Novel application of In Vivo Micro-Optical Coherence Tomography to assess Cornea scarring in an Animal Model

This pilot study uses a micro-optical coherence tomography (micro-OCT) system with ~1 μm axial resolution specifically to image the cornea and corneal scars in vivo. We used an established murine corneal scar model by irregular phototherapeutic keratectomy in ten C57BL/6 mice, with serial imaging using the micro-OCT and compared to anterior segment (AS-OCT) (RTvue, Optovue, Fremont, CA) before and after scar induction. Main outcome was agreement between the AS-OCT and micro-OCT using Bland-Altman plots (95% limits of agreement, LoA).We analysed 10 control eyes and 10 eyes with corneal scars and found that there was good agreement between AS-OCT and micro-OCT (P > 0.05) LOA: lower limit −14 µm (95% CI: −19 to −8.8 µm) upper limit 23 µm (95% CI: 18 to 28.5 µm) in terms of central corneal thickness. There was also good agreement between AS-OCT and micro-OCT in terms of corneal scar measurements (P > 0.5; correlation coefficient >0.99) LOA lower limit −2.1 µm (95% CI: −2.8 to −1.5 µm); upper limit 1.8 µm (95% CI: 1.1 to 2.4 µm). Our pilot study suggests that this novel in vivo micro-OCT imaging technique was able to measure central corneal thickness and scar thickness in agreement with current AS-OCT techniques.

Optical coherence tomography (OCT) uses low-coherence interferometry to enable non-contact, in vivo high-resolution imaging of the cornea 1 . However, most commercial anterior segment OCT (AS-OCT) systems are only able to achieve up to 5 μm axial resolution 2 and thus, not able to differentiate layers within the cornea 3 . We have previously described a micro-OCT that uses broadband light to produce images of the cornea with up to 1-2 μm spatial resolution [4][5][6] . The micro-OCT was able to image corneal endothelial cells 4,5 with a much higher resolution compared to a spectral domain OCT 7 .
Corneal scarring secondary to various insults may cause permanent visual impairment, where corneal transplantation may be the only way to restore vision 8 . As most corneal lesions or scars do not exist throughout the cornea, lamellar replacement of the diseased corneal layers 9 , may provide better tectonic outcomes 10 , or faster visual rehabilitation [11][12][13] . Thus, delineation of the cornea is important to allow for targeted treatment -while real-time AS-OCT guidance has now become possible in our surgical practice 14 . However, current AS-OCT systems have limited resolution, which are only able to detect gross changes to the cornea without detailed imaging of corneal layers, especially in the presence of corneal scars or intrastromal lesions.
Therefore, we performed a pilot study using a superluminescent diode array based micro-OCT system with ~1 μm axial resolution to detect diseased corneal layers, i.e. even in the presence of corneal scarring. Here, we compared this in vivo micro-OCT to a commercially available AS-OCT system in an established animal model to assess normal and scarred corneas.

Results
The analysis was done across all corneal samples (10 control and 10 injured). First, we compared normal corneal thickness measurements between established AS-OCT and micro-OCT to validate that the novel micro-OCT was comparable - Fig. 1. We found that the overall mean at baseline central corneal thickness (CCT) was 94.8 ± 4.0 µm using the AS-OCT and 91.4 ± 6.0 µm using the micro-OCT. There were no significant differences in mean CCT measured under AS-OCT and micro-OCT (Table 1). There was a good agreement between these two imaging techniques (P > 0.05) as the LOA were ranged from a lower limit of −14 µm (95% CI: −19 to −8.8 µm) to upper limit of 23 µm (95% CI: 18 to 28.5 µm) - Fig. 2.
Next, we compared the ability of the micro-OCT to measure the corneal scar thickness as compared to the AS-OCT. The mean scar thickness was 6.9 ± 3.3 µm measured by AS-OCT and 6.9 ± 3.0 µm by micro-OCT wherein no significant difference was observed between both the imaging modalities ( Table 2). The central scar was developed to the greatest thickness at week 2 and subsequently reduced by week 4. There was a good agreement between these two imaging techniques (P > 0.5; correlation coefficient >0.99) with the LOA ranged from a lower limit of −2.1 µm (95% CI: −2.8 to −1.5 µm) to upper limit of 1.8 µm (95% CI: 1.1 to 2.4 µm) - Fig. 3.
The micro-OCT successfully identified the area of corneal scarring as confirmed by corresponding in vivo confocal microscopy imaging (Fig. 4) and histology (Fig. 5). Hematoxylin and eosin histochemistry in scarred corneas at three weeks confirmed corresponding scarring with increased inflammatory cell infiltration in the corneal stroma, and distinct epithelial disorganization. We also used the micro-OCT to identify the areas of active scarring, confirmed by the presence of myofibroblasts and fibroblasts using immunostaining for alpha smooth

