Open Access
Cornea & External Disease  |   June 2025
Artificial Intelligence Aided Analysis of Anterior Segment Optical Coherence Tomography Imaging to Monitor the Device-Cornea Joint After Synthetic Cornea Implantation
Author Affiliations & Notes
  • Esen Karamursel Akpek
    The Ocular Surface Disease Clinic, The Wilmer Eye Institute, The Johns Hopkins University School of Medicine, Baltimore, MD, USA
  • Gavin Li
    The Ocular Surface Disease Clinic, The Wilmer Eye Institute, The Johns Hopkins University School of Medicine, Baltimore, MD, USA
    Icahn School of Medicine at Mount Sinai, New York, NY, USA
  • Anthony J. Aldave
    Stein Eye Institute, David Geffen School of Medicine at UCLA, Los Angeles, CA, USA
  • Guillermo Amescua
    Bascom Palmer Eye Institute, University of Miami School of Medicine, Miami, FL, USA
  • Kathryn A. Colby
    Department of Ophthalmology at New York University Grossman School of Medicine, New York University, New York, NY, USA
  • Maria S. Cortina
    Illinois Eye and Ear Infirmary, University of Illinois, Chicago, IL, USA
  • Jose de la Cruz
    Illinois Eye and Ear Infirmary, University of Illinois, Chicago, IL, USA
  • Jean-Marie A. Parel
    Bascom Palmer Eye Institute, University of Miami School of Medicine, Miami, FL, USA
  • Thomas Schmiedel
    W.L. Gore and Associates, Ophthalmic Venture, Elkton, MD, USA
  • Correspondence: Esen Karamursel Akpek, The Ocular Surface Disease Clinic, The Wilmer Eye Institute, Johns Hopkins University School of Medicine, 1800 Orleans Street, Woods 372, Baltimore, MD 21287-9238, USA. e-mail: [email protected] 
Translational Vision Science & Technology June 2025, Vol.14, 1. doi:https://doi.org/10.1167/tvst.14.6.1
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Esen Karamursel Akpek, Gavin Li, Anthony J. Aldave, Guillermo Amescua, Kathryn A. Colby, Maria S. Cortina, Jose de la Cruz, Jean-Marie A. Parel, Thomas Schmiedel; Artificial Intelligence Aided Analysis of Anterior Segment Optical Coherence Tomography Imaging to Monitor the Device-Cornea Joint After Synthetic Cornea Implantation. Trans. Vis. Sci. Tech. 2025;14(6):1. https://doi.org/10.1167/tvst.14.6.1.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose: The purpose of this study was to assess the utility of artificial intelligence (AI) assisted analysis of anterior segment optical coherence tomography (AS-OCT) imaging of the device-cornea joint in predicting outcomes of an intrastromal synthetic cornea device in a rabbit model.

Methods: Sixteen rabbits underwent intrastromal synthetic cornea implantation. Baseline anterior lamellar thickness was established using AS-OCT intraoperatively. Monthly postoperative clinical examinations and AS-OCT imaging were performed, focusing on the peri-optic zone. A convolutional neural network was trained using a subset of manually marked images to automatically detect anterior lamellar tissue. Images were aligned manually using reference coordinates. The tissue volume data were evaluated as both absolute volume and percentage change from baseline using AI.

Results: Sixteen rabbits were observed for 6 (n = 8) and 12 (n = 8) months. Mild focal anterior lamella thinning without retraction was seen near tight sutures in 2 rabbits (2/8) in the 6-month cohort, whereas 2 rabbits (2/8) in the 12-month cohort showed mild focal retraction from the optic stem with thinning. AI-assisted AS-OCT image analyses detected tissue volume reduction up to 3 months before clinical examination, with a reliable threshold of 5% change in tissue volume.

Conclusions: AI-assisted AS-OCT can detect peri-prosthetic tissue loss and predicting postoperative complications following an intrastromal synthetic cornea implantation in a rabbit model. Further studies are warranted to explore its clinical utility in human patients.

Translational Relevance: AI-assisted monitoring of peri-optic corneal tissue volume may be a useful screening modality to detect subclinical thinning after artificial corneal implantation and inform clinical decision making.

