren, Switzerland) was used for AO-OCT imaging. The 790 nm AO-SLO light source has a full width at half maximum (FWHM) bandwidth of 16 nm that is further reduced to approximately 15 nm through the use of a clean-up spectral filter (ET775/50x, Chroma, Bellow Falls, VT, USA). The 880 nm wavefront sensing source has a FWHM bandwidth of 46 nm that is further reduced to 20 nm through the use of a clean-up spectral filter (FF01-900/32, Semrock, Rochester, NY, USA). The 1080 nm AO-OCT light source has a FWHM bandwidth of 91 nm. The measured optical power at the cornea was below 135 µW for the 790 nm source, below 43 µW for the 880 nm source, and below 1.43 mW for the 1080 nm source for all subjects. Although not all were used simultaneously, when used in combination, the optical power of these sources is below the maximum permissible exposure limit established by the American National Standards Institute standard Z136.1-2014 . As the AO-OCT acquisition rate in this instrument is limited by the spectrometer speed, independent, optically-conjugate horizontal scanners are used for the two subsystems: a 15 kHz resonant scanner (SC-30, Electro-Optical Products Corp, Fresh Meadows, NY, USA) for the AO-SLO system, and a non-resonant galvanometric scanner (GVS011, Thorlabs, Newton, NJ, USA) for AO-OCT (for C-scan acquisitions). The AO-SLO and AO-OCT beams are combined using a short pass dichroic mirror (T970dcspxr-UF3, Chroma, Bellows Falls, VT, USA) inserted into the AO-SLO beam path between the horizontal scanners of each subsystem (RS and GS respectively in Fig. 1) and the shared vertical tip-tilt scanner (TT in Fig. 1, S-334, PI-USA, Auburn, MA, USA). Telescope design and adaptive optics subsystems are based on the previous implementation of this custom-built multimodal AO-SLO system [30,31].
The backscattered 790 nm AO-SLO light returning from the eye is collected with a photomultiplier tube (PMT) in confocal (Confocal PMT in Fig. 1) and non-confocal (Split PMTs in Fig. 1) detection channels that split the light after an annular aperture that reflects the central 2 Airy disk diameters (ADD), and transmits light outside 2 ADD, while fluorescence from the eye (AO-IRAF and AO-ICG) is collected in a separate PMT (Fluorescence PMT in Fig. 1) with a larger confocal aperture (7.5 ADD and 5.0 ADD for AO-IRAF and AO-ICG, respectively) after passing through a fluorescence emission filter (FF01-832/37, Semrock, Rochester, NY, USA). In addition to images of the RPE mosaic, photoreceptor images are collected through confocal (1.2 ADD pinhole) and non-confocal split detection  channels primarily for eye motion compensation and multimodal image registration. Backscattered 1080 nm light from the eye is combined with a reference beam (Ref. Arm in Fig. 1) and collected using a high-speed spectrometer (C-980-1180-GL2KR, Wasatch Photonics, Durham, NC, USA) for AO-OCT. The theoretical axial resolution of the AO-OCT is 5.7 µm in air (approximately 4.1 µm in the retina).
2.2. Imaging procedure
Human subjects were recruited through the National Institutes of Health Clinical Center for AO imaging. The four healthy subjects (age 29 ± 11 years; mean ± SD) included in the study showed no signs of ocular disease after a comprehensive dilated eye examination was performed. To assess a condition in which the RPE mosaic was thought to be affected, one patient with choroideremia (age 41 years) was recruited. Research procedures adhered to the tenets of the Declaration of Helsinki and were approved by the local Institutional Review Board (National Institutes of Health). Written informed consent was obtained for all participants.
Eyes were dilated with 2.5% phenylephrine hydrochloride and 1% tropicamide before imaging. AO-SLO and AO-OCT videos were acquired at selected retinal locations for imaging RPE cells. Retinal locations ranging from 0 to 2.5 mm in the temporal direction relative to the fovea were chosen for this study in order to provide a direct comparison to locations imaged in previous studies . A field-of-view (FOV) ranging from 1.5–2.0 degrees (750 × 605 pixels) was used to acquire AO-SLO videos, while a FOV of 1.5 degrees (300 × 300 pixels) was used to acquire AO-OCT videos. Imaging was performed in three stages. To avoid crosstalk of the two fluorescence modalities that share a common detection path, AO-IRAF and AO-darkfield images were collected first. Next, at least 45 minutes following intravenous ICG injection at the standard clinical dose administered at the National Eye Institute Eye Clinic (25 mg in 3 mL), late-phase AO-ICG and AO-darkfield images were collected as previously described after allowing the heterogeneous RPE pattern to stabilize [16,18]. Finally, volumetric AO-OCT videos were collected in a manner similar to previous studies . Volumetric AO-OCT imaging was performed using only the 1080 nm source for imaging and the 880 nm source for wavefront sensing. For each retinal location, 25–30 videos, each containing 3–5 volumes, were collected. For AO-OCT imaging, retinal locations were paired (acquisition of videos alternated between the two retinal locations) to ensure sufficient decorrelation of the speckle pattern observed in the RPE layer between consecutive videos for improved contrast of RPE cells.
