250 kHz, 1.5 µm resolution SD-OCT for in-vivo cellular imaging of the human cornea

: We present the first spectral domain optical coherence tomography (SD-OCT) system that combines an isotropic imaging resolution of ~1.5 µm in biological tissue with a 250 kHz image acquisition rate, for in vivo non-contact, volumetric imaging of the cellular structure of the human cornea. OCT images of the healthy human cornea acquired with this system reveal the cellular structure of the corneal epithelium, cellular debris and mucin clusters in the tear film, the shape, size and spatial distribution of the sub-basal corneal nerves and keratocytes in the corneal stroma, as well as reflections from endothelial nuclei. The corneal images presented here demonstrate the potential clinical value of the new high speed, high resolution OCT system for non-invasive diagnostics and monitoring the treatment of corneal diseases.

pain, and increase the risk of infections and abrasions; b) limited field of view (typically ~400 µm × 400 µm), therefore, imaging larger areas of the cornea requires acquisition and subsequent "stitching" of multiple images, thus increasing significantly the image acquisition and processing time; and c) depth location ambiguity due to poor axial resolution and slow scanning in axial direction relative to the fast axial eye motion.
Optical coherence tomography (OCT) is a non-invasive optical imaging method capable of generating cross-sectional and volumetric images of biological tissue with cellular level resolution [9,10]. Over the past 25 years, OCT has found a wide range of biomedical applications [10], including imaging of the human cornea in health and disease [11,12]. Recent developments of broad-bandwidth light sources and high speed, large pixel number cameras resulted in development of spectral domain (SD) and full field (FF) ultrahigh resolution OCT (UHR-OCT) technology with axial resolution close to or below 1 µm, suitable for imaging the cellular and sub-cellular structure of biological tissue [13][14][15][16][17]. When used for corneal imaging, UHR-OCT is able to visualize in-vivo, identify and in some occasions even measure the thickness of some of the major corneal layers such as the epithelium (EPI), the Bowman's membrane (BM), the stroma (STR) and the Descemet's membrane (DM) the DM-endothelial complex (DEC) [17][18][19][20][21]. It is also able to image exvivo [22] and in-vivo [19,21] and the cellular structure of different corneal layers, visualize invivo [21] and count [23] keratocyte cells in the corneal stroma, measure in-vivo physiological changes in the cornea [24], assess keratoconus induced damage to the human cornea at cellular level [19,25], as well as quantify the tear film thickness and track tear film dynamics [26]. Mazlin et.al [21]. utilized a novel FF-OCT image in-vivo healthy human cornea and were able to image keratocytes and nerves in the corneal stroma, as well as endothelial cells. However, due to the limited axial resolution and the susceptibility of the FF-OCT technology to strong reflections arising from scattering of the imaging beam from the collagen structure of the corneal BM, they were not able to image the tear film and the cellular structure of the corneal epithelium. Recently, our research group reported the use of a sub-micrometer axial resolution SD-OCT system to visualize in-vivo for the first time the corneal pre-Descemet's layer (PDL, also referred to as the 'Dua" layer in some publications), which in the past has been identified and measured only in ex-vivo human corneal preparations [27]. The Descemet's membrane and the corneal endothelium were also clearly identified in crosssectional UHR-OCT images in that study, and the thickness of all 3 posterior corneal layers was measured in-vivo in healthy subjects. Furthermore, our group utilized the same UHR-OCT technology to image the cellular structure of the corneal epithelium in healthy and KC corneas, to assess the damage to the corneal stroma in subjects with mild to advanced stages of KC [19], as well as generate in-vivo and without tissue contact, volumetric images of the limbal crypts and Palisades of Vogt in the healthy human limbus [28]. However, due to the fairly slow camera readout rate (34 kHz) of the UHR-OCT technology used for these studies [19,27,28], it was difficult to generate volumetric images of the cellular structure of the human cornea and limbus without imaging artefacts generated by the fast eye motion.
To image the cellular structure of the human cornea with UHR-OCT in clinical environment in-vivo, volumetrically, over a wide filed-of-view and with minimal effect of eye motion related image artefacts, it is important for the OCT technology to provide both micrometer scale spatial resolution and high image acquisition rates. Swept source OCT technology (SS-OCT) offers impressive A-scan rates in the order of MHz [29]. However, the axial resolution of SS-OCT systems is typically > 5 µm in biological tissue, due to the limited spectral range of the tunable light sources available for the 800 nm and 1060 nm spectral regions. Spectral domain OCT (SD-OCT) systems with close to 1 µm axial resolution in biological tissue and A-scan rates of up to 70 kHz have been reported for various biomedical applications including corneal imaging [15][16][17]20]. Furthermore, high speed CMOS cameras have also been utilized in SD-OCT systems to generate A-scan rates of up to ~500 kHz [30][31][32][33]. However, the axial OCT resolution in those cases was in the range of 3 µm to 8  (achromat doublet, f = 10 mm, Thorlabs, USA), a pair of galvanometric scanners (Cambridge Technologies, USA), a beam expander (achromat doublets, f = 40 mm and f = 80 mm, Thorlabs, USA) and a 10×/0.26 NIR corrected microscope objective. Fiberoptic polarization controllers are used to optimize the shape and amplitude of the system's point-spreadfunction (PSF). The detection end of the UHR-OCT system is comprised of a customized, commercially available spectrometer (Cobra-S 800, Wasatch Photonics, Durham, USA) integrated with a 2048-pixel, monochrome, line-scan CMOS camera (OCTOPLUS CL, e2v, Teledyne Dalsa, Canada). The camera offers a tall pixel design (10 µm x 200 µm) and a maximum readout rate of 250 kHz. A frame grabber (X64 Xcelera-CL + PX8 Full, Teledyne Dalsa, Canada) is used to acquire images at the maximum camera rate. A custom LabVIEW code was developed for operation of the UHR-OCT system. For in-vivo imaging of the human cornea, the optical power of the imaging beam was set to ~800 µW, significantly lower than the maximum permissible exposure power as specified by the ANSI (American National Standards Institute) standard [34].