Introduction
Keratoprosthesis, or artificial corneal transplantation, may be considered in patients with corneal blindness and limited expectation of success from donor keratoplasty. The Boston type I keratoprosthesis (Boston KPro; Massachusetts Eye and Ear Infirmary, Boston, MA, USA) is currently the most frequently used device worldwide. Although the Boston KPro leads to much better intermediate-term visual outcomes compared to repeat donor keratoplasty in high risk cases,13 longer-term studies report increasing incidences of complications with each year of postoperative follow-up, including sterile or infectious keratitis, endophthalmitis, retinal detachment, and glaucoma.46 Progressive peri-prosthetic corneal tissue loss is a major risk factor for secondary infections and device exposure and potentially extrusion.4 The mechanisms involved in this tissue thinning and degradation are not well-known. Tumor necrosis factor (TNF) alpha and interleukin (IL)-1 beta were previously found elevated in the corneal tissue of a preclinical mouse model of miniature Boston KPro.7 Clinical studies linked elevated serum levels of TNF-alpha in patients with Boston KPro to the occurrence of sterile corneal melt.8 Increased tear film TNF-alpha and IL-1 beta were also noted in patients with Boston KPro associated glaucoma.9 In addition, matrix metalloproteinase (MMP-9) in the tears of patients with Boston KPro was found in patients with such complications.10 Indeed, sterile corneal lysis leading to device exposure and extrusion is not uncommon in clinical practice with reported rates between 3% and 18% 1 year after surgery.11,12 Although, secondary surgeries to salvage the device have been reported in the literature using various techniques,13,14 device explantation is not uncommon.15 
One proposed mechanism for this is chronic inflammation caused by the perpetual flexural oscillation of the device due to the rigidity of the material (polymethylmethacrylate) and lack of biointegration at the device-cornea joint. Constant kinetic energy is transmitted to the donor stroma housing the optical stem, as each blink applies an inward force on the device, while the intraocular pressure exerts a constant outward force.16,17 Therefore, the current efforts on developing an ideal artificial cornea, composed of a flexible material that is permissive to keratocyte adhesion/integration, continue. Toward that goal, we prototyped a novel flexible and fully synthetic device that is compliant with native human cornea. Feasibility of the minimally invasive surgical technique, biocompatibility of the materials used, and short-term clinical outcomes in an animal model have previously been published.17 
Anterior segment optical coherence tomography (AS-OCT), first introduced in 1994,18 has advanced rapidly with regard to scanning times and spatial resolution, and is now an integral tool for ophthalmologists, aiding in the diagnosis of various anterior segment pathologies.19 AS-OCT has been used to reliably assess postsurgical anatomic changes in the anterior segment depth and iridocorneal angle following Boston KPro.2023 An understudied application of AS-OCT in the keratoprostheses field is the evaluation of the device-cornea joint which is most relevant to retention of the device. A few previous reports highlighted the usefulness of the technology for providing the ability to examine the tissue-device interaction in situ and real time.2325 However, the challenges with repeatability of the imaging across time points as well as quantitative interpretation of changes have been acknowledged. Whereas fluorescein staining of the ocular surface can help identify epithelial defects and stromal thinning or retraction of the donor cornea when significant, subtle changes in the peri-optic tissue may be missed. Loss of tissue over time creates a peri-optic space (which can be situational with blinking or eye movements) and can serve as a potential entry for microbes or particles. We have previously used AS-OCT to assess the device-cornea joint following implantation of our synthetic cornea device.17,26 During our studies, the need for precise and accurate alignment of the images that shifted invariably from examination to examination became apparent. Therefore, a process to register our data sets in an absolute grid has been developed. 
With the rapid increase in big data availability and computational ability, artificial intelligence (AI) has seen progressive adoption within ophthalmology. AI's ability to integrate multiple parameters allows it to identify subtle patterns that may not be readily apparent to clinicians. AI-assisted AS-OCT imaging has been previously utilized for detecting keratoconus, screening for post-refractive ectasia, predicting complications in donor corneal transplantation, grading inflammation in anterior uveitis, and assessing the iridocorneal angle in angle-closure glaucoma.19,2733 
We developed an AI algorithm to analyze serial postoperative AS-OCT images to monitor the optic-cornea joint to detect subclinical changes, such as thinning or retraction of the recipient tissue, following implantation of a novel intrastromal synthetic corneal device in a rabbit model. 
Materials and Methods
All study protocols were approved by a credentialed review board (protocol # 2635SC by The W.L. Gore & Associates Institutional Animal Care and Use Committee Flagstaff, Arizona, and protocol # RB17M34 and RB23M12 by Johns Hopkins University Institutional Animal Care and Use Committee, Baltimore, Maryland). The tenets of the Declaration of Helsinki regarding the ethical treatment of animal subjects were adhered to throughout the study. 
Device Construct, Surgical Technique, and Intra-Operative Imaging of the Anterior Corneal Lamellae
The intrastromal synthetic cornea device is a single-piece, optic-skirt design, flexible, suturable prosthesis made of transparent and compact perfluoroalkoxy alkane (PFA; proprietary to W.L. Gore & Associates, Inc., Newark, DE, USA; Fig. 1). The skirt houses 16 macroapertures to provide diffusion of fluid and nutrients to the anterior corneal lamella. The skirt and the optic-perimeter are overlaid with a porous ingrowth surface utilizing expanded polytetrafluoroethylene (ePTFE) to allow integration of the corneal stroma. The ingrowth surfaces are rendered hydrophilic with a temperature-resistant polyvinyl alcohol-based coating and become translucent when wetted, to improve cosmesis. The skirt lies within the deep corneal stromal pocket. The optic extends between the ocular surface anteriorly and anterior chamber posteriorly through a full-thickness central trephination site.17,26,34 
Figure 1.
 
Schematic (left) and external appearance (right) of the intrastromal synthetic cornea device. The diagram is drawn to scale and all dimensions are expressed in mm.
Figure 1.
 