2.3. Image processing and analysis
Eye motion in AO-SLO images was compensated through strip-based image registration based on simultaneously-collected images of photoreceptors acquired by AO-SLO confocal imaging . AO-darkfield images were constructed by summation of the two non-confocal reflection images (Split PMTs in Fig. 1). For AO-OCT videos, retinal volumes were digitally flattened based on the outer retinal layers to mitigate axial eye motion. Lateral eye motion was compensated with the same strip-based image registration procedure used for AO-SLO image registration  after generating a 2D en face projection of the photoreceptor layers, similar to what has been described previously for RPE cell imaging . RPE images were extracted from a single en face slice of the averaged OCT volume (1.3 µm between adjacent slices).
A cross-modality registration procedure was developed to further improve cell-to-cell alignment across modalities. First, all modalities were co-registered by scaling and translation of the corresponding photoreceptor images. Next, when eye motion artifacts were minimal, cross-modality strip-based registration was performed by registering frames from separate video acquisitions to a common reference frame (see Visualization 1 and Visualization 2).
Spatial frequency content of all RPE imaging modalities was objectively compared through their 2D power spectra. Row-to-row spacing was estimated from the power spectra and converted to cell-to-cell (center-to-center) spacing based on the assumption of hexagonal packing . Cell-to-cell spacing and density were further quantified through manual identification of RPE cell centers in each acquired modality by a single expert reader and subsequently edited and validated independently by a second expert reader for all analyzed retinal locations in this study. Only images from which the locations of cell centers could be inferred were included in this analysis. Cells were identified from each modality in accordance to previously published methodologies, based on the following criteria:
- • In AO-darkfield, AO-IRAF, and AO-OCT images, cells were identified by dark cell centers surrounded by bright cell boundaries. In image regions of low contrast within AO-darkfield and AO-IRAF images, cell centers were inferred based on the approximate hexagonal arrangement and size of neighboring higher-contrast RPE cells.
- • As has been described previously [16,18,36], initial cell identifications in AO-ICG images were made in high contrast image regions where the heterogeneous ICG pattern enabled delineation of borders between individual RPE cells based on varying levels of ICG in neighboring cells. The remainder of cells present in the image were inferred based on the expected size of neighboring RPE cells, presence of partial cell borders, and approximate hexagonal arrangement of RPE cells relative to the initial cell identifications.
These data were used to calculate density recovery profiles based on a hexagonal packing arrangement  and quantify cell-to-cell spacing and density. Finally, to characterize discrepancies between cell centers identified in each modality, a final set of RPE cell centers was manually established by a single expert reader taking into consideration the combined information provided by all RPE cell imaging modalities in a single co-registered image stack. This new set of cell centers was used as the seed for k-means clustering of the validated cell centers from each modality to initialize the cluster locations near the expected cell centers based on the combination of all modalities. Following k-means clustering, cell center error was calculated for each modality as the distance between the expert-validated cell center positions and cluster centroids.
2.4 Statistical analysis
All data are presented as mean ± SD. Cell center error across imaging modalities was compared with Kruskal-Wallis one-way ANOVA. Changes in RPE cell-to-cell spacing and density with eccentricity were examined through linear regression. Linear regression coefficients obtained from analysis on k-means clustered cell centers were compared to those obtained from normative data with the Chow test. For all test statistics, p < 0.05 was considered significant.
3.1. Visualization of RPE cells based on multiple contrast sources from AO-SLO and AO-OCT
Quad-modality images for side-by-side comparisons of the RPE mosaic were successfully obtained using the combined AO-SLO and AO-OCT system. An example set of co-registered images acquired from the fovea (Fig. 2) demonstrates multiple contrast mechanisms allowing visualization of different aspects of RPE cells. Images appeared qualitatively similar to those reported from previous studies. AO-darkfield imaging [Fig. 2(A)] revealed dark cell centers and brighter cell boundaries, albeit with additional low spatial-frequency content (spanning tens of RPE cells across), of unknown origin. The heterogeneous pattern formed by ICG uptake of RPE cells [Fig. 2(B)] provides exogenous fluorescence contrast where a subset of cell boundaries is discerned based on varying levels of ICG fluorescence between two neighboring cells. RPE cells imaged with AO-IRAF [Fig. 2(C)] showed dark cell centers and bright cell boundaries. Finally, en face AO-OCT images from the RPE cell layer [Fig. 2(D)] showed bright cell boundaries contrasting with dark cell centers. In many cases, cells are identified across all modalities (green arrows in Fig. 2). However, in other instances, cells could not be easily delineated in an individual modality but were more evident when comparing the same location across all modalities (magenta arrows in Fig. 2). In this example demonstrating quad-modality imaging of the RPE cells at the fovea, RPE cell contrast is relatively low in the AO-darkfield image, but much higher in the AO-ICG, AO-IRAF, and AO-OCT images, allowing visualization of cellular structure through the combination of these imaging modalities.
