Fully automatic evaluation of the corneal endothelium from in vivo confocal microscopy

Automatic Determination of Endothelial Cell Density From Donor Cornea Endothelial Cell Images

Purpose

To determine endothelial cell density (ECD) from real-world donor cornea endothelial cell (EC) images using a self-supervised deep learning segmentation model.

Methods

Two eye banks (Eversight, VisionGift) provided 15,138 single, unique EC images from 8169 donors along with their demographics, tissue characteristics, and ECD. This dataset was utilized for self-supervised training and deep learning inference. The Cornea Image Analysis Reading Center (CIARC) provided a second dataset of 174 donor EC images based on image and tissue quality. These images were used to train a supervised deep learning cell border segmentation model. Evaluation between manual and automated determination of ECD was restricted to the 1939 test EC images with at least 100 cells counted by both methods.

Results

The ECD measurements from both methods were in excellent agreement with rc of 0.77 (95% confidence interval [CI], 0.75-0.79; P < 0.001) and bias of 123 cells/mm2 (95% CI, 114-131; P < 0.001); 81% of the automated ECD values were within 10% of the manual ECD values. When the analysis was further restricted to the cropped image, the rc was 0.88 (95% CI, 0.87-0.89; P < 0.001), bias was 46 cells/mm2 (95% CI, 39-53; P < 0.001), and 93% of the automated ECD values were within 10% of the manual ECD values.

Conclusions

Deep learning analysis provides accurate ECDs of donor images, potentially reducing analysis time and training requirements.

Translational Relevance

The approach of this study, a robust methodology for automatically evaluating donor cornea EC images, could expand the quantitative determination of endothelial health beyond ECD.

Quality Control in the Corneal Bank with Artificial Intelligence: Comparison of a New Deep Learning-based Approach with Conventional Endothelial Cell Counting by the "Rhine-Tec Endothelial Analysis System"

Endothelial cell density (ECD) is a crucial parameter for the release of corneal grafts for transplantation. The Lions Eye Bank of Baden-Württemberg uses the "Rhine-Tec Endothelial Analysis System" for ECD quantification, which is based on a fixed counting frame method considering only a small sample of 15 to 40 endothelial cells. The measurement result therefore depends on the frame placement and manual correction of the cells counted within the frame. To increase the sample size and create higher objectivity, we developed a new method based on "deep learning" that automatically detects all visible endothelial cells in the image. This study aims to compare this new method with the conventional Rhine-Tec system. 9375 archived phase-contrast microscopic images of consecutive grafts from the Lions Eye Bank were evaluated with the deep learning method and compared with the corresponding archived analyses of the Rhine-Tec system. Means, Bland-Altman and correlation analyses were compared. Comparable results were obtained for both methods. The mean difference between the Rhine-Tec system and the deep learning method was only - 23 cells/mm2 (95% confidence interval - 29 to - 17). There was a statistically significant positive correlation between the two methods, with a correlation coefficient of 0.748. What was striking in the Bland-Altman analysis were clustered deviations in the cell density range between 2000 and 2500 cells/mm2 - with higher values in the Rhine-Tec system. The comparable results for cell density measurement values underline the validity of the deep learning-based method. The deviations around the formal threshold for graft release of 2000 cells/mm2 are most likely explained by the higher objectivity of the deep learning method and the fact that measurement frames and manual corrections were specifically selected to reach the formal threshold of 2000 cells/mm2 when the full area endothelial quality was good. This full area assessment of the graft endothelium cannot currently be replaced by deep learning methods and remains the most important basis for graft release for keratoplasty.

Mobile-CellNet: Automatic Segmentation of Corneal Endothelium Using an Efficient Hybrid Deep Learning Model

PURPOSE

The corneal endothelium, the innermost layer of the human cornea, exhibits a morphology of predominantly hexagonal cells. These endothelial cells are believed to have limited regeneration capacity, and their density decreases over time. Endothelial cell density (ECD) can therefore be used to measure the health of the corneal endothelium and the overall cornea. In clinical settings, specular microscopes are used to image this layer. Owing to the unavailability of reliable automatic tools, technicians often manually mark the cell centers and borders to measure ECD for such images, a process that is time and resource-consuming.

METHODS

In this article, we propose Mobile-CellNet, a novel completely automatic, efficient deep learning-based cell segmentation algorithm to estimate ECD. This uses 2 similar image segmentation models working in parallel along with image postprocessing using classical image processing techniques. We also compare the proposed algorithm with widely used biomedical image segmentation networks U-Net and U-Net++.