Schematic (left) and external appearance (right) of the intrastromal synthetic cornea device. The diagram is drawn to scale and all dimensions are expressed in mm.
Details of the surgical technique have been previously published.17,26 Briefly, a scleral fixation ring and stay sutures are used to stabilize the globe. The cornea is marked for centration. First, a baseline AS-OCT imaging was obtained from the recipient cornea to capture baseline measurements. These images were used to help the AI model distinguish normal corneal architecture from changes induced by the implanted device. Then, a partial thickness central trephination is made at \(\frac{2}{3}\) to \(\frac{3}{4}\) depth using a 4.0 mm trephine. An intrastromal lamellar pocket measuring 8 mm in outer diameter is then created 360 degrees manually, using a disposable blade, aiming for 60% to 80% of the recipient corneal thickness assessed using intraoperative OCT. The anterior chamber is entered through a paracentesis using a surgical blade and filled with viscoelastic after irrigating with heparinized balanced salt solution 20 IU/mL to prevent clotting. The corneal button is removed using a pair of curved microscissors. The device is then folded and inserted with the optic captured within the full-thickness trephination site and the skirt secured within the deep stromal pocket using 16 interrupted 10-0 nylon sutures. The paracentesis is closed using a single 10-0 nylon suture after having exchanged the viscoelastic with balanced salt solution. A surgical video is included in a previously published report.17 
Sixteen New Zealand white rabbits aged 12 months and older underwent intrastromal device implantation in their right eye, using the above surgical technique. The rabbits were randomly assigned to the 6-month (n = 8) or 12-month (n = 8) cohorts. An operating microscope with built-in AS-OCT capability (Proveo 8 Ophthalmic Microscope and EnFocus Ultra-Deep OCT HR, Leica Microsystems, Wetzlar, Germany) was used for imaging. The images taken immediately after creation of the stromal pocket served as the baseline anterior lamellar thickness (Fig. 2, top section). Five cross-sectional OCT images, providing 10 anterior lamella regions (every 36 degrees, with 2 per cross section), were used to calculate the average baseline thickness as follows. Three points (proximal, medial, and distal), spaced equally along the boundary of the anterior lamella, were identified radially starting from the trephination site extending 1 mm distally, using a trained AI point detection algorithm. These points were then used to fit circular segments to each group of three points. The shortest distance between these circles is measured in a radial direction for subsequent analyses (see Fig. 2, bottom section). Twenty such measurements along the circular segments were taken in each region. Five of these measurements of anterior lamella thickness were selected approximately 0.7 mm distal to the 4 mm central trephination site. An average of these constitutes the calculated baseline thickness in the region. This process was repeated for all 10 regions using a LabVIEW (National Instruments Corp., Austin, TX, USA) program to automate the calculations starting with the AI-detected points. The resulting 10 average reference thickness values, spanning 360 degrees around the optic-cornea joint, were utilized to compute the “Gaussian Weighted Average” curve, a weighted average using a Gaussian curve with a standard deviation of 1.3 points, to represent the anterior lamellar thickness at implantation as a reference set for subsequent calculations. 
Figure 2.
 
Anterior segment optical coherence tomography image, taken after partial thickness trephination, followed by dissection of the lamellar pocket and prior to full-thickness trephination, depicting the method used to determine baseline anterior corneal lamella thickness value. After a 4 mm diameter central trephination is performed at \(\frac{2}{3}\) to \(\frac{3}{4}\) stromal depth (red arrow), an intra-stromal pocket with an 8 mm radius is created distally in the recipient cornea. The anterior lamella (green arrow) is identified relative to the trephination site and the measured tissue thickness serving as the baseline (A). The series of white and green points depict the automated point detection, identifying the top and bottom surfaces of the anterior lamella, with each of the three-point groups fitted to a circle segment. Five measurements (blue indicated circles), at 0.7 mm peripheral to the 4 mm diameter trephination site (B) where the optic-cornea joint resides, were taken and averaged to yield the anterior lamella thickness.
Figure 2.
 
Anterior segment optical coherence tomography image, taken after partial thickness trephination, followed by dissection of the lamellar pocket and prior to full-thickness trephination, depicting the method used to determine baseline anterior corneal lamella thickness value. After a 4 mm diameter central trephination is performed at \(\frac{2}{3}\) to \(\frac{3}{4}\) stromal depth (red arrow), an intra-stromal pocket with an 8 mm radius is created distally in the recipient cornea. The anterior lamella (green arrow) is identified relative to the trephination site and the measured tissue thickness serving as the baseline (A). The series of white and green points depict the automated point detection, identifying the top and bottom surfaces of the anterior lamella, with each of the three-point groups fitted to a circle segment. Five measurements (blue indicated circles), at 0.7 mm peripheral to the 4 mm diameter trephination site (B) where the optic-cornea joint resides, were taken and averaged to yield the anterior lamella thickness.
Post-Implantation Imaging of Anterior Corneal Stromal Lamella and Developing an Artificial Intelligence Algorithm for Image Analyses
High resolution averaged AS-OCT images were taken of the implanted eyes at monthly intervals postoperative, following clinical examination. Clinical examination included direct visualization under the operating microscope as well as fluorescein dye instillation to examine patterns of “pooling in the recipient cornea” to detect tissue thinning or “guttering at the optic-cornea joint” to detect tissue retraction. Because of strong cornea-device bonding and full collagen deposition into the porous ingrowth surface around the optic wall and skirt, the AS-OCT images failed to differentiate the tissue from the thin layer of synthetic porous material.17 Therefore, the lamella and ingrowth surface were digitally labeled as a unit, using Photoshop (Adobe Inc., San Jose, CA, USA; Fig. 3A). Twenty-five images from 8 AS-OCT stacks were taken as a representative cross section of scans which provided a total of 200 training images with 2 marked regions each. This “ground truth” data were used to train a convolutional neural network (CNN) to automatically detect the tissue (including the ingrowth surface) during all other scans at all time points. 
Figure 3.
 