Discussion
Our study using a novel in vivo micro-OCT imaging technique suggested that the measurement of central corneal thickness and scar thickness were generally of good agreement with AS-OCT images, with at least 95% of measurements and differences within the 95% prediction intervals and 95% LOA, respectively. We found that the images acquired were able to discern the five basic corneal layers. When imaging the irr-PTK injured eyes,   Table 2. Comparison of in vivo central corneal scar thickness measurements (milimeters) between anterior segment optical coherence tomography and micro-optical coherence tomography. the scar tissue was also detected prominently. For example, the mean central corneal (80.2 ± 11.28 μm) and scar thickness (9.9 ± 2.28 μm) in mice cornea determined by the micro-OCT showed good agreement with immunofluorescent histology measurements of (82 ± 5.5 μm) CCT and (9.2 ± 3.5 μm) scar obtained at Week 2 follow-up. The resolution of distinction of the corneal layers in the processed micro-OCT acquired images, are comparable to the histology results, suggesting the potential for OCT to produce high resolution images of specific corneal layers non-invasively. When comparing the micro-OCT images with the AS-OCT imaging modality, the former performs better than AS-OCT in prominently displaying the scar distribution and intensity. However, it is important to note the limitations of the in vivo micro-OCT used in this pilot study. One of the main problems encountered was the widespread presence of motion artefacts. The reason behind this could be attributed to the heart beat rate (310-840 beats per minute) of the mice used herein, which was much higher than the image acquisition speed of 60 frames per second. In view of this, future outlook for this area of study may involve improving the image acquisition speed. Moreover, the image acquisition for the micro-OCT is longer (estimated 1-2 minutes per image) compared to IVCM (15-30 seconds) and AS-OCT (5 seconds). Nonetheless, we obtained promising results, where this non-invasive imaging technique could produce images with quality comparable to that of ex vivo histology, and with more detail as compared to the current AS-OCT. Further improvements in the image processing technique and image acquisition speed, may allow better results for human eyes 15,16 .
The recent advances in OCT technology now allows, not only higher resolution imaging of the cornea 17 , but also additional information such as en face reconstruction for the cornea and ocular surface 18 , or even delineating vascular flow within the cornea [19][20][21] . High-resolution OCT is currently able to image corneal layers i.e. epithelium, Bowman's layer, stroma, Descemet membrane, endothelium -and even the ocular surface tear film 22 . However, we have described a micro-OCT system that is able to provide more details similar to in vivo microscopy imaging, which may have a potential role in corneal pathologies that specifically require more information such as surgical planning in corneal dystrophies 23 , atypical corneal infections requiring microscopic detail of the pathogen 24 , or assessing the cornea after refractive surgery where high definition stromal layer assessment is required 25,26 .
In summary, despite the recognized limitations of this in vivo micro-OCT system, we provide promising early results from an animal study that suggests high resolution images with histology like detail are obtained in both normal and scarred corneas. Technical improvements such as developing a swept source micro-OCT system with higher A-line speed (up to 4 MHz) 27 , or introduction of a line scan camera with full-range OCT imaging 28 , may improve the image quality even further. If successful, the increased amount of detail in corneal structure analysis could have clinical applications in assessing disease within corneal layers or planning for selective corneal replacement. Further studies for human corneal imaging use in vivo would be required to establish this promising micro-OCT system in the future.

Methods
This study was approved by the Institutional Animal Care and Use Committee of Singapore Health Services and followed the guidelines in the use of Animals in Ophthalmic and Vision Research. We used a murine corneal scar model by irregular phototherapeutic keratectomy as previously described 13 . Ten C57BL/6 mice were anaesthetized by an intraperitoneal injection of ketamine hydrochloride (20 mg/kg body weight) and xylazine hydrochloride (2 mg/kg body weight), with only one eye treated under topical anesthesia (topical 1% lignocaine  hydrochloride). Briefly, a standard 2 mm diameter central corneal wound was made using #64 Beaver blade and denuded of epithelium. Irregular PTK (irrPTK) was then performed to the ablation zone by firing 105 laser pulses (ablation depth ~10 μm) with an excimer laser. A fine mesh screen was positioned in the path of laser after firing 50% of the pulses performed with PTK. Discomfort was relieved by subcutaneous injection of Buprenorphine, 0.05 mg/kg body weight, twice daily for 1 day. Tobramycin ointment was applied twice daily for 2 days to prevent infection. Both injured and fellow control eyes were imaged under micro-OCT, slit lamp biomicroscopy, in vivo confocal microscopy and standard OCT (described in detail below) at day 3 before and 7, 14, 21, 28 days after scar induction. The mice were sacrificed at the 5 th week.
Micro-Optical Coherence Tomography Imaging. The micro-OCT essentially uses light scattered elastically from within tissue in three dimensions and uses the electric field amplitude and the optical delay of light returned from the sample to measure the depth of the tissue. In this study, the micro-OCT system was modified from that described previously 29 , with significant improvements to axial and lateral resolution. The high-bandwidth (~165 nm centered at 830) and short coherence length light, is obtained by use of a superluminescent diode source (T-850, Superlum, Ireland), allowing axial resolution of 2.5 μm in air. The system includes an interferometer while the reference and sample arms intersect at the beam-splitter. A lateral resolution of 2.5 μm (10X objective lens) and 1.3 μm (20X objective lens) is achieved. Effective power incident on sample was kept less than 2 mW. The galvanometer scanning motors were controlled using customizable software, in correspondence with the spectral-data from a scan camera (E2V, AViiVA EM4). The line rate (1024 per frame), frame rate (60 per second) and the scan geometry (3.5 mm × 3.5 mm) were used. Three dimensional images (3.5 mm × 3.5 mm × 1 mm, W × H × D) were obtained with an acquisition time of less than 3 minutes per 3D image.