Further imaging performed at 10 retinal locations in four healthy subjects showed some variation in RPE cell contrast, both across subjects as well as across retinal location in individual subjects and even across neighboring cells within a single image (Fig. 3). Poor delineation of RPE cells was observed in some cases due to poor SNR [Fig. 3(K)], poor contrast [Figs. 3(E), 3(F), 3(I), 3(J)], or the presence of residual eye motion that could not be corrected [Figs. 3(I), 3(J), 3(L)]. This multimodal approach clarifies the underlying RPE mosaic structure in these situations. As an example, the presence of residual photoreceptor structure is a common artifact in AO-darkfield imaging [green arrows in Fig. 3(A)] leading to difficulties in image interpretation. In this case, images from other modalities were used to definitively identify RPE cells and prevent misinterpretation of AO-darkfield images. In AO-ICG imaging, it is common to observe areas of neighboring cells with similar fluorescence levels [green arrows in Fig. 3(B)], leading to challenges in the identification and enumeration of individual cells in these regions. Here, cellular structure observed in other modalities is used to more reliably determine the number of cells in these regions as well as to pinpoint their locations. Due to the typically lower SNR in AO-IRAF imaging, determination of cell boundaries can be difficult [green arrows in Fig. 3(C)]. This is remedied by looking to the AO-OCT, AO-ICG, and AO-darkfield images where cell boundaries may be more clearly defined. Finally, AO-OCT imaging, as implemented here, is more susceptible to residual motion artifacts resulting in image distortions [green arrows in Fig. 3(D)] owing to the order of magnitude difference in frame/volume rates between AO-SLO images and AO-OCT volumes. Images from other modalities are used to identify these regions and recover a more accurate representation of the underlying cellular structure. These examples illustrate specific cases where difficulties in image interpretation from a single imaging modality are mitigated through the collection and examination of co-registered multimodal datasets.
3.2. Multimodal AO-SLO and AO-OCT images of the RPE mosaic have consistent spatial scales
In addition to the different sources of contrast for RPE imaging, each modality contains consistent information about the spatial scale of the RPE mosaic. This is observed by examining the power spectra of co-registered images (Fig. 4). AO-darkfield [Fig. 4(A)], AO-IRAF [Fig. 4(C)], and AO-OCT [Fig. 4(D)] show similar features in their power spectra [Fig. 4(E), 4(G), and 4(H) respectively] where a bright ring can be observed corresponding to the row-to-row spacing of the RPE cell mosaic originating from the spatial arrangement of bright cell boundaries and dark cell centers of individual RPE cells. In contrast, the power spectrum of the AO-ICG image [Fig. 4(B)] shows a dissimilar, but related pattern in which a band of spatial frequencies is observed with a sharp cutoff frequency. Here, the power spectrum calculated from the heterogeneous ICG pattern contains a mix of lower spatial frequency content corresponding to groups of two or more neighboring cells with similar levels of ICG uptake that is in turn superimposed on higher spatial frequency content corresponding to individual cells up to a cutoff frequency corresponding to the row-to-row spacing of RPE cells. Indeed, overlaying the AO-OCT and AO-ICG power spectra [Fig. 4(I)] shows good correspondence between the cutoff frequency of the AO-ICG power spectrum and the bright ring of the AO-OCT power spectrum.
It should be noted that the dark cell centers observed in AO-IRAF and AO-OCT images are cell centers only and are necessarily smaller than the cells observed on AO-ICG in which uniform fluorescence over the entire cell area is shown. As further evidence supporting the observation of a consistent spatial scale observed across modalities, comparison of the radially averaged power spectrum from each modality also shows good agreement between peaks from AO-darkfield, AO-IRAF, and AO-OCT power spectra and the cutoff frequency (sharp change in slope) of the AO-ICG power spectrum corresponding to a cell-to-cell spacing of approximately 14–15 µm assuming hexagonal packing, in agreement with previous studies .