Imaging procedure
This study was approved by the Research Ethics Committee at the University of Waterloo and was carried out in compliance with the tenets of the Declaration of Helsinki. Healthy subjects (n = 10), aged 20 to 45 years, were recruited for this study. All subjects passed a slit-lamp biomicroscopy screening and provided written consent for participation in the study. During the OCT imaging sessions, a head band and a fixation target were used to minimize the subject's head and eye motion. Cross-sectional (1000 A-scans × 1024 pixels) and volumetric (750 A-scans × 750 B-scans x 1024 pixels) images were acquired from ~0.75 mm × 0.75 mm regions of the cornea, located slightly inferiorly with respect to the corneal apex in order to avoid direct reflection artefacts at the corneal apex. The acquisition time for a volumetric OCT image of 750 A-scans × 750 B-scans was ~2.8 s, taking into account the 80/20 raster scan ratio. Because of the limited depth-of-focus of the microscope objective, the OCT imaging probe was adjusted manually in axial direction to a specific depth location of the cornea and the reference path-length was adjusted accordingly. Multiple volumetric OCT images were acquired from different depth locations in the cornea from each of the subjects.

Image processing
Images were generated from the raw OCT data and numerically dispersion compensated up to the 5 th order with a custom MATLAB algorithm. A cross-correlation algorithm was applied to the 3D OCT images to compensate for eye motion. No additional image post-processing was used for the UHR-OCT images presented in this paper. The volumetric and enface images were generated from the 3D data sets with Amira (Amira Inc.)