RESULTS

The proposed technique achieved a mean absolute error of 4.06% for the ECD on the test set, comparable with the error for U-Net of 3.80% ( P = 0.185 for difference), but requiring almost 31 times fewer floating-point operations (FLOPs) and 34 times fewer parameters.

CONCLUSIONS

Mobile-CellNet accurately segments corneal endothelial cells and reports ECD and cell morphology efficiently. This can be used to develop tools to analyze specular corneal endothelial images in remote settings.

Machine Learning Analysis of Postkeratoplasty Endothelial Cell Images for the Prediction of Future Graft Rejection

Purpose

This study developed machine learning (ML) classifiers of postoperative corneal endothelial cell images to identify postkeratoplasty patients at risk for allograft rejection within 1 to 24 months of treatment.

Methods

Central corneal endothelium specular microscopic images were obtained from 44 patients after Descemet membrane endothelial keratoplasty (DMEK), half of whom had experienced graft rejection. After deep learning segmentation of images from all patients' last and second-to-last imaging, time points prior to rejection were analyzed (175 and 168, respectively), and 432 quantitative features were extracted assessing cellular spatial arrangements and cell intensity values. Random forest (RF) and logistic regression (LR) models were trained on novel-to-this-application features from single time points, delta-radiomics, and traditional morphometrics (endothelial cell density, coefficient of variation, hexagonality) via 10 iterations of threefold cross-validation. Final assessments were evaluated on a held-out test set.

Results

ML classifiers trained on novel-to-this-application features outperformed those trained on traditional morphometrics for predicting future graft rejection. RF and LR models predicted post-DMEK patients' allograft rejection in the held-out test set with >0.80 accuracy. RF models trained on novel features from second-to-last time points and delta-radiomics predicted post-DMEK patients' rejection with >0.70 accuracy. Cell-graph spatial arrangement, intensity, and shape features were most indicative of graft rejection.

Conclusions

ML classifiers successfully predicted future graft rejections 1 to 24 months prior to clinically apparent rejection. This technology could aid clinicians to identify patients at risk for graft rejection and guide treatment plans accordingly.

Translational Relevance

Our software applies ML techniques to clinical images and enhances patient care by detecting preclinical keratoplasty rejection.

Corneal endothelium assessment in specular microscopy images with Fuchs' dystrophy via deep regression of signed distance maps

Specular microscopy assessment of the human corneal endothelium (CE) in Fuchs' dystrophy is challenging due to the presence of dark image regions called guttae. This paper proposes a UNet-based segmentation approach that requires minimal post-processing and achieves reliable CE morphometric assessment and guttae identification across all degrees of Fuchs' dystrophy. We cast the segmentation problem as a regression task of the cell and gutta signed distance maps instead of a pixel-level classification task as typically done with UNets. Compared to the conventional UNet classification approach, the distance-map regression approach converges faster in clinically relevant parameters. It also produces morphometric parameters that agree with the manually-segmented ground-truth data, namely the average cell density difference of -41.9 cells/mm2 (95% confidence interval (CI) [-306.2, 222.5]) and the average difference of mean cell area of 14.8 µm 2 (95% CI [-41.9, 71.5]). These results suggest a promising alternative for CE assessment.

DenseUNets with feedback non-local attention for the segmentation of specular microscopy images of the corneal endothelium with guttae

Corneal guttae, which are the abnormal growth of extracellular matrix in the corneal endothelium, are observed in specular images as black droplets that occlude the endothelial cells. To estimate the corneal parameters (endothelial cell density [ECD], coefficient of variation [CV], and hexagonality [HEX]), we propose a new deep learning method that includes a novel attention mechanism (named fNLA), which helps to infer the cell edges in the occluded areas. The approach first derives the cell edges, then infers the well-detected cells, and finally employs a postprocessing method to fix mistakes. This results in a binary segmentation from which the corneal parameters are estimated. We analyzed 1203 images (500 contained guttae) obtained with a Topcon SP-1P microscope. To generate the ground truth, we performed manual segmentation in all images. Several networks were evaluated (UNet, ResUNeXt, DenseUNets, UNet++, etc.) and we found that DenseUNets with fNLA provided the lowest error: a mean absolute error of 23.16 [cells/mm[Formula: see text]] in ECD, 1.28 [%] in CV, and 3.13 [%] in HEX. Compared with Topcon's built-in software, our error was 3-6 times smaller. Overall, our approach handled notably well the cells affected by guttae, detecting cell edges partially occluded by small guttae and discarding large areas covered by extensive guttae.