3.3. Multimodal imaging enables consistent measurements of RPE cell-to-cell spacing across the human retina
Expert-validated cell centers identified in each modality were generally found to be distributed in clusters surrounding cell centers identified using all modalities with relatively few outliers. Based on this observation, k-means clustering of identified cell centers across modality was performed and resulted in a set of combined cell centers that localize well with images of RPE cells in each modality (Figs. 5(A)–5(H), Visualization 3) demonstrating that corresponding individual RPE cells can be identified and delineated. Comparison of the initial seed locations of cell centers manually identified through a co-registered image stack of all modalities to cell center locations after clustering reveals relatively small changes in the positions of cell centers in an example dataset [Fig. 5(I) black arrows], demonstrating generally good agreement between the cell centers identified using all modalities and the expert-validated cell centers from individual modalities. Further analysis comparing the clustered center locations to the locations of expert-validated cell centers extracted for each modality revealed low cell center error values of approximately 2–2.5 µm [Fig. 5(J)]. When compared to RPE cells, which range in size from 14–16 µm, this corresponds to a relative cell center error of 14–18%. Kruskal-Wallis testing further revealed no statistically significant difference in cell center error across modalities (p = 0.2).
Analysis across healthy subjects revealed a notable range of RPE cell-to-cell spacing and density values obtained across imaging modality, subject, and retinal location. Cell-to-cell spacing was found to vary by up to 4 µm [Fig. 6(A)] and density by up to 3000 cells/mm2 [Fig. 6(C)] within individual subjects at a given retinal location, likely arising from differences in image interpretation and cell center identification. However, good agreement with previously published normative data  was found for both cell-to-cell spacing [Fig. 6(B)] and density [Fig. 6(D)] values based on clustered cell centers. Chow tests revealed no statistically significant differences in linear regression coefficients between clustered values and normative data for both cell-to-cell spacing (p = 0.6) and density (p = 0.1) metrics. These results further support the use of this clustering approach for the analysis of RPE cell-to-cell spacing and density.
3.4. Evaluation of RPE mosaic disruption using multimodal AO imaging
Images of the RPE mosaic in a patient with choroideremia, a disease known to result in disruptions to the RPE mosaic , were acquired (Fig. 7). To the best of our knowledge, this is the first demonstration of combined multimodal AO-SLO and AO-OCT imaging of the RPE mosaic in a diseased eye.
In this case, RPE cells appear remarkably different compared to those seen in healthy subjects for each of the four modalities. The AO-darkfield image [Fig. 7(A)] shows increased RPE cell contrast compared to AO-darkfield images from healthy subjects at the same retinal location. While residual photoreceptor signal may be expected from images of healthy subjects, this was less noticeable here. Cellular structure can be further visualized in AO-ICG images [Fig. 7(B)] based on varying levels of ICG uptake by RPE cells. Interestingly, in some RPE cells in these images, dark cell centers, presumably cell nuclei , can be observed [white arrows in Fig. 7(B)] which are not usually visible in AO-ICG images from healthy subjects. In AO-IRAF images, RPE cells can be distinguished as dark cell centers surrounded by fluorescent cell boundaries. Here, however, a region of hypofluorescence [Fig. 7(C), green arrow] makes interpretation of the cellular structure in this location challenging. RPE cells as observed in AO-OCT images appear strikingly different compared to those seen in healthy subjects. Bright regions surrounding dark cell centers appear more diffuse and uniformly distributed compared to the bright RPE cell boundaries observed in healthy subjects from AO-OCT images.
Considered together, these co-registered multimodal images enable improved interpretation of the underlying RPE mosaic. While in some regions of these images, RPE cell contrast can be clearly observed across all modalities, other regions are more difficult to interpret based on images from an individual modality. Considered independently, dark regions in the AO-ICG pattern [Fig. 7(B), magenta arrow], AO-IRAF [Fig. 7(C), green arrow], and AO-OCT [Fig. 7(D), green arrow] could potentially be misinterpreted as regions of missing RPE cells. By considering images across all modalities, it is clear that RPE cells are present in these regions. These observations demonstrate the importance of using multiple AO imaging modalities to establish a baseline interpretation of images of disrupted RPE mosaics.
Combining AO-SLO and AO-OCT in a single multimodal imaging system permits direct comparison of several RPE mosaic imaging techniques at the single cell level. This provides complementary views of the RPE mosaic and multiple sources of information to help elucidate the cellular structure. Additionally, the acquired multimodal images serve to validate interpretation of ambiguous areas present on any given modality.
