Reliability and efficiency of corneal thickness measurements using sterile donor tomography in the eye bank

To evaluate the reliability and efficiency of sterile pachymetric measurements of donor corneas based on tomographic data using two different methods: a “manual” and a “(semi-)automated” method. Twenty-five (25) donor corneas (50%) stored in MI and 25 (50%) in MII were imaged 5 times consecutively using an anterior segment OCT (AS-OCT). The central corneal thickness (CCT) was measured both with the manual measurement tool of the AS-OCT (= CCTm) and with a MATLAB self-programmed software allowing (semi-)automated analysis (= CCTa). We analyzed the reliability of CCTm and CCTa using Cronbach´s alpha (α) and Wilcoxon signed-Rank Test. Concerning CCTm, 68 measurements (54.4%) in MI and 46 (36.8%) in MII presented distortions in the imaged 3D-volumes and were discarded. Concerning CCTa, 5 (4%) in MI and 1 (0.8%) in MII were not analyzable. The mean (± SD) CCTm was 1129 ± 6.8 in MI and 820 ± 5.1 µm in MII. The mean CCTa was 1149 ± 2.7 and 811 ± 2.4 µm, respectively. Both methods showed a high reliability with a Cronbach´s α for CCTm of 1.0 (MI/MII) and for CCTa of 0.99 (MI) and 1.0 (MII). Nevertheless, the mean SD of the 5 measurements was significantly higher for CCTm compared to CCTa in MI (p = 0.03), but not in MII (p = 0.92). Sterile donor tomography proves to be highly reliable for assessment of CCT with both methods. However, due to frequent distortions regarding the manual method, the (semi-)automated method is more efficient and should be preferred.

increased over time (EEBA 2020). According to the German keratoplasty register, 9173 keratoplasties were performed in 2019 in Germany (Flockerzi et al. 2018). The surgical technique has continuously been improved and adapted over the years in order to respond to new challenges. Those improvements of the ''original'' penetrating keratoplasty  included new suture techniques e.g. the double running cross-stitch suture according to Hoffmann (Hoffmann 1976;Suffo et al. 2021), new trephination techniques, e.g. with excimer laser (Seitz et al. 1999), and new lamellar techniques, e.g. the Descemet Membrane Endothelial Keratoplasty (DMEK) (Melles 2006;Seitz et al. 2020) or the Deep Anterior Lamellar Keratoplasty (DALK) (Luengo-Gimeno et al. 2011), optionally assisted with excimer laser . Other improvements related to corneal grafting are much less recognized or discussed in ophthalmology, such as modern imaging techniques in eye banking, where-in Europe-the donor corneoscleral discs (CD) are prepared and preserved, mostly using organ culture (EEBA 2020). Since the introduction of organ culture in eye banks in 1973 (Summerlin et al. 1973;Doughman et al. 1976), many innovations have been made in terms of improving storage and selection of CD before keratoplasty (Batista et al. 2018). One milestone is the sterile donor tomography which allows easy and sterile tomographic analysis of the CD for clinical decision support in the eye bank before keratoplasty. The possibility of qualifying CD in eye banks using topo-or tomographs had already been considered since 1999 (Terry et al. 1999) but has made significant progress in recent years. Nowadays, donor tomography can be used in clinical routine as a screening method prior to keratoplasty Quintin et al. 2021a) and has already been used, amongst others, to tailor out the best timepoint for changing from MI to MII before penetrating keratoplasty (Wolf et al. 2009;Hamon et al. 2021) to improve the IOL power calculation in classical triple procedure (Quintin et al. 2021b) and to ''harmonize'' the donor and recipient tomography (Mäurer et al. 2020). However, the corneal tomographs were not developed for this purpose and other challenges when measuring CD under sterile conditions are the optical distortions of the culture container and image deterioration with culture media.
The purpose of this study was to investigate the reliability and efficiency (time and resources required for imaging and analyzing the cornea) of the ''manual'' and ''(semi-)automated'' measurement of donor central corneal thickness (CCT) through the plastic cell culture flask and through preservative organ culture medium I (MI) or transport/deswelling organ culture medium II (MII) using sterile non-contact donor tomography in the eye bank.

Materials and methods
No ethical approval was required according to the Ethics Committee of Saarland, Germany. The study was conducted in accordance with the Declaration of Helsinki and surrogate consent for the scientific use of donated corneas was obtained.