Optical Coherence Tomography Image Enhancement and Data collection. Current image qual-
ity from standard OCT is greatly hampered by the presence of shadow artifacts and poor tissue visibility in the deeper layers -due to signal attenuation, whereby signal strength diminishes as a function of tissue depth. Moreover, the animal model for this experiment, mice, have a very high heart rate of 310-840 beats per minute, which being much higher than the frame rate results in noise and distortion in the captured image. Filters were applied to remove the background noise and edge distortions from the images using MATLAB R2017a (The MathWorks, Inc., Natick, Massachusetts, United States). Further, the captured micro-OCT images were improved using compensation algorithms to compensate for light attenuation and contrast enhancement with post-processing software 15 , used to remove the shadow artefacts from dense structures and improve the visibility of the deep tissue layers, as previously described - Fig. 1 16 . The post-processed images were then used to obtain information on the corneal, epithelial and the scar thickness (in case of the injured samples). All measurements were done using ImageJ and each measurement was computed as the average of 10 measurements taken across the sample.
Anterior segment evaluation and histology. We performed serial slit-lamp microscopy (FS-3V Zoom Photo Slit Lamp, Nikon, Tokyo, Japan) and AS-OCT (RTvue, Optovue, Fremont, CA) to compare corneal measurements between the micro-OCT and AS-OCT. In vivo confocal microscopy (Heidelberg Engineering GmbH, Heidelberg, Germany) was also performed for corneal scars in vivo. After sacrifice, the murine corneas were extracted and fixed in 10% formaldehyde (Sigma Aldrich, St. Louis, MO, USA) for 60 min and PBS washes. They were dehydrated and processed for paraffin embedding, with sections (5 μm thick), haematoxylin and eosin (H&E) staining -using Mayer's hematoxylin (Sigma Aldrich, USA), then differentiated with acid alcohol, extensively washed before stain with Eosin solution (Sigma Aldrich, USA) for another 2 minutes, washed, and dehydrated. The section was mounted with Permount reagent (Fisher) and viewed under light microscopy (Nikon C2 confocal microscope). Immunofluorescence staining for scar tissue markers was performed using Mouse On Mouse (MOM) Immunodetection kit (Vector Laboratories Inc., Burlingame, CA, US) with mouse monoclonal antibody against alpha smooth muscle actin (α-SMA) (DAKO M0851; 1:200 dilution) and mouse monoclonal antibody against fibronectin (FN) (Millipore MAB1940; 1:200 dilution; Temecula CA, US). After staining, the samples were mounted with Vectashield containing DAPI (Vector Laboratories) and viewed under fluorescence microscopy (ZEISS Axio Observer.Z1, Oberkochen, Germany).
Statistical analysis. Mean differences in corneal and scar measurements (mean ± standard deviation, SD) between the two machines were assessed using a paired t-test, after tests for normality. Agreements between the AS-OCT and µ-OCT machines were also described using the Bland-Altman, with 95% limits of agreement (LoA). Bland Altman analysis was performed with MedCalc Version 17.1. Intra-Class Correlation Coefficients were computed for inter-machine reliability of data. All other statistical tests were performed with IBM Corp. Released 2011 IBM SPSS Statistics for Windows, Version 20.0. Armonk, NY: IBM Corp.; Statistical significance was considered p < 0.05. Data availability. The data that support the findings of this study are available from the corresponding author.
(J) Injured cornea showing the area of scar with increased inflammatory cell infiltration and epithelial disorganization which corresponded to the area of scar detected by the micro-OCT. (K,L) Examples of slit-lamp photographs showing normal (K) and scarred (L) corneal, slit-beam to demonstrate the scars are mainly in the anterior stromal aspects.