By making direct comparisons of RPE cell imaging techniques, specific advantages and disadvantages of each technique were revealed. Table 1 provides a summary comparison of the four modalities based on the configuration used in this study in healthy subjects including considerations of contrast source, acquisition speed, SNR, relationship between RPE cell contrast and subject or retinal location, and specificity of the contrast source to RPE cells. Overall, side-by-side comparisons of these four modalities revealed that:
- • AO-darkfield imaging, a structural imaging modality based on non-confocal backscattered light, was found to be highly dependent on both subject and retinal location, with improved contrast observed near the fovea compared to eccentric locations where background signal from photoreceptors was found to mask the RPE structure as has been reported previously [Figs. 3(E) and 3(I)] .
- • AO-ICG imaging, based on the heterogeneous pattern formed by ICG uptake from the RPE mosaic, revealed a discernable RPE cell mosaic across all subjects and eccentricities in this study. The contrast of the AO-ICG pattern was observed to be slightly higher near the fovea compared to eccentric locations.
- • AO-IRAF imaging, based on the endogenous fluorescence of melanin [18,22], was observed to have a relatively low SNR compared to other modalities, resulting in many instances where RPE cells could not be clearly identified [green arrows in Fig. 3(C)]. Large variability in RPE cell contrast was observed across collected images, consistent with our earlier report .
- • AO-OCT imaging, based on isolating backscattered light from the RPE layer, revealed the RPE mosaic across all subjects and retinal locations included in this study. In contrast to the other RPE imaging modalities investigated here, the ability to identify and delineate RPE cells in AO-OCT images was qualitatively found to be consistent across the eccentricities investigated here. These observations together with imaging results from previous studies suggest that RPE cell contrast in AO-OCT images does not have a strong dependence on retinal location. However, due to the slower imaging speed of the SD AO-OCT system employed here, this imaging technique is more susceptible to eye motion artifacts [Fig. 3(L)].
In all healthy subjects, the photoreceptor images acquired during all three stages of image acquisition were consistently high quality, suggesting that subject fatigue is unlikely to be a major factor in the observed variations in visibility of RPE cells. Additionally, no notable differences were observed in AO-darkfield images collected across the two AO-SLO imaging sessions (before or after ICG injection). Instead, the source of variability is likely due to modality- and subject-dependent factors.
Notably, the imaging techniques compared here represent only one implementation of a wide range of techniques that have been developed to image the human RPE mosaic in vivo including short wavelength autofluorescence [38,39], transscleral optical imaging , and confocal or non-confocal split detection in regions with missing cones [17,41,42]. The four modalities used in this study were chosen based on their mutual compatibility when implemented in this multimodal system. As has been demonstrated here, the combination of these imaging modalities can provide a more robust interpretation of the RPE mosaic by overcoming limitations of individual imaging modalities including the variation of RPE cell visibility across subjects and retinal locations. This is particularly important for increasing the success rate of cellular-level assessment in patients where the ability to identify and delineate RPE cells is unpredictable due to alterations or disruptions to the RPE mosaic.
In general, when AO imaging is applied to diseased eyes with disruptions to the cellular mosaic, image interpretation and cell contrast does not necessarily follow the trends observed in healthy subjects. It is challenging to predict which modalities will provide useful information regarding the underlying RPE cell structure (Fig. 7). While RPE cell contrast in AO-darkfield images at similar retinal locations tends to be lower in healthy subjects, in the diseased eye imaged here, AO-darkfield provided complementary information not available in the other modalities. Interestingly, a previous AO-based study reported similar observations in a patient with the same disease . In that report, low spatial frequency objects reminiscent of RPE cells were visible along with cone photoreceptors in AO-flood illumination but not in AO-SLO confocal images. In addition to AO-darkfield observations, the appearance of RPE cells in other modalities is distinct from what is expected from imaging in healthy subjects. In AO-ICG images, areas of high contrast between neighboring cells are useful to demarcate the boundary between two distinct cells. In AO-IRAF and AO-OCT images, areas of increased fluorescence are intriguingly colocalized to areas of increased scattering, which suggests that these modalities may have a similar biological source of contrast. RPE cells in AO-OCT images exhibit larger diffuse regions of increased optical scattering surrounding dark cell centers and are remarkably different than what is observed in healthy subjects. This example illustrates the advantage of using a multimodal imaging approach to improve the certainty of RPE cell assessments.
All in all, we demonstrate that assumptions made for assessing RPE cells on individual modalities all lead to consistent quantitative measurements. Nevertheless, there are advantages and disadvantages to every imaging modality and in situations where there are cells within a mosaic that have poor contrast or when the RPE mosaic is disrupted in disease, adopting a multimodal approach can help to provide additional context and validate interpretation of results.