U-Net Convolutional Neural Network for Segmenting the Corneal Endothelium in a Mouse Model of Fuchs Endothelial Corneal Dystrophy

PURPOSE

The purpose of this study was to assess the U-Net-based convolutional neural network performance for segmenting corneal endothelium and guttae of Fuchs endothelial corneal dystrophy.

METHODS

Twenty-eight images of corneal endothelial cells and guttae of Col8a2L450W/L450W knock-in mice were obtained by specular microscopy. We used 20 images as training data to develop the U-Net for analyzing guttae and cell borders. The proposed network was validated using independent test data of 8 images. Cell density, hexagonality, and coefficient of variation were calculated from the predicted cell borders and compared with ground truth.

RESULTS

U-Net allowed the prediction of cell borders and guttae, and overlays of those segmentations on specular microscopy images highly corresponded to ground truth. The average number of guttae per field was 6.25 ± 8.07 for ground truth and 6.25 ± 7.87 when predicted by the network (Pearson correlation coefficient 0.989, P = 3.25 × 10 -6 ). The guttae areas were 1.60% ± 1.79% by manual determination and 1.90% ± 2.02% determined by the network (Pearson correlation coefficient 0.970, P = 6.72 × 10 -5 ). Cell density, hexagonality, and coefficient of variation analyzed by the proposed network for cell borders showed very strong correlations with ground truth (Pearson correlation coefficient 0.989, P = 3.23 × 10 -6 , Pearson correlation coefficient 0.978, P = 2.66 × 10 -5 , and Pearson correlation coefficient 0.936, P = 6.20 × 10 -4 , respectively).

CONCLUSIONS

We demonstrated proof of concept for application of U-Net for objective analysis of corneal endothelial cells and guttae in Fuchs endothelial corneal dystrophy, based on limited ground truth data.

Trainable watershed-based model for cornea endothelial cell segmentation

Segmentation of the medical image plays a significant role when it comes to diagnosis using computer aided system. This article focuses on the human corneal endothelium’s health, which is one of the filed research interests, especially in the human cornea. Various pathological environments fasten the extermination of the endothelial cells, which in turn decreases the cell density in an abnormal manner. Dead cells worsen the hexagonal design. The mutilated endothelial cells can no longer revive back and that gives room for neighbouring cells to migrate and expand so that they can fill in the space. The latter results in cell elongation that is unpredictable as well as increase in size and thinning. Cell density and shape are therefore considered major parameters when it comes to explaining the health condition attributed to corneal endothelium. In this study, medical feature extraction was obtained depending on the segmentation of the endothelial cell boundary, and the task of segmentation of such objects especially the thin, transparent, and unclear cell boundary is considered challenging due to the nature of the image capture during endothelium layer examination by ophthalmologists using confocal or specular microscopy. The resulting image suffers from various issues that affect the quality of the image. Low quality is due to non-uniformity of illumination and the presence of a lot of noise and artefacts resulting from high amounts of distortion, and most of these limitations are present because of the nature of the imaging modality. Usually, images contain certain kind of noise and also continuous shadow. Furthermore, the cells are separated by poor border, thereby leading to great difficulty in the segmentation of the images. The irregular shape of cell and also the contrast of such images seem to be low as they possess blurry boundaries with diverse objects existing in addition to the lack of homogeneity. The main aim of the study is to propose and develop a totally automatic, robust, and real-time model for the segmentation of endothelial cells of the human cornea obtained by in vivo microscopy and computation of different clinical features of endothelial cells. To achieve the aim of this study a new scheme of image enhancement was proposed such as the Contrast-Limited Adaptive Histogram Equalisation (CLAHE) technique to enhance contrast. After that, a new image denoising technique called Wavelet Transform Filter and Butterworth Bandpass for Segmentation is used. Subsequently, brightness level correction is applied by using the moving average filter and the CLAHE to reduce the effects of the non-uniform image lighting produced as a result of the previous step. The main aim of this article is the segmentation of endothelial cells, which involves precise detection of the endothelial contours. So a new segmentation model was proposed such that the shape of the cells will be extracted, and the contours were highlighted. This stage is followed by clinical feature extraction and uses the features for diagnosis. In this stage, several relevant clinical features such as pleomorphism mean cell perimeter, mean cell density, mean cell area, and polymegathism are extracted. The role of these clinical features is crucial for the early detection of corneal pathologies as well as the evaluation of the health of the corneal endothelium layer. The findings of this study were promising.