A ''compromise analysis'' was performed with the Software G*Power (Erdfelder, Faul, Buchner, HHU Düsseldorf, Germany) in order to assess the power (1-ß) and the probability of first order error (a) of the statistical analysis for a total sample of 50 corneas (Faul et al. 2009). Statistical analysis were performed using a Wilcoxon signed-rank test; a power (1-ß) of 0.94 and a probability of error (a) of 0.06 were assessed, with effect size (d) according to Cohen of 0.5 (medium effect size) (correspond to 0.5 pooled SD between the means of both groups).
The CD were imaged with an anterior segment optical coherence tomograph (AS-OCT) CASIA 2 (Tomey, Nagoya, Japan) according to the method developed by Mäurer, Damian and Langenbucher (Janunts et al. 2016;Damian et al. 2017;Mäurer et al. 2019) under sterile conditions through their cell culture flask (Primaria 25 cm 2 canted-neck cell culture flask, Corning Inc., Corning, NY, USA). The culture flasks were maintained on a 3D-printed plastic holder adapted to the chin rest of the AS-OCT (Fig. 1a). The OCT system was aligned to measure the CD from the endothelial side, whereby a central measuring range of approximately 7 mm is achieved, limited by the tissue holder. Out of the initial raster scan of 12 9 12 mm (64 slices, 4 9 repeat), subsequent 3-dimensional (3D) volume data were generated with an axial/lateral resolution of 5.6/6 lm/voxel, respectively (Janunts et al. 2016). Each CD was imaged 5 times consecutively by the same examiner (ophthalmologist working in the eye bank and trained to perform the sterile imaging) in MI or MII.
The central corneal thickness (CCT) was assessed using two alternative methods. Firstly, using the manual measurement tool of the AS-OCT (CASIA 2 Software, version 3G.3, Tomey, Nagoya, Japan) with a modified method inspired by that of Schnitzler et al. (2016). The ''manually measured'' CCT (CCTm) was determined by the corneal thickness at the vertical bisection of the horizontal line between both shaded artefacts caused by the holder (Fig. 1b). All CCTm were measured by the same examiner: a physician specialized in eye banking and trained for ''sterile donor tomography''. Secondly, the (semi-)automated CCT (CCTa) was assessed with a MATLAB (The MathWorks Inc., Natick, Massachusetts, USA) custom-made program developed by Mäurer et al. (2019). This software was validated using a polymethyl methacrylate (PMMA) phantom cornea based on Gullstrand's proportions. While processing the raw data from the AS-OCT, the program eliminates artefacts such as the cell culture flask wall and cornea holder and fits the spherocylindrical model to the corneal front and back surface after correction of image distortion. The corneal thickness is extracted from the model data of both corneal surfaces (Mäurer et al. 2019).
To investigate the reliability of the measurements, Cronbach's alpha (a) was calculated for the sequence of 5 measurements, separately for each culture medium (MI and MII) and for each pachymetry method (CCTm and CCTa) (Cronbach 1951). For clinical purposes, a Cronbach's alpha exceeding 0.9 indicates proper reliability (Nunnally et al. 1993).
We calculated the mean of standard deviations (SD) of the 5 subsequent CCTm and CCTa measurements to determine the repeatability. Differences in intermethod agreements were calculated using linear regressions and a Wilcoxon signed-rank test. The results were presented in Bland-Altman plots. In addition, we analyzed the effect of the storage time in MI or MII (Time in medium-TiM) on the CCT using Pearson's correlation tests.
Statistical analysis was performed with SPSS Version 20.0.0 for Windows (SPSS Inc., Chicago, IL, USA). Values are expressed as mean ± SD (minimum -maximum).

Results
The CCTm was 1129 ± 6.8 lm (710-1923 lm) in MI and 820 ± 5.1 lm (621-1493 lm) in MII. Out of the total of 125 AS-OCT images in each medium, 68 (54.4%) ''manual'' measurements in MI and 46 (36.8%) in MII had to be discarded due to distortions in the corneal volume generated by the AS-OCT Software (Fig. 2). None of the 5 images were even analyzable for 2 corneas (4%) for MI and 1 cornea (2%) for MII. The CCTa was 1149 ± 2.6 lm (730-1705 lm) in MI and 811 ± 4.4 lm (634-1057 lm) in MII. For 5 cases in MI (4%) and 1 case in MII (0.8%), the MATLAB program could not extract the edges from the volume data and, therefore, CCTa was not available. All individual measurements are shown in Tables 1 and 2. Based on available measurements, Cronbachs a showed a very high reliability for CCTm in MI and MII with a value of a = 1.0. The CCTa showed a very high reliability in MI (a = 1.0) and in MII (a = 0.99).