Manual mark-up (A) and artificial intelligence-detected (B) area of anterior lamellar tissue, adjacent to the optic wall of the device (red arrow indicates the device optic and the green arrow indicates the skirt). Because the area includes the epithelium and stroma as well as the porous ingrowth material laid over the optic wall and skirt of the device, the thickness of the ingrowth surface (50 microns) was then subtracted to yield tissue only. The scale of measurements is depicted in image (C), with the proximal (250 microns, marked in blue) and total (1000 microns marked in white) areas of measurement.
Figure 3.
 
Manual mark-up (A) and artificial intelligence-detected (B) area of anterior lamellar tissue, adjacent to the optic wall of the device (red arrow indicates the device optic and the green arrow indicates the skirt). Because the area includes the epithelium and stroma as well as the porous ingrowth material laid over the optic wall and skirt of the device, the thickness of the ingrowth surface (50 microns) was then subtracted to yield tissue only. The scale of measurements is depicted in image (C), with the proximal (250 microns, marked in blue) and total (1000 microns marked in white) areas of measurement.
To allow accurate comparisons between follow-up versus baseline OCT images, positional variation of the eye had to be accounted for. Therefore, after manually labeling the paracentesis, all images were automatically rotated and aligned to the position of the paracentesis suture according to surgeon's view, using LabVIEW (National Instruments Corp., Austin, TX, USA). 
The detection accuracy was evaluated using the “Intersection over Union” (IoU) method and a value of 0.91 was achieved. IoU determines the model accuracy by measuring the overlap between the detected area (segmented region) and the ground truth area (training data). A Python (Python version 3.12.1; Python Software Foundation, Beaverton, OR, USA) script is utilized to conduct image analysis on the marked regions including corrections for over and under detection of tissue. Random visual inspections throughout the data gathering period confirmed the accuracy of the method. 
The data reported evaluates two regions of interest: the corneal stromal tissue immediately adjacent to the optic (250 microns in radial extent) and stromal tissue overlying the rest of the skirt (1000 microns in radial extent; Fig. 3C). Results obtained from the scans are converted to provide the absolute thickness of the tissue expressed in microns, using LabVIEW. The absolute value was then used to compute the percentage change from baseline at every examination time to detect tissue loss (i.e. less than 100% of baseline) or tissue gain (i.e. more than 100% of baseline), using a sine function (allowing for a slightly tilted cut plane) fitted to the “Average Reference Thickness” data. The data from the proximal 250 microns radius of stromal tissue closest to the optic data was considered as being most sensitive to the changes at the optic-cornea junction and relevant to the device retention. The images from both the 6-month and the 12-month cohorts and at all postoperative time points were mixed to train the algorithm to ensure that the model learned from a range of postoperative findings at various intervals. 
Results
The clinical results of the experiments have previously been published.17 All devices remained in situ with clear optics over a period of 6 months (n = 8 devices) and 12 months (n = 8 devices). No retroprosthetic membrane, conjunctivalization or neovascularization of the recipient, glaucoma, cataract formation, or retinal detachment was observed. Many of the rabbits had loosening of sutures which were removed and replaced. Two rabbits in the 6-month group had mild, focal anterior lamella thinning without retraction adjacent to the optic near tight sutures. Three postoperative complications occurred in the 12-month group. One rabbit diagnosed with endophthalmitis was euthanized on day 228. Mild sterile focal retraction of anterior lamella occurred in 2 rabbits, which were terminated on days 225 and 315. 
Color plots were generated based on absolute thickness, to offer a geographic overview of the AI-detected tomographic data, mapped to the area of recipient corneal tissue overlying the device skirt with the 250 microns as well as the 1000 microns radial sections around the optic-cornea joint (Fig. 4). 
Figure 4.
 
Tomographic color plot illustration of peri-prosthetic corneal tissue map showing tissue thickness at 1000 microns radial sections from the optic, overlying the entire skirt (A). The thin inner circle indicates the 250 microns radial sections immediately adjacent to the optic. (B) Magnified view of the 250 microns radial section, considered most relevant to the anatomic outcomes and post-surgical retention, showing the tissue thickness in more detail. All color plots were aligned post-imaging, according to the paracentesis suture at the 12 o'clock position as a reference point.
Figure 4.
 
Tomographic color plot illustration of peri-prosthetic corneal tissue map showing tissue thickness at 1000 microns radial sections from the optic, overlying the entire skirt (A). The thin inner circle indicates the 250 microns radial sections immediately adjacent to the optic. (B) Magnified view of the 250 microns radial section, considered most relevant to the anatomic outcomes and post-surgical retention, showing the tissue thickness in more detail. All color plots were aligned post-imaging, according to the paracentesis suture at the 12 o'clock position as a reference point.
Serial longitudinal tomographic plots were able to detect subclinical focal anterior lamellar tissue thinning correlating with the OCT images in Figure 5
Figure 5.
 
A collage of sequential artificial intelligence (AI) aided anterior lamellar thickness color tomography plots (top), actual anterior segment optical coherence tomography (AS-OCT) images (middle), and clinical pictures (with and without fluorescein staining) (bottom) starting from 6 months through 10 months postoperative. A focal area of thinning in peri-prosthetic tissue immediately adjacent to the optic, which could easily be missed during clinical examination, is evident on the color tomography plot (250 microns) and AS-OCT image as early as month 7.
Figure 5.
 