CellsDeepNet: A Novel Deep Learning-Based Web Application for the Automated Morphometric Analysis of Corneal Endothelial Cells

The quantification of corneal endothelial cell (CEC) morphology using manual and semi-automatic software enables an objective assessment of corneal endothelial pathology. However, the procedure is tedious, subjective, and not widely applied in clinical practice. We have developed the CellsDeepNet system to automatically segment and analyse the CEC morphology. The CellsDeepNet system uses Contrast-Limited Adaptive Histogram Equalization (CLAHE) to improve the contrast of the CEC images and reduce the effects of non-uniform image illumination, 2D Double-Density Dual-Tree Complex Wavelet Transform (2DDD-TCWT) to reduce noise, Butterworth Bandpass filter to enhance the CEC edges, and moving average filter to adjust for brightness level. An improved version of U-Net was used to detect the boundaries of the CECs, regardless of the CEC size. CEC morphology was measured as mean cell density (MCD, cell/mm2), mean cell area (MCA, μm2), mean cell perimeter (MCP, μm), polymegathism (coefficient of CEC size variation), and pleomorphism (percentage of hexagonality coefficient). The CellsDeepNet system correlated highly significantly with the manual estimations for MCD (r = 0.94), MCA (r = 0.99), MCP (r = 0.99), polymegathism (r = 0.92), and pleomorphism (r = 0.86), with p < 0.0001 for all the extracted clinical features. The Bland–Altman plots showed excellent agreement. The percentage difference between the manual and automated estimations was superior for the CellsDeepNet system compared to the CEAS system and other state-of-the-art CEC segmentation systems on three large and challenging corneal endothelium image datasets captured using two different ophthalmic devices.

An automatic approach for cell detection and segmentation of corneal endothelium in specular microscope

PURPOSE

Specular microscopy is an indispensable tool for clinicians seeking to monitor the corneal endothelium. Automated methods of determining endothelial cell density (ECD) are limited in their ability to analyze images of poor quality. We describe and assess an image processing algorithm to analyze corneal endothelial images.

METHODS

A set of corneal endothelial images acquired with a Konan CellChek specular microscope was analyzed using three methods: flex-center, Konan Auto Tracer, and the proposed method. In this technique, the algorithm determines the region of interest, filters the image to differentiate cell boundaries from their interiors, and utilizes stochastic watershed segmentation to draw cell boundaries and assess ECD based on the masked region. We compared ECD measured by the algorithm with manual and automated results from the specular microscope.

RESULTS

We analyzed a total of 303 images manually, using the Auto Tracer, and with the proposed image processing method. Relative to manual analysis across all images, the mean error was 0.04% in the proposed method (p = 0.23 for difference) whereas Auto Tracer demonstrated a bias towards overestimation, with a mean error of 5.7% (p = 2.06× 10^-8). The relative mean absolute errors were 6.9% and 7.9%, respectively, for the proposed and Auto Tracer. The average time for analysis of each image using the proposed method was 2.5 s.

CONCLUSION

We demonstrate a computationally efficient algorithm to analyze corneal endothelial cell density that can be implemented on devices for clinical and research use.

Automated Image Segmentation of the Corneal Endothelium in Patients With Fuchs Dystrophy

Purpose

To perform segmentation of specular microscopy (SM) images of the corneal endothelium for comparing average perimeter length (APL) between Fuchs endothelial corneal dystrophy (FECD) patients and healthy subjects.

Methods

A retrospective review of clinical records of FECD patients and those with healthy endothelium was carried out to collect images of the endothelium. The images were segmented by modified U-Net, a deep learning architecture, followed by the Watershed algorithm to resolve merged cell borders (<5%). The segmented images were analyzed for endothelial cell density (ECDUW) and APL.

Results

The combination of the U-Net and Watershed algorithm, referred to as the UW approach, enabled a complete segmentation of the endothelium. In healthy, ECDUW was close to estimates by SM and manual segmentation (31 subjects; P > 0.1). However, in FECD, ECDUW was closer to estimates by manual segmentation but not by SM (27 patients; P < 0.001). ECDUW in FECD (2547 ± 499 cells/mm2; 60 patients) was smaller compared to that in the healthy (2713 ± 401 cells/mm2; 70 subjects) (P < 0.001). APL in the healthy was 66.87 ± 7.68 µm/cell (70 subjects), but it increased with %Guttae in FECD (56.60-195.30 µm/cell; 60 patients) (P < 0.0001).

Conclusions

The UW approach is precise for the segmentation of SM images from the healthy and FECD. Our analysis has revealed that APL increases with %Guttae.

Translational Relevance

The average perimeter length of the corneal endothelium, which represents the length of the paracellular pathway for fluid flux into the stroma, is increased in Fuchs dystrophy.

