Study Population

This study was approved by the University of Miami Institutional Review Board. All participants provided written informed consent before enrollment. The study design followed the tenets of the Declaration of Helsinki for biomedical research.

Forty-nine eyes of 49 patients and 32 eyes of 32 normal controls were prospectively and consecutively recruited from December 2016 to January 2019 at Bascom Palmer Eye Institute, University of Miami (Miami, Florida, USA). Inclusion criteria for participation included uneventful penetrating keratoplasty (PK) or Descemet stripping automated endothelial keratoplasty (DSAEK) surgery performed greater than 1 month prior. Exclusion criteria included corneal grafts with microbial infection or past history of a rejection episode. For the unoperated control eyes, a best-corrected visual acuity (BCVA) better than 20/25 on the Snellen scale and no corneal abnormalities as detected by slit-lamp examination were required to be included in the study. Individuals with recent contact lens use, ocular diseases, previous ocular surgery, and systemic diseases with ocular involvement were excluded. Slit-lamp examination was performed on each eye by a masked cornea specialist (either M.A. or S.Y.) in order to assign the examined cornea into either a clear graft or actively rejecting category. Active endothelial graft rejection was diagnosed by detecting new keratic precipitates (KPs) or a Khodadoust line in the presence of anterior chamber cells and new persistent corneal graft edema on 2 consecutive visits in a graft that was previously clear. ^, Clinical grading of corneal graft rejection based on the ejection severity was not attempted.

Image Acquisition and Analysis

Each participant received anterior segment HD-OCT (Envisu R2210; Bioptigen, Buffalo Grove, Illinois, USA) with 6-mm radial cuts centered on the corneal vertex. This device uses a super-luminescent diode light source with a central wavelength of 840 nm, and it has an axial resolution of 3 μm and a scanning speed of 32,000 A-scans per second with 36 frames per scan. Each participant was asked to look at a central fixation target and the presence of a visible specular reflection in all images confirmed optimal centration. In decentered grafts and post-PK eyes with high irregular astigmatism, the patient was asked to look at an optimized fixation target to maintain the geometric centration of the scan upon the graft. Custom-built segmentation software was used to segment the corneal epithelium, Descemet membrane, and endothelium automatically, with manual editing if needed ( Figure 1 ). Manual editing was executed by 2 masked operators when the automatic segmentation failed to delineate the irregularities on the endothelial layer in the rejecting corneal grafts. The segmentation software used the Random Sample Consensus method with a polynomial model to estimate the corneal layer boundaries from potential points obtained from thresholding the OCT image. Then, the estimates are refined by searching locally in the original image for the best points. The segmentation method is robust to the specular reflection, as each boundary is fitted to a polynomial model that depends on all points and is not much affected by the specular reflection, as shown in Figure 1 . Then, the segmentation of different frames of the scan is mapped into 3D and interpolated using bi-cubic interpolation to obtain the corneal surfaces. After that, 3D ray tracing is applied iteratively at each interpolated surface to correct for the refraction in the OCT imaging light by applying the vector form of Snell’s law at the refractive interface between each 2 successive layers ( Figure 2 A). Finally, the inter-surface distances are measured as the shortest axial distance between each consecutive surface to generate the thickness maps. We used the refractive index of 1.376 for the corneal layers. Thickness maps were regionally divided into a central 2-mm circle, surrounded by 2-mm-wide paracentral and 2-mm-wide peripheral concentric rings ( Figure 2 B). The average central, paracentral, and peripheral thickness parameters were calculated and used for further analysis. Validation studies of our automated image processing techniques were published in our previous work and the results were comparable to those of the manual operators.