A collage of sequential artificial intelligence (AI) aided anterior lamellar thickness color tomography plots (top), actual anterior segment optical coherence tomography (AS-OCT) images (middle), and clinical pictures (with and without fluorescein staining) (bottom) starting from 6 months through 10 months postoperative. A focal area of thinning in peri-prosthetic tissue immediately adjacent to the optic, which could easily be missed during clinical examination, is evident on the color tomography plot (250 microns) and AS-OCT image as early as month 7.
The threshold to detect changes in tissue volume is approximately 5%. This estimate is based on the fact that the original IoU for the primary data is 0.91. Further correction for missing or over-detected pixels yields an approximate 95% of tissue detection accuracy. Because the volumes are assessed on the basis of a fixed radial length (250 and 1000 microns), the average height of the tissue is the dominant factor in variability which should be 5% or less. 
The clinical results of the 6- and 12-month cohorts have been reported previously.17 Overall, 2 of 8 animals in the 6-month cohort and 3 of 8 in the 12-month cohort showed focal thinning/retraction of the anterior corneal lamella immediately adjacent to the peri-prosthetic section, which could be detected with AS-OCT several months before the slit-lamp examination with fluorescein staining. All areas of focal tissue thinning/retraction were invariably associated with tight sutures presumably due to compression creating desiccation as well as outward traction of the corneal tissue separating it the optical wall.17 
Discussion
Our study demonstrates that AI-assisted analyses of AS-OCT images can help detect subclinical tissue thinning with or without retraction in a rabbit model of an intrastromal synthetic cornea implantation. Current practices for postoperative monitoring of the Boston KPro recipients include close observation with slit-lamp examination, digital palpation intraocular pressure measurement, and regular visual field examinations and posterior segment OCT to assess for glaucoma progression.16 However, AS-OCT is not routinely utilized to monitor the status of the recipient cornea, even though it is well known that peri-prosthetic epithelial defects or thinning may lead to keratolysis and device extrusion.35 Two previous studies reported serial AS-OCT imaging for measuring anterior chamber angle for glaucoma progression.20,21 However, these studies also noted challenges such as difficulties with image alignment due to patient eye movement, the lack of software specifically designed for the KPro interface, which makes it difficult to detect subtle changes between visits, and prolonged image acquisition times. These studies did not use AI to facilitate analysis and required manual interpretation of the serial scans to quantify the degree of angle closure. We herein demonstrate that AI-assisted analysis of anterior lamellar tissue thickness measured with AS-OCT can be used to detect changes in recipient tissue thickness earlier than clinical examination. This may help facilitate timely appropriate intervention to prevent secondary infection and device extrusion. 
Machine learning algorithms have been widely implemented in the field of retina to detect progression of diabetic retinopathy and age-related macular degeneration.3639 With regard to corneal transplantation, the majority of efforts have focused on predicting the likelihood of mechanical or physiological complications following endothelial keratoplasty using AS-OCT images.30,31,40 There currently is lack of knowledge regarding the utility of AI-assisted AS-OCT imaging in appropriate patient selection or postoperative monitoring following penetrating keratoplasty or keratoprosthesis implantation. 
There are several limitations of this study. First, although our image post-processing enables accurate alignment of all longitudinal OCT images by using the paracentesis suture as a reference point, the suture’s position had to be manually labeled before the images could be automatically rotated for alignment. Further work is needed to train an additional algorithm to eliminate the need for manual labeling which is time-consuming and provides room for human error. Second, our analysis focused specifically on the optic-cornea joint, which is the area known to be most sensitive to focal retraction and thinning. However, there may be corneal changes distal to this area that our algorithm is currently unable to assess. A third limitation is that our algorithm cannot differentiate changes in corneal thickness due to various etiologies such as edema, inflammation/infection, or scarring. Therefore, tomography plots must be clinically correlated. Last, whereas our results show promise in detecting early or subclinical recipient corneal changes, the clinical utility in keratoprosthesis recipients remains to be assessed with longer-term studies, particularly to establish the thresholds of thinning that warrant intervention. 