A combined multimodal AO-SLO and AO-OCT imaging platform was developed and used to directly compare techniques for imaging RPE cells in vivo and to obtain multiple views of the RPE mosaic. Images across several retinal locations in healthy subjects show that this multimodal approach helps to improve interpretation and quantification of these cells by combining the strengths of these modalities while simultaneously overcoming inherent limitations that affect individual modalities. Multimodal images acquired of the RPE mosaic from a patient with choroideremia revealed noticeably different RPE cell contrast compared to images from healthy subjects, limiting the ability to assess the cellular structure of the RPE mosaic in individual modalities. However, when analyzed together, RPE cell features in each modality could be combined to better interpret the underlying structure. These results illustrate the benefits of a multimodal approach in a situation where RPE cell contrast or visibility in any one modality is unpredictable. By combining information from all modalities, inferences about normal vs. disrupted cellular structure can be validated, a step towards improving the reliability, accuracy, and overall consistency of in vivo assessment of the human RPE mosaic.
U.S. Food and Drug Administration (Critical Path Initiative); Research to Prevent Blindness; Alcon Research Institute; National Institutes of Health (Intramural Research Program, P30EY026877, R01EY025231, R01EY028287, U01EY025477).
The authors would like to thank B. Brooks, W. Zein, and C. Cukras for assistance with clinical assessment of subjects and helpful clinical discussions, H. Jung for assistance with adaptive optics instrumentation, and D. Cunningham, D. Claus, A. Bamji, W. Holland, B. Gildersleeve, and G. Babilonia-Ayukawa for assistance with clinical assessments and procedures. Schematic components in Fig. 1 were created using ComponentLibrary by Alexander Franzen (http://www.gwoptics.org/ComponentLibrary).
The authors declare that there are no conflicts of interest related to this article. Disclaimer: The mention of commercial products, their sources, or their use in connection with material reported herein is not to be construed as either an actual or implied endorsement of such products by the US Department of Health and Human Services.
1. J. R. Sparrrow, D. Hicks, and C. P. Hamel, “The retinal pigment epithelium in health and disease,” Curr. Mol. Med. 10(9), 802–823 (2010). [CrossRef]
2. O. Strauss, “The retinal pigment epithelium in visual function,” Physiol. Rev. 85(3), 845–881 (2005). [CrossRef]
3. T. Ach, C. Huisingh, G. McGwin Jr, J. D. Messinger, T. Zhang, M. J. Bentley, D. B. Gutierrez, Z. Ablonczy, R. T. Smith, K. R. Sloan, and C. A. Curcio, “Quantitative autofluorescence and cell density maps of the human retinal pigment epithelium,” Invest. Ophthalmol. Visual Sci. 55(8), 4832–4841 (2014). [CrossRef]
4. J. A. Gambril, K. R. Sloan, T. A. Swain, C. Huisingh, A. V. Zarubina, J. D. Messinger, T. Ach, and C. A. Curcio, “Quantifying retinal pigment epithelium dysmorphia and loss of histologic autofluorescence in age-related macular degeneration,” Invest. Ophthalmol. Visual Sci. 60(7), 2481–2493 (2019). [CrossRef]
5. S. Panda-Jonas, J. B. Jonas, and M. Jakobczyk-Zmija, “Retinal pigment epithelial cell count, distribution, and correlations in normal human eyes,” Am. J. Ophthalmol. 121(2), 181–189 (1996). [CrossRef]
6. R. C. Watzke, J. D. Soldevilla, and D. R. Trune, “Morphometric analysis of human retinal pigment epithelium: correlation with age and location,” Curr. Eye Res. 12(2), 133–142 (1993). [CrossRef]
7. T. Ach, E. Tolstik, J. D. Messinger, A. V. Zarubina, R. Heintzmann, and C. A. Curcio, “Lipofuscin redistribution and loss accompanied by cytoskeletal stress in retinal pigment epithelium of eyes with age-related macular degeneration,” Invest. Ophthalmol. Visual Sci. 56(5), 3242–3252 (2015). [CrossRef]
8. I. M. MacDonald, L. Russell, and C.-C. Chan, “Choroideremia: new findings from ocular pathology and review of recent literature,” Surv. Ophthalmol. 54(3), 401–407 (2009). [CrossRef]
9. A. H. Milam, Z.-Y. Li, and R. N. Fariss, “Histopathology of the human retina in retinitis pigmentosa,” Prog. Retinal Eye Res. 17(2), 175–205 (1998). [CrossRef]