The level of agreement between CCTm and CCTa measurements were compared using linear regression and Bland-Altman plots (Fig. 3). In both culture media (MI and MII), CCTa and CCTm did not differ significantly for MI (mean difference between the measurements of 5 ± 76 lm, significant level of agreement, p = 0.006) and for MII (mean difference between the measurements of 18 ± 83 lm, significant level of agreement, p = 0.002). A Wilcoxon signedrank test comparing the mean SD of both methods showed no significant difference for CD imaged in MI (respectively 6.8 lm and 2.6 lm, p = 0.09), but showed a significant larger value for SD with CCTm (5.1 lm) in comparison to SD with CCTa (2.4 lm) (p = 0.03) in MII, indicating a better repeatability of measures conducted (semi-)automatically in comparison to manually in MII.
The CD storage time in medium I (TiM-MI) or medium II (TiM-MII) did not correlate significantly with CCTm (p = 0.34 and p = 0.46, respectively) ( Fig. 4a, b) or with CCTa (p = 0.15 and p = 0.92, respectively) (Fig. 4c, d). The disparity of OCT devices (using different optical sources and spatial resolutions (Ang et al. 2018)) and the different measurement conditions do not allow a direct comparison of the results presented in the studies mentioned above. None of these studies reported the use of an optical correction for image distortion in addition to those presumably already applied by the AS-OCT systems themselves. Wolf et al. as well as Amato et al., showed-as mentioned above-tomographic CCT values as being similar to ultrasonic CCT measurements (performed for model validation). However, these measurements were performed with an old OCT model on whole globes and on posterior donor lenticle, respectively. These results have therefore a low clinical relevance for current CD measurements.
In this study, we compared the results of two types of corneal measurements of the raw data obtained with the AS-OCT CASIA 2 (Tomey, Nagoya, Japan), a Swept Source (SS) AS-OCT, using a raster scan, through a plastic flask filled with MI or MII. The ''manual'' measurements (CCTm) were performed with the CASIA 2 Software version 3G.3, using the raw 3D volume data imaged and post-processed by the AS-OCT system. The instrument manufacturer does not provide information about spatial and optical built-in corrections applied in the images themselves, and we did not apply any additional correction factor after the measurements. The (semi-)automated measurements (CCTa) were perform by the customized MATLAB software, according to Mäurer, Damian, Langenbucher et al. (Janunts et al. 2016;Damian et al. 2017;Mäurer et al. 2019), that processed the same raw 3D volume data imaged with the AS-OCT. The software fits a spherocylindrical model to both corneal surfaces based on the 3D volume data after removing artefacts from the cell culture flask and the medium, and corrects the spatial (geometric) and optical (refractive) distortions that may be caused by the axial scan through the plastic flask and the organ culture medium. This method was validated using a corneal-like PMMA phantom with design data according to the Gullstrand schematic model eye (Damian et al. 2017;Mäurer et al. 2019). The two methods showed very similar CCT results, with CCTm = 1129 ± 6.8 lm versus CCTa = 1149 ± 2.6 lm in MI and CCTm = 820 ± 5.1 lm versus CCTa = 811 ± 4.4 lm in MII, which were comparable (p = 0.006 in MI; p = 0.002 in MII). As the CCTa values have previously been validated, these results suggest that CCTm values do not require additional correction and could be used as generated by the AS-OCT system. This result may seem surprising since even with the plastic cell culture flask and the medium (I or II) on both sides of the CD the tomographer is not working under normal conditions with air/cornea/aqueous humor interfaces.
