Morphological comparison of specular microscopy images may be a more robust indicator for endothelial stability than cell density estimations

Automated segmentation of the corneal endothelium in a large set of 'real-world' specular microscopy images using the U-Net architecture

Monitoring the density of corneal endothelial cells (CEC) is essential in the management of corneal diseases. Its manual calculation is time consuming and prone to errors. U-Net, a neural network for biomedical image segmentation, has shown promising results in the automated segmentation of images of healthy corneas and good quality. The purpose of this study was to assess its performance in “real-world” CEC images (variable quality, different ophthalmologic diseases). The outcome measures were: precision and recall of the extraction of CEC, correctness of CEC density estimation, detection of ungradable images. A classical approach based on grayscale morphology and water shedding was pursued for comparison. There was good agreement between the automated image analysis and the manual annotation from the U-Net. R-square from Pearson’s correlation was 0.96. Recall of CEC averaged 0.34 and precision 0.84. The U-Net correctly predicted the CEC density in a large set of images of healthy and diseased corneas, including images of poor quality. It robustly ignored image regions with poor visibility of CEC. The classical approach, however, did not provide acceptable results. R-square from Pearson’s correlation with the ground truth was as low as 0.35.

On the reliability (repeatability) of single pachymetry measures taken with specular microscope-based assessments of the human corneal endothelium

PURPOSE

To assess agreement between 3 measures of central corneal thickness (CCT) taken over a few minutes from nominally normal eyes with a non-contact specular microscope.

METHODS

100 eyes from 100 healthy adults (with an average age of 22 y) were assessed using the Topcon 3000 P instrument to obtain a high quality image of the endothelium and pachymetry.

RESULTS

The group mean CCT values from the 1st, 2nd and 3rd measures were 0.519, 0.520 and 0.520 mm, but the sets of values could differ by between - 0.020 mm and + 0.029 mm, i.e. between - 4.0 and + 5.6% of the average values. Paired comparisons (e.g. 2nd vs. 1st set) indicated limits of agreement (LoA) to be between - 0.020 and + 0.020 mm of the averaged value. Across the 3 measures, the averaged SD was 0.005 mm to give an estimate of the intra-subject variability (as the coefficient of variation, COV) of 0.91% (range 0 to 2.3%). The variability in the pachymetry measures was not predictably related to the averaged values of CCT (r = 0.022) or self-reported refractive error of the subjects (r = 0.012). Some repeat pachymetry assessments obviously included the same region of the endothelium.

CONCLUSIONS

Single pachymetry measures with a non-contact specular microscope are only likely to be able to generate CCT estimates within +/- 4% of the expected average values. This repeatability is comparable to single image estimates of endothelial cell density and therefore acceptable in most cases.

Evaluations of Corneas in Eyes with Isolated Iris Coloboma

PURPOSE

We aimed to evaluate the differences in the corneas of eyes with isolated iris coloboma.

METHODS

This study was designed as a prospective study. Among patients who applied to our hospital within the last 3-year period, those with isolated iris coloboma in one eye and a completely normal contralateral eye were included in the study. In addition to a complete ophthalmologic examination, all patients were scanned by Pentacam Scheimpflug camera in the dark, and a corneal endothelium cell count was taken using a Topcon Specular Microscope. Data entered using SPSS software were then evaluated by paired t-test; p < 0.05 value was considered statistically significant.

RESULTS

Nine patients, seven men and two women, with a mean age of 35.77 ± 22.50 (range 8-60) years, were included in the study. In comparison to normal eyes, the colobomatous eyes were statistically found to have lower keratometry values in diopters (43.82 ± 2.17, 44.46 ± 1.86; p = 0.040), a greater negative asphericity (Q value) (-0.38 ± 0.07, -0.22 ± 0.20; p = 0.042), increased corneal thickness (569.78 ± 25.42, 537.33 ± 27.36; p < 0.001), decreased anterior chamber depth (2.50 ± 0.32, 2.81 ± 0.50; p = 0.005), a shallower angle (30.51 ± 10.68, 35.29 ± 9.99; p = 0.057), higher best-fit sphere (BFS) values of the anterior and posterior corneal surfaces (7.88 ± 0.39, 7.67 ± 0.37; p = 0.001; 6.40 ± 0.26, 6.23 ± 0.19; p = 0.006), larger pupil diameter (3.48 ± 0.49, 2.80 ± 0.34; p = 0.001), smaller horizontal corneal diameter (HCD) (11.06 ± 0.42; 11.26 ± 0.45; p = 0.009), and higher value of endothelial cell density (ECD) (3161 ± 431; 2923 ± 590; p = 0.031).

CONCLUSION

We have determined that a number of factors such as keratometry values, particularly of the corneal endothelium, are different from those of normal eyes. These differences should be taken into account in the follow-up and treatment of eyes with iris coloboma.

Cell-by-cell alignment of repeated specular microscopy images from the same eye

PURPOSE

Modern specular microscopes (SM) robustly depict the same central area of the corneal endothelium at different time points through a built-in fixation light. However, repeated image acquisitions slightly shift and rotate because of minute changes in head position in the chin and forehead rest. This prevents the manual retrieval of individual corneal endothelial cells (CECs) in repeated measurements because SM images usually lack obvious landmarks. We devised and validated an image registration algorithm that aligns SM images from the same eye to make corresponding CECs coincide.

METHODS

We retrospectively selected 27 image pairs for the presence of significant image overlap. Each image pair had been recorded on the same day and of the same eye. We applied our registration method in each image pair. Two observers independently validated, by means of alternation flicker, that the image pairs had been correctly aligned. We also repeatedly applied our registration method on unrelated image pairs by randomly drawing images and making certain that the images did not originate from the same eye. This was done to assess the specifity of our method.

RESULTS

All automated registrations of the same-day and same-eye image pairs were accurate. However, one single image incorrectly failed to trigger the non-match diagnosis twice in 81 registration attempts between unrelated images. As it turned out, this particular image depicted only 73 CECs. The average number of CECs was 253 (range 73-393).

CONCLUSION

Repeated non-contact SM images can be automatedly aligned so that the corresponding CECs coincide. Any successful alignment can be considered as proof of the retrieval of identical CECs as soon as at least 100 CEC centroids have been identified. We believe our method is the first to robustly confirm endothelial stability in individual eyes.