10. J. Liang, D. R. Williams, and D. T. Miller, “Supernormal vision and high-resolution retinal imaging through adaptive optics,” J. Opt. Soc. Am. A 14(11), 2884–2892 (1997). [CrossRef]
11. A. Roorda, F. Romero-Borja, W. J. Donnelly Iii, H. Queener, T. J. Hebert, and M. C. W. Campbell, “Adaptive optics scanning laser ophthalmoscopy,” Opt. Express 10(9), 405–412 (2002). [CrossRef]
12. Y. Zhang, J. Rha, R. S. Jonnal, and D. T. Miller, “Adaptive optics parallel spectral domain optical coherence tomography for imaging the living retina,” Opt. Express 13(12), 4792–4811 (2005). [CrossRef]
13. S. A. Burns, A. E. Elsner, K. A. Sapoznik, R. L. Warner, and T. J. Gast, “Adaptive optics imaging of the human retina,” Prog. Retinal Eye Res. 68, 1–30 (2019). [CrossRef]
14. R. S. Jonnal, O. P. Kocaoglu, R. J. Zawadzki, Z. Liu, D. T. Miller, and J. S. Werner, “A review of adaptive optics optical coherence tomography: technical advances, scientific applications, and the future,” Invest. Ophthalmol. Visual Sci. 57(9), OCT51–OCT68 (2016). [CrossRef]
15. D. Scoles, Y. N. Sulai, and A. Dubra, “In vivo dark-field imaging of the retinal pigment epithelium cell mosaic,” Biomed. Opt. Express 4(9), 1710–1723 (2013). [CrossRef]
16. J. Tam, J. Liu, A. Dubra, and R. Fariss, “In vivo imaging of the human retinal pigment epithelial mosaic using adaptive optics enhanced indocyanine green ophthalmoscopy,” Invest. Ophthalmol. Visual Sci. 57(10), 4376–4384 (2016). [CrossRef]
17. H. Jung, T. Liu, J. Liu, L. A. Huryn, and J. Tam, “Combining multimodal adaptive optics imaging and angiography improves visualization of human eyes with cellular-level resolution,” Commun. Biol. 1(1), 189 (2018). [CrossRef]
18. H. Jung, J. Liu, T. Liu, A. George, M. G. Smelkinson, S. Cohen, R. Sharma, O. Schwartz, A. Maminishkis, K. Bharti, C. Cukras, L. A. Huryn, B. P. Brooks, R. Fariss, and J. Tam, “Longitudinal adaptive optics fluorescence microscopy reveals cellular mosaicism in patients,” JCI Insight 4(6), e124904 (2019). [CrossRef]
19. T. Liu, H. Jung, J. Liu, M. Droettboom, and J. Tam, “Noninvasive near infrared autofluorescence imaging of retinal pigment epithelial cells in the human retina using adaptive optics,” Biomed. Opt. Express 8(10), 4348–4360 (2017). [CrossRef]
20. C. E. Granger, Q. Yang, H. Song, K. Saito, K. Nozato, L. R. Latchney, B. T. Leonard, M. M. Chung, D. R. Williams, and E. A. Rossi, “Human retinal pigment epithelium: in vivo cell morphometry, multispectral autofluorescence, and relationship to cone mosaic,” Invest. Ophthalmol. Visual Sci. 59(15), 5705–5716 (2018). [CrossRef]
21. K. Grieve, E. Gofas-Salas, R. D. Ferguson, J. A. Sahel, M. Paques, and E. A. Rossi, “In vivo near-infrared autofluorescence imaging of retinal pigment epithelial cells with 757 nm excitation,” Biomed. Opt. Express 9(12), 5946–5961 (2018). [CrossRef]
22. C. N. Keilhauer, F. o, and C. Delori, “Near-infrared autofluorescence imaging of the fundus: visualization of ocular melanin,” Invest. Ophthalmol. Visual Sci. 47(8), 3556–3564 (2006). [CrossRef]
23. Z. Liu, O. P. Kocaoglu, and D. T. Miller, “3D imaging of retinal pigment epithelial cells in the living human retina,” Invest. Ophthalmol. Visual Sci. 57(9), OCT533 (2016). [CrossRef]
24. Z. Liu, K. Kurokawa, D. X. Hammer, and D. T. Miller, “In vivo measurement of organelle motility in human retinal pigment epithelial cells,” Biomed. Opt. Express 10(8), 4142–4158 (2019). [CrossRef]
25. F. Felberer, J.-S. Kroisamer, B. Baumann, S. Zotter, U. Schmidt-Erfurth, C. K. Hitzenberger, and M. Pircher, “Adaptive optics SLO/OCT for 3D imaging of human photoreceptors in vivo,” Biomed. Opt. Express 5(2), 439–456 (2014). [CrossRef]
26. D. X. Hammer, R. Daniel Ferguson, M. Mujat, A. Patel, E. Plumb, N. Iftimia, T. Y. P. Chui, J. D. Akula, and A. B. Fulton, “Multimodal adaptive optics retinal imager: design and performance,” J. Opt. Soc. Am. A 29(12), 2598–2607 (2012). [CrossRef]
27. Z. Liu, J. Tam, O. Saeedi, and D. X. Hammer, “Trans-retinal cellular imaging with multimodal adaptive optics,” Biomed. Opt. Express 9(9), 4246–4262 (2018). [CrossRef]
28. A. Meadway, C. A. Girkin, and Y. Zhang, “A dual-modal retinal imaging system with adaptive optics,” Opt. Express 21(24), 29792–29807 (2013). [CrossRef]
29. “American National Standard for Safe Use of lasers ANSI Z136. 1-2014,” (American National Standards Institute, 2014).