The results were strongly reliable for both CCTm and CCTa methods in both organ culture media but showed a higher repeatability using the CCTa, not significant in MI (p = 0.09), however significant in MII (p = 0.03). However, the CCTm was impaired in 54.4% of the cases in MI and 36.8% of the cases in MII due to the presence of distortions in the post-processed volumes of the AS-OCT Software due to artefacts, i.e. light reflections on the plastic surface or micromovements during the examination (Fig. 2). A solution to this problem may be to perform several successive images to obtain ''at least'' one analyzable measurement. Nevertheless, we showed that in 6% of the corneas, all 5 images were not sufficient to properly measure the CCTm of one image. According to our experience prior to the implementation of the custom MATLAB software, the number of AS-OCT imaging required to obtain an analyzable 3D model can be up to 10, considerably increasing the time required to measure the CCT for each individual cornea. This might put into question the statements about ''repeatability'' or ''reproducibility'', despite the seemingly high reliability of CCTm (based on measurable 3D-volumes). In contrast, the MATLAB software was able to generate a measurable 3Dvolume for 96% of the CD in MI and 99.2% of the CD in MII. Considering the issue caused by the artefacts and the considerable time saving by using the custom MATLAB software (both for imaging and analysis), the semi-automated method seems to be much more efficient to analyze corneal transplants in clinical practice.
As part of this study, we also analyzed the effect of the CD storage time in MI or MII (TiM) on the absolute CCT values (Fig. 4). This analysis showed no effect of TiM on absolute CCT values (neither CCTm nor CCTa). While this seems to apply under stable conditions (prolonged storage in the same medium), there is evidence nonetheless that the cornea swells in MI (Pels 1997;Doughman 1980) and deswells in MII during the first hours after the transfer (Borderie et al. 1997;Hamon et al. 2021).
In practice, an AS-OCT remains an expensive device and the procurement of such equipment for the eye bank is questionable. Being part of a university corneal transplantation center with on-site eye bank, we use the same AS-OCT that is used daily in the framework of our consultations. As a standard procedure, tomographic images of all CD are performed preoperatively at our Department of Ophthalmology since 2018 . This option-available in the majority of large ophthalmology centers-does not require any additional costs. For the pachymetrical measurements, we use the (semi-)automated method using our self-programmed MATLAB software, which almost systematically allows us to obtain a CCT value. The generation of a 3D model presents advantages compared to a single central measurement with cross-line scan. Firstly, a 3D model allows a complete mapping with not only a central measurement (CCT) but also peripheral measurements. If the peripheral measurements are generally of minor importance, they may enable the detection of specific abnormalities such as marginal pellucid degeneration or an ulcer, for example. Secondly, this self-programmed MATLAB software contains others functionalities which require 3D modeling, such as a keratometric analysis (steep and flat anterior and posterior radii of curvature) with detection of curvature anomalies (Damian et al. 2017) and an automated detection of corneal opacities (stromal scars, corneal flap after laser in situ keratomileusis (LASIK), corneal dystrophies, …) ). This information is of great importance for eye banks, considering that a cornea with curvature anomaly or stromal densification can be used-for example-for DMEK, if the endothelial cell count (ECC) is sufficient. All these functionalities run in an integrated and simultaneous sequence after a single imaging, which supports the efficiency of this semi-automated method in comparison to a manual CCT measurement or a single central cross-line scan. In the future, curvature mapping could allow to orient the transplant during the surgery in order to align the steep and flat meridians of donor and recipient together and thus minimize postoperative astigmatism (''harmonization'' of donor and recipient tomography) ). Until today, this self-programmed MATLAB software was used to analyze raw data from 3DV-files from CASIA or CASIA 2 AS-OCT (Tomey, Nagoya, Japan) and from AVI-files from Spectralis AS-OCT (Heidelberg Engineering, Heidelberg, Germany). However, the software should be able to use 3DV or AVI-files from others models of AS-OCT with little or no specific adjustments. This software is currently not available for sale or download for other departments or eye banks and is still subject to research projects. In the future, it could be made available if there is sufficient demand.
This study has two limitations. First, the fixation of the CD to the cell culture flask holder may cause a slight deformation of the cornea, whereby the measured geometry may not fully correspond with ''in vivo'' conditions. Second, the measurements do not take into account the inhomogeneous epithelial layer, which could cause a bias of approximately 50 lm (Neubauer et al. 2002). Since the condition of the corneal epithelium is assumed to be comparable with sequential measurements of the same CD, this limitation could affect the absolute measurement of CCT, but is not expected to affect measures such as reliability or repeatability.
In conclusion, both manual and (semi-)automated methods using the AS-OCT CASIA 2 showed a high reliability in both culture media I and II. Despite the similarity between manual and (semi-)automated measurements, manual measurements were frequently hampered by artifacts or distortions. For this reason, semi-automated measurements seem to be more efficient and should be preferred.