High-definition optical coherence tomography (HD-OCT) images of a healthy unoperated cornea (First row, left), clear corneal graft (Second row, left), and an actively rejecting graft (Third row, left). Presets (Right column) show magnified images of the posterior part of the corresponding cornea on the left. The red dashed line represents the anterior boundary of the corneal epithelium (Epi), the green dashed line indicates the corneal endothelium (En), and the blue dashed line indicates the Descemet membrane (DM). In the presets, the thick arrows indicate the segmentation of the endothelium/Descemet membrane complex (En/DM). Endothelium/Descemet membrane complex thickness (En/DMT) is measured as the inter-surface distance between the segmented En and DM layers. The specular reflection (SR) confirmed optimal centration; however, the segmentation method is robust, as each boundary is fitted to a polynomial model, and thus does not get affected that much. In both healthy corneas and clear grafts, corneal En and DM were visualized as a band formed by 2 smooth hyperreflective lines with a translucent space in between (First and Second rows, right). In actively rejecting grafts (Third row), En/DM appeared as a thickened band bounded by 2 hyperreflective lines. The anterior line and the translucent space in between the lines were like those of the healthy corneas and clear grafts, while the posterior line had a broader hyperreflectivity (Third row, right) and occasional nodular excrescences (Third row, right, yellow arrows). Bars are 100 μm.

(A) Diagram showing the concept of thickness measurement using 3-dimensional ray tracing. The axial distances between the uncorrected surfaces (red arrows) represent the optical path length and it is converted to geometric distance (blue arrows) using the layer refractive index. Three-dimensional ray tracing is applied iteratively at each surface to correct for the refraction in the OCT imaging light by applying the vector form of Snell’s law at the refractive interface between each 2 successive layers. Then, the thickness is measured as the shortest distance (black arrows) between each consecutive surface. (B) The scheme of arrangement of the corneal regions for quantitative evaluation of the layer thickness: the central region (C1, C2) lies within a 1-mm radius, surrounded by 2 concentric paracentral (M1, M2, M3, M4, M5, M6) and outer (O1, O2, O3, O4, O5, O6) rings, each with a 2-mm width. I = inferior; N = nasal; S = superior; T = temporal.

Diagnostic Indices

We used 3 diagnostic indices to describe the regional microstructural characteristics of the endothelial/Descemet membrane (En/DM) complex: central, paracentral, and peripheral En/DMT. We previously reported the 2 interfaces of the En/DM complex as the 2 most posterior hyperreflective bands of the cornea on HD-OCT images. ^,

Statistical Analysis

Statistical analyses were performed using SPSS software version 22.0 (SPSS, Chicago, Illinois, USA) to calculate descriptive statistics for all eyes. The obtained TCT measurements were verified to have a normal distribution by assessment of histograms and a Shapiro-Wilk test of normality, while En/DMT was not. Therefore, mean ± standard deviation (SD) was used to characterize the distribution of the corneal thickness values, while the median values were used to characterize both En/DMT. In addition, 1-way analysis of variance (ANOVA) with post hoc comparisons was performed to account for the differences in central, paracentral, and peripheral corneal thickness, while a Kruskal-Wallis test with pairwise comparisons was used for regional En/DMT. Additionally, factorial ANOVA was performed to verify if the changes in the corneal thickness and En/DMT depend on the type of graft (PK vs DSAEK). The sensitivity and specificity of regional En/DMT and regional TCT in differentiating between studied groups were determined by generating receiver operating characteristic curves. In order to determine if central, paracentral, and peripheral En/DMT would be descriptive of graft rejection severity, coefficient of correlation (r-value) of those indices and rejection severity based on central TCT was computed. ^, Two-sided P values less than .05 were considered to be statistically significant. Moreover, intraclass correlation coefficients (ICC) were used to assess the inter-operator reliability of the manual measurements in the selected eyes. The ICC is defined as the ratio of the between-subjects variance to the sum of the pooled within-subjects variance and the between-subjects variance. The ICC interpretation that was used considered the reliability of the values as poor for values less than 0.2, fair for values from 0.21 to 0.40, moderate for values from 0.41 to 0.60, good for values from 0.61 to 0.80, and excellent for values higher than 0.80.