Importantly, our results should be interpreted with caution. We do not have any human patient data over short or long-term follow-up to compare these results to other devices currently in use with respect to the rates of tissue thinning, sterile keratolysis/retraction with or without device extrusion. The potential of improved outcomes with this device owing to the biocompatibility of the materials used34 and the unique anatomical design of it allowing biointegration at device-cornea joint41 remain to be seen in further studies. Patents on the various aspects of the technology are being pursued jointly by the Johns Hopkins University and W.L. Gore & Associates. 
In conclusion, this feasibility study demonstrates the utility of AI-assisted analyses of corneal thickness measurements at the device-cornea joint using AS-OCT images in a rabbit model using a novel intracorneal device. Although there may be significant differences with regard to the ability of AS-OCT to image human cornea versus rabbit cornea, and the materials used in devices that are in clinical use versus the materials used in our novel device, the potential positive impact of this technique in clinical practice justifies further studies. A semi-automated tool integrated into AS-OCT software may help with real-time imaging and objective quantification of changes. 
Acknowledgments
The authors thank Theo Fleck, B.S., of W.L. Gore and Associates for his contributions in establishing the AI/machine learning (ML) to data interface and implementing the error correction algorithm. Yannet Interian, PhD, and Michael Ruddy, PhD, of University of San Francisco, California, helped create, test, and tune the AI/ML algorithms for both reference point and tissue area detection steps of this study. 
Supported by W.L. Gore & Associates Inc. 
Disclosure: E.K. Akpek, Johns Hopkins University (E), W.L. Gore & Associates Inc. (F); G. Li, Johns Hopkins University (E), W.L. Gore & Associates Inc. (F); A.J. Aldave, W.L. Gore & Associates (C); G. Amescua, W.L. Gore & Associates (C); K.A. Colby, W.L. Gore & Associates (C); M.S. Cortina, W.L. Gore & Associates (C); J. de la Cruz, W.L. Gore & Associates (C); J.-M.A. Parel, W.L. Gore & Associates (C); T. Schmiedel, W.L. Gore & Associates (E) 
References
Ahmad S, Mathews PM, Lindsley K, et al. Boston type 1 keratoprosthesis versus repeat donor keratoplasty for corneal graft failure: a systematic review and meta-analysis. Ophthalmology. 2016; 123(1): 165–177. [CrossRef] [PubMed]
Bonneau S, Tong CM, Yang Y, Harissi-Dagher M. The treatment of end-stage corneal disease: penetrating keratoplasty compared with Boston type 1 keratoprosthesis. Graefes Arch Clin Exp Ophthalmol. 2022; 260(9): 2781–2790. [CrossRef] [PubMed]
Akpek EK, Cassard SD, Dunlap K, Hahn S, Ramulu PY. Donor corneal transplantation vs Boston type 1 keratoprosthesis in patients with previous graft failures: a retrospective single center study (an American Ophthalmological Society Thesis). Trans Am Ophthalmol Soc. 2015; 113: T3. [PubMed]
Aravena C, Yu F, Aldave AJ. Long-term visual outcomes, complications, and retention of the Boston type I keratoprosthesis. Cornea. 2018; 37(1): 3. [CrossRef] [PubMed]
Kanu LN, Niparugs M, Nonpassopon M, Karas FI, de la Cruz JM, Cortina MS. Predictive factors of Boston type I keratoprosthesis outcomes: a long-term analysis. Ocul Surf. 2020; 18(4): 613–619. [CrossRef] [PubMed]
Tsou BC, Koseoglu ND, Akpek EL, Liu TYA. Ten-year outcome of Boston type I keratoprosthesis surgery at a tertiary care center. Cornea. 2024; 43: 982–988. [CrossRef] [PubMed]
Crnej A, Omoto M, Dohlman TH, Dohlman CH, Dana R. Corneal inflammation after miniature keratoprosthesis implantation. Invest Ophthalmol Vis Sci. 2014; 56(1): 185–189. [CrossRef] [PubMed]
Paschalis EI, Taniguchi EV, Chodosh J, et al. Blood levels of tumor necrosis factor alpha and its type 2 receptor are elevated in patients with Boston type I keratoprosthesis. Curr Eye Res. 2019; 44(6): 599–606. [CrossRef] [PubMed]
Geoffrion D, Robert MC, Di Polo A, Koenekoop RK, Agoumi Y, Harissi-Dagher M. Tear film cytokine profile of patients with the Boston keratoprosthesis type 1: comparing patients with and without glaucoma. Invest Ophthalmol Vis Sci. 2021; 62(4): 20. [CrossRef] [PubMed]
Arteaga AC, Weiss MC, Perez R, Cortina MS. Metalloproteinase-9 in the ocular surface of patients with implanted Boston type 1 keratoprosthesis. Cornea Open. 2023; 2(1): e0008. [CrossRef] [PubMed]
Ciolino JB, Belin MW, Todani A, Al-Arfaj K, Rudnisky CJ, Boston Keratoprosthesis Type 1 Study Group. Retention of the Boston keratoprosthesis type 1: multicenter study results. Ophthalmology. 2013; 120(6): 1195–1200. [CrossRef] [PubMed]
Aldave AJ, Sangwan VS, Basu S, et al. International results with the Boston type I keratoprosthesis. Ophthalmology. 2012; 119(8): 1530–1538. [CrossRef] [PubMed]
Feng MT, Burkhart ZN, McKee Y, Price FW. A technique to rescue keratoprosthesis melts. Cornea. 2013; 32(10): 1407–1411. [CrossRef] [PubMed]
Chan CC, LoVerde L, Qiang J, Nordlund ML, Holland EJ. Incidence, risk factors, and surgical management of Boston type 1 keratoprothesis corneal melts, leaks, and extrusions. Cornea. 2016; 35(8): 1049–1056. [CrossRef] [PubMed]
Bouhout S, Robert MC, Deli S, Harissi-Dagher M. Corneal melt after Boston keratoprosthesis: clinical presentation, management, outcomes and risk factor analysis. Ocul Immunol Inflamm. 2018; 26(5): 693–699. [CrossRef] [PubMed]