Advances in Imaging of Subbasal Corneal Nerves With Micro-Optical Coherence Tomography

Purpose

To investigate the most peripheral corneal nerve plexus using high-resolution micro-optical coherence tomography (µOCT) imaging and to assess µOCT's clinical potential as a screening tool for corneal and systemic diseases.

Methods

An experimental high-resolution (1.5 × 1.5 × 1 µm) µOCT setup was applied for three-dimensional imaging of the subbasal nerve plexus in nonhuman primates (NHPs) and swine within 3 hours postmortem. Morphologic features of subbasal nerves in µOCT were compared to β3 tubulin-stained fluorescence confocal microscopy (FCM). Parameters such as nerve density, nerve distribution, and imaging repeatability were evaluated, using semiautomatic image analysis in form of a custom corneal surface segmentation algorithm and NeuronJ.

Results

Swine and NHP corneas showed the species-specific nerve morphology in both imaging modalities. Most fibers showed a linear course, forming a highly parallel pattern, converging in a vortex with overall nerve densities varying between 9.51 and 24.24 mm/mm2. The repeatability of nerve density quantification of the µOCT scans as approximately 88% in multiple image recordings of the same cornea.

Conclusions

Compared to the current gold standard of FCM, µOCT's larger field of view of currently 1 × 1 mm increases the conclusiveness of density measurements, which, coupled with µOCT's feature of not requiring direct contact, shows promise for future clinical application. The nerve density quantification may be relevant for screening for systemic disease (e.g., peripheral neuropathy).

Translational Relevance

Technological advances in OCT technology may enable a quick assessment of corneal nerve density, which could be valuable evaluating ophthalmic and systemic peripheral innervation.

A Review On digital image processing techniques for in-Vivo confocal images of the cornea

This work reviews the scientific literature regarding digital image processing for in vivo confocal microscopy images of the cornea. We present and discuss a selection of prominent techniques designed for semi- and automatic analysis of four areas of the cornea (epithelium, sub-basal nerve plexus, stroma and endothelium). The main context is image enhancement, detection of structures of interest, and quantification of clinical information. We have found that the preprocessing stage lacks of quantitative studies regarding the quality of the enhanced image, or its effects in subsequent steps of the image processing. Threshold values are widely used in the reviewed methods, although generally, they are selected empirically and manually. The image processing results are evaluated in many cases through comparison with gold standards not widely accepted. It is necessary to standardize values to be quantified in terms of sensitivity and specificity of methods. Most of the reviewed studies do not show an estimation of the computational cost of the image processing. We conclude that reliable, automatic, computer-assisted image analysis of the cornea is still an open issue, constituting an interesting and worthwhile area of research.

Unbiased corneal tissue analysis using Gabor-domain optical coherence microscopy and machine learning for automatic segmentation of corneal endothelial cells

SIGNIFICANCE

An accurate, automated, and unbiased cell counting procedure is needed for tissue selection for corneal transplantation.

AIM

To improve accuracy and reduce bias in endothelial cell density (ECD) quantification by combining Gabor-domain optical coherence microscopy (GDOCM) for three-dimensional, wide field-of-view (1  mm2) corneal imaging and machine learning for automatic delineation of endothelial cell boundaries.

APPROACH

Human corneas stored in viewing chambers were imaged over a wide field-of-view with GDOCM without contacting the specimens. Numerical methods were applied to compensate for the natural curvature of the cornea and produce an image of the flattened endothelium. A convolutional neural network (CNN) was trained to automatically delineate the cell boundaries using 180 manually annotated images from six corneas. Ten additional corneas were imaged with GDOCM and compared with specular microscopy (SM) to determine performance of the combined GDOCM and CNN to achieve automated endothelial counts relative to current procedural standards.

RESULTS

Cells could be imaged over a larger area with GDOCM than SM, and more cells could be delineated via automatic cell segmentation than via manual methods. ECD obtained from automatic cell segmentation of GDOCM images yielded a correlation of 0.94 (p  <  0.001) with the manual segmentation on the same images, and correlation of 0.91 (p  <  0.001) with the corresponding manually counted SM results.

CONCLUSIONS

Automated endothelial cell counting on GDOCM images with large field of view eliminates selection bias and reduces sampling error, which both affect the gold standard of manual counting on SM images.

Deep Learning for Assessing the Corneal Endothelium from Specular Microscopy Images up to 1 Year after Ultrathin-DSAEK Surgery

Purpose

To present a fully automatic method to estimate the corneal endothelium parameters from specular microscopy images and to use it to study a one-year follow-up after ultrathin Descemet stripping automated endothelial keratoplasty.