In-vivo confocal microscopy (IVCM) and histology
IVCM images of the cornea were acquired from healthy subjects with the Heidelberg Retinal Tomograph III (HRT III), using the Rostock Cornea Module (Heidelberg, Germany). The size of the imaged area in the cornea was 400 µm × 400 µm. The instrument was set up according to standard techniques outlined in the manufacturer's operation manual and described elsewhere [35,36]. A high viscosity gel, Tear-Gel, was applied to the front surface of the microscope lens, prior to mounting a new, sterile TomoCap. One drop of a topical anesthetic (0.5% proparacaine hydrochloride) was instilled into the participant's eye prior to the imaging procedure. The HRT III system was aligned so that the TomoCap made slight contact with the cornea and the imaging probe was translated in axial direction to focus the imaging beam at different axial positions inside the cornea.
Histological images were also acquired using the following protocol. Healthy corneas were harvested postmortem and after initial fixation in 10% neutral buffered formalin, the tissue was embedded in paraffin, serially sectioned into 5 µm thick sections, and stained with hematoxylin paraffin secti fibrillary acid nerves. The DM1000, ICC

System's
The performa resolution an normalized re detection end overlap betwe nm, thus corre PSF was mea shown in Fig compensation PSF in Fig. 2 using a custom space, which index of n = 1 of the refracti Results fro Fig. 2 The lateral resolution of the UHR-OCT system was measured using a standard United States Air Force resolution target (USAF, Thorlabs, USA). An image of the USAF target acquired at 250 kHz is shown in Fig. 2(D). A magnified view of the central region marked with the blue square in Fig. 2(D) is shown in Fig. 2(E). The line pairs of groups 6 and 7 or Element 7 are clearly resolved. The red and green lines in Fig. 2(E) mark the locations from which the intensity profiles shown in Fig. 2(F) were extracted. These results suggest that the lateral resolution measured in free space is < 2 µm. Figure 3(A) shows a representative cross-sectional H&E stained histological image of the anterior healthy human cornea. The image shows the cellular structure of the corneal epithelium (EPI), as well as the Bowman's membrane (BM) and keratocytes in the corneal stroma (STR) marked with red arrows. Figure 3(B) shows a representative cross-sectional UHR-OCT image, that was acquired in-vivo from the anterior cornea of a healthy subject at the maximum camera readout rate of 250 kHz. The cellular structure of the corneal EPI is clearly visible on the UHR-OCT B-scan. Reflective white dots are also observed inside the cells, most likely corresponding to reflections from the cellular nuclei. The interface between the corneal epithelium and the Bowman's membrane appears as a thin, highly reflective, almost solid white line, while the interface between the BM and the stroma is only marked by the first, most anterior layer of keratocyte cells, marked with red arrows. The high axial resolution (1.5 µm physical resolution), also allows for visualization of the thin tear film layer ( Fig. 3(B), blue arrow) and measurement of its thickness, which in this particular case was ~4 µm. A volumetric image of the corneal epithelium is shown in Fig. 3(C). In addition to the clearly visible cellular morphology of the EPI layer, the image also shows hyper-reflective (white) structures within the tear film, marked with yellow arrows. An enface view of the tear film surface is shown in Fig. 3(D). Some of the reflective structures have more or less round shape of different dimensions and they, most likely, correspond to clumps of cellular debris. Other reflective structures appear as thin, occasionally branching lines of various length. We hypothesize that those structures may correspond to clusters of mucins that are present in the tear film. Figures 3(E)-3(G) show enface UHR-OCT images of the corneal epithelium that were acquired at different depths within the EPI layer. The image in Fig. 3(G) shows the cellular structure of the basal cell layer of the EPI. Reflective white dots are visible inside the cells and most likely they correspond to reflections from cellular nuclei. Figure 3(H) shows a larger field of view (~250 µm × 250 µm) enface UHR-OCT image of the basal cell layer of the corneal epithelium, which correlates well with an IVCM image acquired at a similar location and in the cornea and with similar magnification.