Akpek EK, Karakus S, Yohannan J, et al. Reliability of several glaucoma tests in patients with Boston type 1 keratoprosthesis. Cornea. 2022; 41(3): 310–316. [CrossRef] [PubMed]
Akpek EK, Aldave AJ, Amescua G, et al. Twelve-month clinical and histopathological performance of a novel synthetic cornea device in rabbit model. Transl Vis Sci Technol. 2023; 12(8): 9. [CrossRef] [PubMed]
Izatt JA, Hee MR, Swanson EA, et al. Micrometer-scale resolution imaging of the anterior eye in vivo with optical coherence tomography. Arch Ophthalmol Chic Ill 1960. 1994; 112(12): 1584–1589.
Garcia Marin YF, Alonso-Caneiro D, Vincent SJ, Collins MJ. Anterior segment optical coherence tomography (AS-OCT) image analysis methods and applications: a systematic review. Comput Biol Med. 2022; 146: 105471. [CrossRef] [PubMed]
Qian CX, Hassanaly S, Harissi-Dagher M. Anterior segment optical coherence tomography in the long-term follow-up and detection of glaucoma in Boston type I keratoprosthesis. Ophthalmology. 2015; 122(2): 317–325. [CrossRef] [PubMed]
Kang JJ, Allemann N, de la Cruz J, Cortina MS. Serial analysis of anterior chamber depth and angle status using anterior segment optical coherence tomography after Boston keratoprosthesis. Cornea. 2013; 32(10): 1369–1374. [CrossRef] [PubMed]
Kang JJ, Allemann N, Vajaranant TS, Vajaranant T, de la Cruz J, Cortina MS. Anterior segment optical coherence tomography for the quantitative evaluation of the anterior segment following Boston keratoprosthesis. PLoS One. 2013; 8(8): e70673. [CrossRef] [PubMed]
Garcia JPS, Ritterband DC, Buxton DF, De la Cruz J. Evaluation of the stability of Boston type I keratoprosthesis-donor cornea interface using anterior segment optical coherence tomography. Cornea. 2010; 29(9): 1031–1035. [CrossRef] [PubMed]
Zarei-Ghanavati S, Betancurt C, Mas AM, Wang J, Perez VL. Ultra high resolution optical coherence tomography in Boston type I keratoprosthesis. J Ophthalmic Vis Res. 2015; 10(1): 26–32. [PubMed]
Shapiro BL, Cortés DE, Chin EK, et al. High-resolution spectral domain anterior segment optical coherence tomography in type 1 Boston keratoprosthesis. Cornea. 2013; 32(7): 951–955. [CrossRef] [PubMed]
Akpek EK, Van Court S, Glass S, Schmiedel T, Troutman M. Short-term clinical outcomes of a novel corneal prosthetic device in a rabbit model. Cornea. 2020; 39(6): 706–712. [CrossRef] [PubMed]
Waldstein SM, Vogl WD, Bogunovic H, Sadeghipour A, Riedl S, Schmidt-Erfurth U. Characterization of drusen and hyperreflective foci as biomarkers for disease progression in age-related macular degeneration using artificial intelligence in optical coherence tomography. JAMA Ophthalmol. 2020; 138(7): 740–747. [CrossRef] [PubMed]
Ting DSW, Pasquale LR, Peng L, et al. Artificial intelligence and deep learning in ophthalmology. Br J Ophthalmol. 2019; 103(2): 167–175. [CrossRef] [PubMed]
Ting DSJ, Foo VH, Yang LWY, et al. Artificial intelligence for anterior segment diseases: emerging applications in ophthalmology. Br J Ophthalmol. 2021; 105(2): 158–168. [CrossRef] [PubMed]
Treder M, Lauermann JL, Alnawaiseh M, Eter N. Using deep learning in automated detection of graft detachment in descemet membrane endothelial keratoplasty: a pilot study. Cornea. 2019; 38(2): 157. [CrossRef] [PubMed]
Patefield A, Meng Y, Airaldi M, et al. Deep learning using preoperative AS-OCT predicts graft detachment in DMEK. Transl Vis Sci Technol. 2023; 12(5): 14. [CrossRef] [PubMed]
Nongpiur ME, Haaland BA, Friedman DS, et al. Classification algorithms based on anterior segment optical coherence tomography measurements for detection of angle closure. Ophthalmology. 2013; 120(1): 48–54. [CrossRef] [PubMed]
Sorkhabi MA, Potapenko IO, Ilginis T, Alberti M, Cabrerizo J. Assessment of anterior uveitis through anterior-segment optical coherence tomography and artificial intelligence-based image analyses. Transl Vis Sci Technol. 2022; 11(4): 7. [CrossRef] [PubMed]
Li G, Aldave AJ, Amescua G, et al. Design and biocompatibility of a novel, flexible artificial cornea. Transl Vis Sci Technol. 2024; 13(5): 19. [CrossRef] [PubMed]
Goins KM, Kitzmann AS, Greiner MA, et al. Boston type 1 keratoprosthesis: visual outcomes, device retention, and complications. Cornea. 2016; 35(9): 1165–1174. [CrossRef] [PubMed]
Interlenghi M, Sborgia G, Venturi A, et al. A radiomic-based machine learning system to diagnose age-related macular degeneration from ultra-widefield fundus retinography. Diagn Basel Switz. 2023; 13(18): 2965.
Grzybowski A, Brona P, Lim G, et al. Artificial intelligence for diabetic retinopathy screening: a review. Eye Lond Engl. 2020; 34(3): 451–460.
Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016; 316(22): 2402–2410. [CrossRef] [PubMed]
Yan Q, Jiang Y, Huang H, et al. Genome-wide association studies-based machine learning for prediction of age-related macular degeneration risk. Transl Vis Sci Technol. 2021; 10(2): 29. [CrossRef] [PubMed]
Hayashi T, Tabuchi H, Masumoto H, et al. A deep learning approach in rebubbling after Descemet's membrane endothelial keratoplasty. Eye Contact Lens. 2020; 46(2): 121–126. [CrossRef] [PubMed]
Akpek EK, Aldave AJ, Amescua G, et al. Prognostic significance of biointegration at the optic-cornea joint in keratoprosthesis implantation. Cornea. 2024; 44: 628–636. [PubMed]
Figure 1.
 