Methods

We analyzed 383 post ultrathin Descemet stripping automated endothelial keratoplasty images from 41 eyes acquired with a Topcon SP-1P specular microscope at 1, 3, 6, and 12 months after surgery. The estimated parameters were endothelial cell density (ECD), coefficient of variation (CV), and hexagonality (HEX). Manual segmentation was performed in all images.

Results

Our method provided an estimate for ECD, CV, and HEX in 98.4% of the images, whereas Topcon's software had a success rate of 71.5% for ECD/CV and 30.5% for HEX. For the images with estimates, the percentage error in our method was 2.5% for ECD, 5.7% for CV, and 5.7% for HEX, whereas Topcon's software provided an error of 7.5% for ECD, 17.5% for CV, and 18.3% for HEX. Our method was significantly better than Topcon's (P < 0.0001) and was not statistically significantly different from the manual assessments (P > 0.05). At month 12, the subjects presented an average ECD = 1377 ± 483 [cells/mm2], CV = 26.1 ± 5.7 [%], and HEX = 58.1 ± 7.1 [%].

Conclusions

The proposed method obtains reliable and accurate estimations even in challenging specular images of pathologic corneas.

Translational Relevance

CV and HEX, not currently used in the clinic owing to a lack of reliability in automatic methods, are useful biomarkers to analyze the postoperative healing process. Our accurate estimations allow now for their clinical use.

Convolutional neural network-based regression for biomarker estimation in corneal endothelium microscopy images

The morphometric parameters of the corneal endothelium - cell density (ECD), cell size variation (CV), and hexagonality (HEX) - provide clinically relevant information about the cornea. To estimate these parameters, the endothelium is commonly imaged with a non-contact specular microscope and cell segmentation is performed to these images. In previous work, we have developed several methods that, combined, can perform an automated estimation of the parameters: the inference of the cell edges, the detection of the region of interest (ROI), a post-processing method that combines both images (edges and ROI), and a refinement method that removes false edges. In this work, we first explore the possibility of using a CNN-based regressor to directly infer the parameters from the edge images, simplifying the framework. We use a dataset of 738 images coming from a study related to the implantation of a Baerveldt glaucoma device and a standard clinical care regarding DSAEK corneal transplantation, both from the Rotterdam Eye Hospital and both containing images of unhealthy endotheliums. This large dataset allows us to build a large training set that makes this approach feasible. We achieved a mean absolute percentage error (MAPE) of 4.32% for ECD, 7.07% for CV, and 11.74% for HEX. These results, while promising, do not outperform our previous work. In a second experiment, we explore the use of the CNN-based regressor to improve the post-processing method of our previous approach in order to adapt it to the specifics of each image. Our results showed no clear benefit and proved that our previous post-processing is already highly reliable and robust.

Quantitative and qualitative evaluation of deep learning automatic segmentations of corneal endothelial cell images of reduced image quality obtained following cornea transplant

We are developing automated analysis of corneal-endothelial-cell-layer, specular microscopic images so as to determine quantitative biomarkers indicative of corneal health following corneal transplantation. Especially on these images of varying quality, commercial automated image analysis systems can give inaccurate results, and manual methods are very labor intensive. We have developed a method to automatically segment endothelial cells with a process that included image flattening, U-Net deep learning, and postprocessing to create individual cell segmentations. We used 130 corneal endothelial cell images following one type of corneal transplantation (Descemet stripping automated endothelial keratoplasty) with expert-reader annotated cell borders. We obtained very good pixelwise segmentation performance (e.g., Dice coefficient = 0.87 ± 0.17 , Jaccard index = 0.80 ± 0.18 , across 10 folds). The automated method segmented cells left unmarked by analysts and sometimes segmented cells differently than analysts (e.g., one cell was split or two cells were merged). A clinically informative visual analysis of the held-out test set showed that 92% of cells within manually labeled regions were acceptably segmented and that, as compared to manual segmentation, automation added 21% more correctly segmented cells. We speculate that automation could reduce 15 to 30 min of manual segmentation to 3 to 5 min of manual review and editing.

In vivo corneal endothelium imaging using ultrahigh resolution OCT

We investigate the influence of optical coherence tomography (OCT) system resolution on high-quality in vivo en face corneal endothelial cell images of the monkey eye, to allow for quantitative analysis of cell density. We vary the lateral resolution of the ultrahigh resolution (UHR) OCT system (centered at 850 nm) by using different objectives, and the axial resolution by windowing the source spectrum. By suppressing the motion of the animal, we are able to obtain a high-quality en face corneal endothelial cell map in vivo using UHR OCT for the first time with a lateral resolution of 3.1 µm. Increasing lateral resolution did not result in a better image quality but a smaller field of view (FOV), and the axial resolution had little impact on the visualization of corneal endothelial cells. Quantitative analysis of cell density was performed on in vivo en face OCT images of corneal endothelial cells, and the results are in agreement with previously reported data. Our study may offer a practical guideline for designing OCT systems that allow for in vivo corneal endothelial cell imaging with high quality.