Cornea
The high sp acquisition ra representative shows the sam basal cell laye sub-basal corn interface betw stained histol brown in the i view, enface IVCM respec one, it may be 3 Figure 6 sho UHR-OCT im beam was foc contrast in th region of inte that ROI (5x Fig. 6  keratocytes in dle stroma are tion UHR-OC and 5(E) respe rved in these U F)) acquired fr ble in both the e posterior cor 6(A). Since in part of the corn s of the 10x m ine rectangle a n Fig. 6    OCT (FF-OCT) system, designed for in-vivo non-contact, volumetric imaging of the cellular structure of the human cornea [21]. One major advantage of the FF-OCT technology is its ability to acquire larger field-of-view, cellular resolution images of the cornea that are free of eye motion artefacts. This is due to the fact that FF-OCT generates fast enface 2D images and scans slower in axial direction. By using FF-OCT, Mazlin et.al., were able to acquire impressive images of keratocytes and nerves in the corneal stroma, and of the honeycomb arrangement of the corneal endothelial cells. However, due to the limited axial resolution and the susceptibility of the FF-OCT technology to strong reflections arising from scattering of the imaging beam from the collagen structure of the corneal BM, they were unable to image the tear film and the cellular structure of the corneal epithelium. The high spatial resolution offered by the SD-OCT system described here, allows for detection of reflections from endothelial cell nuclei, as shown in Fig. 6(E). The shape of the nuclei does not appear spherical in this image due to presence of optical aberrations that have not been compensated. Note that Fig. 6(E) shows a single-plane enface image, acquired at a depth location corresponding to about half way through the thickness of the corneal endothelial layer (marked with a blue arrow in Fig. 6(B)). For this reason, this image does not show the typical hexagonally-shaped endothelial cells as seen in the IVCM image ( Fig. 6(D)), or as generated by FF-OCT [21]. Both the IVCM and the FF-OCT [21] images are generated with axial resolution of ~10 µm and ~8 µm respectively, which is ~2x larger than the thickness of the corneal endothelial cells. Therefore, IVCM and FF-OCT integrate the light scattered from the nuclei and other cellular organelles, as well as some light scattered from the collagen fibrilae in the Descemet's membrane. Thus, these imaging technologies are able to generate enface images that show the hexagonal shape of the corneal endothelial cells. As shown in Fig. 6(G), SD-OCT technology with 1 µm axial resolution is also able to generate images of the corneal endothelial cells of similar appearance, by utilizing maximum intensity projection over the entire thickness of the endothelial layer. Note that the image in Fig. 6(G) was acquired ex-vivo from a post-mortem corneal tissue sample. Therefore, the lack of eye motion artefacts allowed for generation of high resolution, high contrast images of the corneal endothelial cells.

