Images of the full and contiguous photoreceptor mosaic at retinal locations between 5° and 15° temporal to the fovea along the horizontal meridian were recorded in 9 healthy human subjects. Foveal fixation was confirmed by examining the appearance of the cone mosaic while the subject was fixating at the center of the raster scan (data not shown). This eccentricity range spans from a cone-dominated to a rod-dominated location in terms of area, as shown in
below. Note that most of the rod photoreceptor mosaic can be resolved in the logarithmic image, despite the 80% increase in the FWHM of the PSF with respect to the image displayed with a linear grayscale.
Fig. 1 Reflectance images of the human photoreceptor mosaic at three retinal locations along the temporal meridian for subject DLAB_0008, collected using 680 nm light and 0.4 Airy disk pinhole size. The same images are shown with linear (top row) and logarithmic (more ...)
Notably, the transmission images recorded by Packer et al.
and ) resemble the in vivo
images (e.g., ). The bright spots in the histological images correspond to light guided by both rod and cone photoreceptor outer segments. In both our images and those by Packer et al
., the area around the central bright spot of each cone typically appears as a dark ring, which represents the boundary of its inner segment.
Fig. 2 The two images on the left and middle were collected in an excised primate retina, imaged in a bright field microscope in transmission . These images, reproduced with permission from the Journal of Neuroscience, show the cone and rod outer segment (more ...)
The images in
illustrate how averaging multiple registered frames reduces the photon and readout noise that gives single raw frames a grainy appearance. The optimal number of images to average depends on the inherent eye motion, signal-to-noise ratio (SNR) of each particular image series, the image registration algorithm, and the AO correction, among other factors. Three regimes can be observed, depending on the number of images averaged. First, when the number of images is low, assuming perfect image registration and AO correction, the image quality would be limited by readout, photon, or background (i.e., stray light) noise. Second, if the number of averaged images is adequate, and provided the image registration is acceptable, diffraction or residual aberrations would limit the image resolution. Third, when the number of images is too high, the accumulation of imperfectly registered images would blur the average of registered images, and thus the resolution would be limited by the registration algorithm itself. Determining what is an “acceptable” or the “optimal” number of images is by no means trivial, and any attempt to obtain a “rule of thumb” is complicated by tear film disruption, intraocular scattering, residual accommodation, pupil diameter, and other experimental factors. Interestingly, averaging images collected at different times during the day reduces the large cell differences in rod (and cone) reflectivity across the mosaic, as illustrated by the images in . In addition, the uniformity of the reflectivity profile of certain cells also becomes more uniform as a result of averaging images from different time points [46
Fig. 3 Reflectance images of the human photoreceptor mosaic at 10° temporal to the fovea for subject JC_0138, collected using 680 nm light and 1.1 Airy disk pinhole size. From top to bottom the images are: a single frame, a registered average of 50 frames, (more ...)
While the images shown here demonstrate the ability to resolve the rod photoreceptor mosaic, the clinical utility of these images ultimately requires extraction of quantitative information. Shown in
are rod:cone ratio and cell density measurements for all 9 subjects. The estimated values are in reasonable agreement with the histological findings from Curcio et al.
], even at the point of maximum rod density. The small difference between our values and those obtained via histological analysis are likely due to inter-subject variability. Other sources of discrepancy could be errors in determining the retinal location, not accounting for shorter axial lengths at eccentric locations, or even tissue shrinkage artifacts in the histological data. Regardless, these data represent the first in vivo
estimates of rod photoreceptor density in the normal retina. The fact that the rod mosaic can be fully resolved in young people is encouraging in terms of studying advanced stages of eye disease in older patients. While imaging of older subjects would be expected to result in images of lower resolution due to the opacification of the intraocular media and smaller pupil size, the loss and enlargement of rod photoreceptors observed in the aging retina [12
] might partially compensate for this resolution loss.
Fig. 4 Comparison of in vivo rod and cone metrics with those from Curcio et al. . Shown on the left is a plot of the ratio of rods to cones as a function of retinal eccentricity. The solid line is the mean of Curcio’s measurements taken in the temporal (more ...)
With the ability to resolve the complete peripheral photoreceptor mosaic, we sought to investigate the spatial organization of the interleaved rod and cone mosaics. The normal foveal and parafoveal cone mosaic has been well described as being triangular (hexagonally packed) both in histological [47
] and more recently, using in vivo
imaging tools [25
]. Given the normal variation in cone density, these metrics have also been employed to examine disruptions in mosaic regularity caused by retinal disease [21
]. Shown in
is a Voronoi representation of an image of the photoreceptor mosaic taken at 10° temporal to the fovea. Nearly all the cones in the image have more than 8 rods surrounding them (red and dark blue cells in panel b). When examining only the cones, we see that 48% of them have six-sided Voronoi domains (green cells, panel c), indicating that even at this location, the cone mosaic is roughly organized in a triangular lattice. However, this is reduced in comparison to the parafoveal mosaic, which has been reported to have as many as 60%–70% of cones in a given patch of retina having six-sided Voronoi domains [23
]. We find that at this eccentricity, the rods are similarly packed, with 48% of the rods having six-sided Voronoi domains. Of course, since at this location only a minority of the rods do not have a cone as an immediately adjacent neighbor, it is difficult to assess the packing geometry of the isolated rod mosaic without incorporating the disruptive effect of the interleaved cone mosaic. Further work is needed to assess the normal geometry of the rod mosaic in the context of the cone mosaic.