Comparison of automated vs manual analysis of corneal endothelial cell density and morphology in normal and corneal endothelial dystrophy-affected dogs

OBJECTIVE

To determine the efficacy of automated imaging software of the Nidek ConfoScan 4 confocal biomicroscope at analyzing canine corneal endothelial cell density and morphology in health and disease, by comparing to a manual analysis method.

ANIMAL STUDIED

Nineteen eyes of 10 dogs were evaluated and include three Beagles, three Jack Russell Terriers, and four miscellaneous breeds. Twelve clinically normal and seven eyes affected with corneal endothelial dystrophy (CED) were scanned and analyzed.

PROCEDURES

Endothelial cell density (ECD), mean and standard deviation (SD) of cell area, percent polymegathism, mean and SD of the number of cell sides, and percent pleomorphism were calculated using automated and manual methods for each scan.

RESULTS

The automated analysis showed significantly greater ECD in comparison with the manual frame method due to misidentification of cell domains in CED-affected dogs. No significant differences in ECD were observed between normal and CED-affected dogs in automated analysis, while CED-affected dogs showed significantly lower ECD in manual frame method and planimetry. Using both automated and manual methods, CED-affected dogs showed greater variability of cell area or the number of cell sides than normal dogs.

CONCLUSION

The automated imaging software is unable to accurately identify cell borders in CED-affected dogs resulting in inaccurate estimates of ECD. Thus, manual analysis is recommended for use in clinical trials assessing adverse events associated with novel medical treatments and/or surgical procedures.

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.

Machine learning for segmenting cells in corneal endothelium images

Images of the endothelial cell layer of the cornea can be used to evaluate corneal health. Quantitative biomarkers extracted from these images such as cell density, coefficient of variation of cell area, and cell hexagonality are commonly used to evaluate the status of the endothelium. Currently, fully-automated endothelial image analysis systems in use often give inaccurate results, while semi-automated methods, requiring trained image analysis readers to identify cells manually, are both challenging and time-consuming. We are investigating two deep learning methods to automatically segment cells in such images. We compare the performance of two deep neural networks, namely U-Net and SegNet. To train and test the classifiers, a dataset of 130 images was collected, with expert reader annotated cell borders in each image. We applied standard training and testing techniques to evaluate pixel-wise segmentation performance, and report corresponding metrics such as the Dice and Jaccard coefficients. Visual evaluation of results showed that most pixel-wise errors in the U-Net were rather non-consequential. Results from the U-Net approach are being applied to create endothelial cell segmentations and quantify important morphological measurements for evaluating cornea health.

Fully convolutional architecture vs sliding-window CNN for corneal endothelium cell segmentation

Background

Corneal endothelium (CE) images provide valuable clinical information regarding the health state of the cornea. Computation of the clinical morphometric parameters requires the segmentation of endothelial cell images. Current techniques to image the endothelium in vivo deliver low quality images, which makes automatic segmentation a complicated task. Here, we present two convolutional neural networks (CNN) to segment CE images: a global fully convolutional approach based on U-net, and a local sliding-window network (SW-net). We propose to use probabilistic labels instead of binary, we evaluate a preprocessing method to enhance the contrast of images, and we introduce a postprocessing method based on Fourier analysis and watershed to convert the CNN output images into the final cell segmentation. Both methods are applied to 50 images acquired with an SP-1P Topcon specular microscope. Estimates are compared against a manual delineation made by a trained observer.

Results

U-net (AUC=0.9938) yields slightly sharper, clearer images than SW-net (AUC=0.9921). After postprocessing, U-net obtains a DICE=0.981 and a MHD=0.22 (modified Hausdorff distance), whereas SW-net yields a DICE=0.978 and a MHD=0.30. U-net generates a wrong cell segmentation in only 0.48% of the cells, versus 0.92% for the SW-net. U-net achieves statistically significant better precision and accuracy than both, Topcon and SW-net, for the estimates of three clinical parameters: cell density (ECD), polymegethism (CV), and pleomorphism (HEX). The mean relative error in U-net for the parameters is 0.4% in ECD, 2.8% in CV, and 1.3% in HEX. The computation time to segment an image and estimate the parameters is barely a few seconds.