Poster
In general, similar maximum projection approach can be applied to SD-OCT images of the corneal endothelium acquired in-vivo. Figure 7(A) shows a maximum intensity projection image of the healthy corneal endothelium over the entire thickness of the layer, that was acquired in-vivo with the 250 kHz OCT system. A small area in the image marked with the red rectangle shows hexagonally shaped endothelial cells. A magnified view of that area is presented in Fig. 7(B). Lateral saccades are clearly visible in the enface image (7A). B-scans ( Fig. 7(C) and 7(D)) obtained from different locations of the 3D imaging stack (marked with the blue and orange arrows in Fig. 7(A)), demonstrate the loss of contrast due defocus resulting from fast axial eye motion. I this case, the combination of high spatial resolution (1.5 µm), limited depth-of-focus (~20 µm) and limited camera speed (250 kHz) offered by the current design of the SD-OCT system, limited severely the area over which hexagonal endothelial cells can be observed. Future improvements of the optical design of the SD-OCT system, or the use of faster cameras, could make this task feasible.
One interesting fact of note is that the nuclei of corneal endothelial cells appear as low reflective dots on a highly reflective background of the cell body in IVCM (Fig. 6(D)) and FF-OCT images. However, the same nuclei appear as highly reflective dots on a low reflective background in the OCT enface images generated by the 250 kHz SD-OCT ( Fig.  6(E)). Some factors that may contribute to the apparent inversion of the image contrast are: a) the 3D shape of the light scattering probability profile, which is dependent on the size, shape and refractive index of the cellular nuclei and other organelles, relative to the wavelength range of the optical imaging beam; b) the numerical aperture and the spatial resolution of the respective imaging system.  The micrometer scale axial resolution of the proposed high speed UHR-OCT system will also permit more precise thickness measurement of the thin layers of the human cornea, such as the BM and Descemet's membranes and the corneal endothelium, since the precision of the layer thickness assessment is strongly dependent on the OCT axial resolution. This advantage of high axial resolution OCT has already been demonstrated by other research groups that used OCT technology with ~1.2 µm axial resolution in tissue to measure the thickness of the corneal epithelium and BM [17], and the tear film [26]. Our research group used a 0.95 µm axial resolution SD-OCT system to image in-vivo and quantify with high precision the thickness of the corneal endothelial layer, which is only ~4 µm thick, as well as the Descemet's membrane and the pre-Descemet's layer in the posterior healthy human cornea [27].
Although the current optical design of the fast scanning UHR-OCT system allows for invivo imaging of the cellular and sub-cellular structure of biological tissues, it has some limitations that offer opportunity for future optimization of the system's performance. For example, the 250 kHz data acquisition rate certainly helps with suppressing significantly eye motion related artefacts. However, as shown in the images presented in Figs. 3(H), 4(A), 4(B) and 4(D), fast axial motion in the range of 10 µm to 100 µm, as well as fast lateral saccades, can introduce image artefacts in the volumetric UHR-OCT images even in the case when those are acquired from relatively small areas of the cornea (250 µm x 250 µm) with a total image acquisition time of ~300 ms. One approach to minimizing the effect of axial eye motion artefacts in the volumetric UHR-OCT images is to extend the depth-of-focus, for example by redesigning the core of the interferometer to incorporate axicon lenses [37]. Significant suppression or elimination of OCT image artefacts arising from fast lateral saccades in the eye, would require either increasing the camera readout rate in a SD-OCT design of a system, introducing eye tracking in the system, or the use of FF-OCT or line scanning (LS-OCT) [38] designs of the OCT technology. All of these approaches offer both advantages and limitations: increasing the camera readout rate will result in reduction of the SNR; parallel scanning OCT technology such as FF-OCT is highly susceptible to directly reflected or strongly scattered light in biological tissue, which would prevent imaging of certain regions or layers of the tissue; the optical design of LS-OCT, the number of camera pixels, as well as the cost and availability of ultrahigh speed area cameras with sufficient number of camera pixels, will limit the lateral and axial resolution, the scanning range and the SNR roll-off of the LS-OCT technology and as such affect adversely the affordability of such system.
A simple and fairly low-cost modification of the imaging probe of the current high speed UHR-OCT system would allow its use for non-invasive cellular resolution imaging of the human retina. In that case, the use of hardware or software adaptive optics will be required to visualize in-vivo individual retinal cells, as the lateral resolution of the unprocessed OCT retinal images is mainly determined by the optics of the eye and the diameter of the optical imaging beam. In the past, both hardware and computational adaptive optics (AO-OCT) [39] have been applied successfully to in-vivo retinal images and in those cases allowed for visualization of individual photoreceptors. Imaging of other parts of the human body with the same UHR-OCT system are also feasible, however, the optical density and light scattering of the imaged tissue will pose a limit to the penetration depth at which individual cells can be resolved in the OCT images.
In conclusion, we have developed a high speed (250 kHz) UHR-OCT system that offers theoretical isotropic imaging resolution of ~1.5 µm in corneal tissue. Volumetric images acquired in-vivo with the system from different depth locations in the healthy human cornea, demonstrate the ability of the system to image without contact with the corneal tissue, the cellular structure of the corneal epithelium, stroma and endothelium. The corneal images presented here demonstrate the potential clinical value of the new high speed, UHR-OCT system for non-invasive diagnostics and monitoring the treatment of corneal diseases.

Funding
Canadian Institutes of Health Research (CIHR, 446387); Natural Sciences and Engineering Research Council of Canada (NSERC, 312037 and 446387); the University of Waterloo Research Incentive Fund.