30. A. Dubra and Y. Sulai, “Reflective afocal broadband adaptive optics scanning ophthalmoscope,” Biomed. Opt. Express 2(6), 1757–1768 (2011). [CrossRef]
31. V. Akondi and A. Dubra, “Accounting for focal shift in the Shack–Hartmann wavefront sensor,” Opt. Lett. 44(17), 4151–4154 (2019). [CrossRef]
32. D. Scoles, Y. N. Sulai, C. S. Langlo, G. A. Fishman, C. A. Curcio, J. Carroll, and A. Dubra, “In vivo imaging of human cone photoreceptor inner segments,” Invest. Ophthalmol. Visual Sci. 55(7), 4244–4251 (2014). [CrossRef]
33. A. Dubra and Z. Harvey, “Registration of 2D images from fast scanning ophthalmic instruments,” in Proceedings of the 4th International Conference on Biomedical Image Registration (Springer-Verlag, Lübeck, 2010), pp. 60–71.
34. O. P. Kocaoglu, R. D. Ferguson, R. S. Jonnal, Z. Liu, Q. Wang, D. X. Hammer, and D. T. Miller, “Adaptive optics optical coherence tomography with dynamic retinal tracking,” Biomed. Opt. Express 5(7), 2262–2284 (2014). [CrossRef]
35. R. F. Cooper, G. K. Aguirre, and J. I. W. Morgan, “Fully automated estimation of spacing and density for retinal mosaics,” Trans. Vis. Sci. Tech. 8(5), 26 (2019). [CrossRef]
36. J. Liu, Y. J. Han, T. Liu, N. Aguilera, and J. Tam, “Spatially aware dense-linkNet based regression improves fluorescent cell detection in adaptive optics ophthalmic images,” IEEE J. Biomed. Health Inform. 24(12), 3520–3528 (2020). [CrossRef]
37. R. F. Cooper, M. A. Wilk, S. Tarima, and J. Carroll, “Evaluating descriptive metrics of the human cone mosaic,” Invest. Ophthalmol. Visual Sci. 57(7), 2992–3001 (2016). [CrossRef]
38. J. I. W. Morgan, A. Dubra, R. Wolfe, W. H. Merigan, and D. R. Williams, “In vivo autofluorescence imaging of the human and macaque retinal pigment epithelial cell mosaic,” Invest. Ophthalmol. Visual Sci. 50(3), 1350–1359 (2009). [CrossRef]
39. E. A. Rossi, P. Rangel-Fonseca, K. Parkins, W. Fischer, L. R. Latchney, M. A. Folwell, D. R. Williams, A. Dubra, and M. M. Chung, “In vivo imaging of retinal pigment epithelium cells in age related macular degeneration,” Biomed. Opt. Express 4(11), 2527–2539 (2013). [CrossRef]
40. T. Laforest, M. Künzi, L. Kowalczuk, D. Carpentras, F. Behar-Cohen, and C. Moser, “Transscleral optical phase imaging of the human retina,” Nat. Photonics 14(7), 439–445 (2020). [CrossRef]
41. A. Roorda, Y. Zhang, and J. L. Duncan, “High-resolution in vivo imaging of the RPE mosaic in eyes with retinal disease,” Invest. Ophthalmol. Visual Sci. 48(5), 2297–2303 (2007). [CrossRef]
42. T. Y. P. Chui, Z. Zhong, and S. A. Burns, “The relationship between peripapillary crescent and axial length: Implications for differential eye growth,” Vision Res. 51(19), 2132–2138 (2011). [CrossRef]
43. R. Syed, S. M. Sundquist, K. Ratnam, S. Zayit-Soudry, Y. Zhang, J. B. Crawford, I. M. MacDonald, P. Godara, J. Rha, J. Carroll, A. Roorda, K. E. Stepien, and J. L. Duncan, “High-resolution images of retinal structure in patients with choroideremia,” Invest. Ophthalmol. Visual Sci. 54(2), 950–961 (2013). [CrossRef]