Conclusions

Both methods presented here provide a statistically significant improvement over the state of the art. U-net has reached the smallest error rate. We suggest a segmentation refinement based on our previous work to further improve the performance.

Corneal Endothelial Cell Segmentation by Classifier-Driven Merging of Oversegmented Images

Corneal endothelium images obtained by in vivo specular microscopy provide important information to assess the health status of the cornea. Estimation of clinical parameters, such as cell density, polymegethism, and pleomorphism, requires accurate cell segmentation. State-of-the-art techniques to automatically segment the endothelium are error-prone when applied to images with low contrast and/or large variation in cell size. Here, we propose an automatic method to segment the endothelium. Starting with an oversegmented image comprised of superpixels obtained from a stochastic watershed segmentation, the proposed method uses intensity and shape information of the superpixels to identify and merge those that constitute a cell, using support vector machines. We evaluated the automatic segmentation on a data set of in vivo specular microscopy images (Topcon SP-1P), obtaining 95.8% correctly merged cells and 2.0% undersegmented cells. We also evaluated the parameter estimation against the results of the vendor's built-in software, obtaining a statistically significant better precision in all parameters and a similar or better accuracy. The parameter estimation was also evaluated on three other data sets from different imaging modalities (confocal microscopy, phase-contrast microscopy, and fluorescence confocal microscopy) and tissue types (ex vivo corneal endothelium and retinal pigment epithelium). In comparison with the estimates of the data sets' authors, we achieved statistically significant better accuracy and precision in all parameters except pleomorphism, where a similar accuracy and precision were obtained.

En Face and Cross-sectional Corneal Tomograms Using Sub-micron spatial resolution Optical Coherence Tomography

Accurate diagnosis of corneal pathology and morphological identification of different corneal layers require clear delineation of corneal three-dimensional structures and en face or cross-sectional imaging of palisade of Vogt (POV), neovascularization (NV) or corneal nerves. Here we report a prototype of full-field optical coherence tomography (FF-OCT) system with isotropic sub-micron spatial resolution in the en face and cross-sectional views. It can also provide three-dimensional reconstructed images and a large field of view (FOV) by stitching tomograms side by side. We validated the imaging power of this prototype in in vivo rat and rabbit eyes, and quantified anatomical characteristics such as corneal layer thickness, endothelial cell density and the intensity profile of different layers. This FF-OCT delineated the ridge-like structure of POV, corneal nerve bundles, and conjunctival vessels in rat eyes. It also clearly identified the vessel walls and red blood cells in rabbit model of corneal NV. The findings provided by this FF-OCT are expected to facilitate corneal disease diagnosis and treatment.

A fully automated cell segmentation and morphometric parameter system for quantifying corneal endothelial cell morphology

BACKGROUND AND OBJECTIVE

Corneal endothelial cell abnormalities may be associated with a number of corneal and systemic diseases. Damage to the endothelial cells can significantly affect corneal transparency by altering hydration of the corneal stroma, which can lead to irreversible endothelial cell pathology requiring corneal transplantation. To date, quantitative analysis of endothelial cell abnormalities has been manually performed by ophthalmologists using time consuming and highly subjective semi-automatic tools, which require an operator interaction. We developed and applied a fully-automated and real-time system, termed the Corneal Endothelium Analysis System (CEAS) for the segmentation and computation of endothelial cells in images of the human cornea obtained by in vivo corneal confocal microscopy.

METHODS

First, a Fast Fourier Transform (FFT) Band-pass filter is applied to reduce noise and enhance the image quality to make the cells more visible. Secondly, endothelial cell boundaries are detected using watershed transformations and Voronoi tessellations to accurately quantify the morphological parameters of the human corneal endothelial cells. The performance of the automated segmentation system was tested against manually traced ground-truth images based on a database consisting of 40 corneal confocal endothelial cell images in terms of segmentation accuracy and obtained clinical features. In addition, the robustness and efficiency of the proposed CEAS system were compared with manually obtained cell densities using a separate database of 40 images from controls (n = 11), obese subjects (n = 16) and patients with diabetes (n = 13).

RESULTS

The Pearson correlation coefficient between automated and manual endothelial cell densities is 0.9 (p < 0.0001) and a Bland-Altman plot shows that 95% of the data are between the 2SD agreement lines.

CONCLUSIONS

We demonstrate the effectiveness and robustness of the CEAS system, and the possibility of utilizing it in a real world clinical setting to enable rapid diagnosis and for patient follow-up, with an execution time of only 6 seconds per image.