Schematic (left) and external appearance (right) of the intrastromal synthetic cornea device. The diagram is drawn to scale and all dimensions are expressed in mm.
Figure 1.
 
Schematic (left) and external appearance (right) of the intrastromal synthetic cornea device. The diagram is drawn to scale and all dimensions are expressed in mm.
Figure 2.
 
Anterior segment optical coherence tomography image, taken after partial thickness trephination, followed by dissection of the lamellar pocket and prior to full-thickness trephination, depicting the method used to determine baseline anterior corneal lamella thickness value. After a 4 mm diameter central trephination is performed at \(\frac{2}{3}\) to \(\frac{3}{4}\) stromal depth (red arrow), an intra-stromal pocket with an 8 mm radius is created distally in the recipient cornea. The anterior lamella (green arrow) is identified relative to the trephination site and the measured tissue thickness serving as the baseline (A). The series of white and green points depict the automated point detection, identifying the top and bottom surfaces of the anterior lamella, with each of the three-point groups fitted to a circle segment. Five measurements (blue indicated circles), at 0.7 mm peripheral to the 4 mm diameter trephination site (B) where the optic-cornea joint resides, were taken and averaged to yield the anterior lamella thickness.
Figure 2.
 
Anterior segment optical coherence tomography image, taken after partial thickness trephination, followed by dissection of the lamellar pocket and prior to full-thickness trephination, depicting the method used to determine baseline anterior corneal lamella thickness value. After a 4 mm diameter central trephination is performed at \(\frac{2}{3}\) to \(\frac{3}{4}\) stromal depth (red arrow), an intra-stromal pocket with an 8 mm radius is created distally in the recipient cornea. The anterior lamella (green arrow) is identified relative to the trephination site and the measured tissue thickness serving as the baseline (A). The series of white and green points depict the automated point detection, identifying the top and bottom surfaces of the anterior lamella, with each of the three-point groups fitted to a circle segment. Five measurements (blue indicated circles), at 0.7 mm peripheral to the 4 mm diameter trephination site (B) where the optic-cornea joint resides, were taken and averaged to yield the anterior lamella thickness.
Figure 3.
 
Manual mark-up (A) and artificial intelligence-detected (B) area of anterior lamellar tissue, adjacent to the optic wall of the device (red arrow indicates the device optic and the green arrow indicates the skirt). Because the area includes the epithelium and stroma as well as the porous ingrowth material laid over the optic wall and skirt of the device, the thickness of the ingrowth surface (50 microns) was then subtracted to yield tissue only. The scale of measurements is depicted in image (C), with the proximal (250 microns, marked in blue) and total (1000 microns marked in white) areas of measurement.
Figure 3.
 
Manual mark-up (A) and artificial intelligence-detected (B) area of anterior lamellar tissue, adjacent to the optic wall of the device (red arrow indicates the device optic and the green arrow indicates the skirt). Because the area includes the epithelium and stroma as well as the porous ingrowth material laid over the optic wall and skirt of the device, the thickness of the ingrowth surface (50 microns) was then subtracted to yield tissue only. The scale of measurements is depicted in image (C), with the proximal (250 microns, marked in blue) and total (1000 microns marked in white) areas of measurement.
Figure 4.
 
Tomographic color plot illustration of peri-prosthetic corneal tissue map showing tissue thickness at 1000 microns radial sections from the optic, overlying the entire skirt (A). The thin inner circle indicates the 250 microns radial sections immediately adjacent to the optic. (B) Magnified view of the 250 microns radial section, considered most relevant to the anatomic outcomes and post-surgical retention, showing the tissue thickness in more detail. All color plots were aligned post-imaging, according to the paracentesis suture at the 12 o'clock position as a reference point.
Figure 4.
 
Tomographic color plot illustration of peri-prosthetic corneal tissue map showing tissue thickness at 1000 microns radial sections from the optic, overlying the entire skirt (A). The thin inner circle indicates the 250 microns radial sections immediately adjacent to the optic. (B) Magnified view of the 250 microns radial section, considered most relevant to the anatomic outcomes and post-surgical retention, showing the tissue thickness in more detail. All color plots were aligned post-imaging, according to the paracentesis suture at the 12 o'clock position as a reference point.
Figure 5.
 
A collage of sequential artificial intelligence (AI) aided anterior lamellar thickness color tomography plots (top), actual anterior segment optical coherence tomography (AS-OCT) images (middle), and clinical pictures (with and without fluorescein staining) (bottom) starting from 6 months through 10 months postoperative. A focal area of thinning in peri-prosthetic tissue immediately adjacent to the optic, which could easily be missed during clinical examination, is evident on the color tomography plot (250 microns) and AS-OCT image as early as month 7.
Figure 5.
 
A collage of sequential artificial intelligence (AI) aided anterior lamellar thickness color tomography plots (top), actual anterior segment optical coherence tomography (AS-OCT) images (middle), and clinical pictures (with and without fluorescein staining) (bottom) starting from 6 months through 10 months postoperative. A focal area of thinning in peri-prosthetic tissue immediately adjacent to the optic, which could easily be missed during clinical examination, is evident on the color tomography plot (250 microns) and AS-OCT image as early as month 7.
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×