Fig. 5 Analysis of the regularity of the peripheral photoreceptor mosaic. Shown in a is the 6-hour averaged image (logarithmic display) from subject JC_0138, taken at about 10° temporal to fixation, collected using 680 nm light and 1.1 Airy disk pinhole (more ...)
Another well-known feature of the human rod mosaic is the existence of a rod-free zone at the center of the fovea. The resolution of our AOSOs also enables visualization of this anatomical feature of the rod mosaic. Images of the parafoveal photoreceptor mosaic show rod photoreceptors as close as 190 μm away from the foveal center (
). This is consistent with previous ex vivo
data, in which rod photoreceptors were found to first appear between 100 and 200 μm from the foveal center [1
], and also with in vivo
AO imaging of blue cone monochromacy, which enabled direct visualization of both the rod free zone and S-cone free zone of the human fovea [32
Fig. 6 Reflectance image of the human photoreceptor mosaic from subject JC_0138, collected using 680 nm light and 1.1 Airy disk pinhole size and displayed with linear (left) and logarithmic (right) gray scale mappings. The image is a montage of two overlapping (more ...)
Critical to successful visualization of the rod mosaic is correct focusing of the AOSO. The sequence of images in
shows that even though many rods can be resolved over the 35 μm range covered by the sequence, the full mosaic can only be reliably resolved in one of the images (8.7 μm). This suggests that the coarsest focus step used to image the rod mosaic using 680 nm light should be no larger than 10 μm.
Fig. 7 Reflectance images of the human photoreceptor mosaic from subject DLAB_007 at 10° temporal along the horizontal meridian at different retinal depths, shown with linear (top row) and logarithmic (bottom row) gray scales. The image series was collected (more ...)
Across all subjects, we find that the cones and rods do not necessarily appear in focus at the same time, with the rods appearing in focus 10 to 20 μm shallower than the cones. In fact, when the rods are in focus, the cones typically show a complex irregular intensity profile that can mislead both researchers and automated cell counting algorithms into identifying cones as rod clusters. The dark ring characteristic of cone photoreceptors at eccentricities larger than 5°–10° is not always distinct enough to confidently identify each cone as such (see ). It can be seen in the sequence of images in that the complex intensity profiles corresponding to large cone photoreceptors vary with focus, and in fact, more rapidly than the rod photoreceptor mosaic. This suggests that these patterns might be artifactual, but more experiments following the mosaic over time, using light sources with various wavelengths and coherence lengths, as well as pinhole sizes, need to be performed to determine their nature.
From the practical point of view, cones can be disambiguated from rods by recording two images at each retinal location, one with each of the photoreceptor mosaics in focus. While combining images taken over extended time periods also results in improved visualization of the rod and cone mosaics, this technique would be impractical for routine imaging of patients.
Axial intensity profiles of the cone and rod mosaics estimated from image sequences such as the one in (shown in
) indicate that the rods are brightest 10 to 20 μm shallower than the cones at eccentricities greater than 8°. This is consistent with our qualitative observation of rods appearing in focus shallower than the cones. It is also worth noting that despite the inferior axial resolution of the AOSO (~15 μm) compared with current AO optical coherence tomographs (OCTs) (2–3 μm), the axial intensity profiles suggest two peaks that might correspond to both ends of the cone outer segments [27
]. If this were indeed the case, the single peak in the axial intensity profile of the rod mosaic would correspond to the interface between the external limiting membrane (ELM) and the rod inner segments (IS). This, however, would be inconsistent with spectral-domain OCT images that show a very faint ELM/IS boundary with respect to the reflections from both ends of the outer segment at all retinal eccentricities (e.g., [27
].). Further experiments are needed to establish more refined AOSO axial intensity profiles of the cone and rod mosaics at different retinal eccentricities, to clarify the interpretation of the observed axial peaks.
Axial intensity profiles for the cone and rod photoreceptor mosaics shown in . The (very small) error bars correspond to the errors associated with identifying the boundary of the cones in the images
Pinhole sizes between 0.4 and 1.5 Airy disks were used to collect the images for this work in order to explore the 15%–20% lateral resolution benefit that confocal microscopy theory predicts will arise from using sub-Airy disk pinholes [50
]. The data collected for this work does not appear to reflect such increase in resolution, although a more exhaustive study is necessary. A plausible explanation for this inconsistency is that the reflective properties of the photoreceptor mosaic are not accurately described by the multiplicative 2D function implied in the mathematical modeling by Wilson and Carlini [50
Finally, most of the images collected for this study use 680 nm light, although the photoreceptor mosaic can also be resolved with 775 nm light (see
), despite the 14% loss of resolution expected due to the longer wavelength. Image contrast when using 680 nm light for imaging is noticeably higher that when using 775 nm light, but the ability to image the photoreceptor mosaic by using infrared light might prove critical when imaging subjects with increased sensitivity to light. Also, note that photoreceptors that appear brighter/darker when imaged with 680 nm light might not necessarily appear so in the 775 nm image and vice versa.
Reflectance images of the human photoreceptor mosaic collected using 680 and 775 nm light and 1.1 and 1.6 Airy disk pinhole sizes, respectively, shown with linear (left) and logarithmic (right) gray scales. Scale bars are 10 μm across.