Creating realistic anterior segment optical coherence tomography images using generative adversarial networks.

AIMS: To develop a generative adversarial network (GAN) capable of generating realistic high-resolution anterior segment optical coherence tomography (AS-OCT) images. METHODS: This study included 142 628 AS-OCT B-scans from the American University of Beirut Medical Center. The Style and WAvelet based GAN architecture was trained to generate realistic AS-OCT images and was evaluated through the Fréchet Inception Distance (FID) Score and a blinded assessment by three refractive surgeons who were asked to distinguish between real and generated images. To assess the suitability of the generated images for machine learning tasks, a convolutional neural network (CNN) was trained using a dataset of real and generated images over a classification task. The generated AS-OCT images were then upsampled using an enhanced super-resolution GAN (ESRGAN) to achieve high resolution. RESULTS: The generated images exhibited visual and quantitative similarity to real AS-OCT images. Quantitative similarity assessed using FID scored an average of 6.32. Surgeons scored 51.7% in identifying real versus generated images which was not significantly better than chance (p value >0.3). The CNN accuracy improved from 78% to 100% when synthetic images were added to the dataset. The ESRGAN upsampled images were objectively more realistic and accurate compared with traditional upsampling techniques by scoring a lower Learned Perceptual Image Patch Similarity of 0.0905 compared with 0.4244 of bicubic interpolation. CONCLUSIONS: This study successfully developed and leveraged GANs capable of generating high-definition synthetic AS-OCT images that are realistic and suitable for machine learning and image analysis tasks.

Enhancing the Automated Detection of Implantable Collamer Lens Vault Using Generative Adversarial Networks and Synthetic Data on Optical Coherence Tomography.

PURPOSE: To investigate the efficacy of incorporating Generative Adversarial Network (GAN) and synthetic images in enhancing the performance of a convolutional neural network (CNN) for automated estimation of Implantable Collamer Lens (ICL) vault using anterior segment optical coherence tomography (AS-OCT). METHODS: This study was a retrospective evaluation using synthetic data and real patient images in a deep learning framework. Synthetic ICL AS-OCT scans were generated using GANs and a secondary image editing algorithm, creating approximately 100,000 synthetic images. These were used alongside real patient scans to train a CNN for estimating ICL vault distance. The model's performance was evaluated using statistical metrics such as mean absolute percentage error (MAPE), mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R2) for the estimation of ICL vault distance. RESULTS: The study analyzed 4,557 AS-OCT B-scans from 138 eyes of 103 patients for training. An independent, retrospectively collected dataset of 2,454 AS-OCT images from 88 eyes of 56 patients, used prospectively for evaluation, served as the test set. When trained solely on real images, the CNN achieved a MAPE of 15.31%, MAE of 44.68 µm, and RMSE of 63.3 µm. However, with the inclusion of GAN-generated and algorithmically edited synthetic images, the performance significantly improved, achieving a MAPE of 8.09%, MAE of 24.83 µm, and RMSE of 32.26 µm. The R2 value was +0.98, indicating a strong positive correlation between actual and predicted ICL vault distances (P < .01). No statistically significant difference was observed between measured and predicted vault values (P = .58). CONCLUSIONS: The integration of GAN-generated and edited synthetic images substantially enhanced ICL vault estimation, demonstrating the efficacy of GANs and synthetic data in enhancing OCT image analysis accuracy. This model not only shows potential for assisting postoperative ICL evaluations, but also for improving OCT automation when data paucity is an issue. [J Refract Surg. 2024;40(4):e199-e207.].

Repeatability of Pyramidal Aberrometer Measurements in Keratoconus and Normal Eyes.

PURPOSE: To evaluate the intrasubject repeatability of pyramidal aberrometer measurements in a sample of keratoconus and normal eyes. SETTING: American University of Beirut Medical Center, Beirut, Lebanon. DESIGN: Prospective comparative repeatability analysis. STUDY POPULATION: Keratoconus and Normal eyes from adult patients. OBSERVATION PROCEDURES: Each eye was evaluated with 3 consecutive acquisitions using a pyramidal aberrometer. MAIN OUTCOME MEASURES: The repeatability of different ocular Higher Order and Lower Order Aberrations (HOAs and LOAs, respectively), and Zernike coefficients down to the fifth order, was evaluated. Repeatability was assessed by within-subject standard deviations (Sw), repeatability limits (r), and intraclass correlation coefficients (ICC), among other parameters. RESULTS: 72 keratoconus patients (72 eyes), and 76 normal patients (76 eyes) were included. In normal and keratoconus eyes, the ICC of total LOA and HOAs, as well as each of the Zernike coefficients, was >0.9. The Sw for keratoconus eyes with mean maximal keratometry (Kmax)<50D was 0.1345 for total LOAs, 0.0619 for total HOAs, 0.0292 for horizontal coma, 0.0561 for vertical coma, and 0.0221 for spherical aberration as compared to 0.2696, 0.1486, 0.0972, 0.1497 and 0.0757 for Keratoconus eyes with Kmax>=50D. Similar trend of better repeatability for grade 1 keratoconus and HOA<2D as compared to grade 2 and 3 keratoconus and eyes with HOA>2D were also noted. CONCLUSIONS: Ocular aberrometer measurements generated by high definition pyramidal aberrometers have high repeatability in both normal and mild keratoconus eyes; and moderate repeatability, yet still clinically acceptable, in advanced keratoconus. This is of particular importance in ocular wavefront-guided treatments.

Accelerated corneal crosslinking with 20'-soaking hydroxypropyl methyl cellulose/riboflavin vs conventional crosslinking with 30'-soaking dextran/riboflavin.

PURPOSE: To evaluate and compare functional and structural outcomes of accelerated corneal crosslinking (A-CXL) using riboflavin with hydroxypropyl methyl cellulose (HPMC) vs conventional corneal crosslinking (C-CXL) using riboflavin with dextran. SETTING: American University of Beirut Medical Center, Beirut, Lebanon. DESIGN: Retrospective analysis. METHODS: Retrospective analysis of 83 eyes of 73 patients with mild to moderate keratoconus. First group (n = 44 eyes) underwent C-CXL using a 30-minute riboflavin/dextran soaking between June 2014 and March 2016. Second group (n = 39 eyes) underwent A-CXL using a 20-minute riboflavin/HPMC soaking between April 2016 and December 2017. Patients were evaluated preoperatively and at 1, 3, and 12 months postoperatively. Main outcome measures were simulated keratometry (simK), maximum axial keratometry (Kmax), demarcation line depth, and haze intensity measured using optical coherence tomography-based image analysis software. RESULTS: Demarcation line (DL) was 298.30 ± 64.60 µm and 335.61 ± 99.76 µm for C-CXL and A-CXL groups, respectively ( P = .04). Haze profile was similar for both groups. The mean simK values were reduced from 46.93 ± 3.50 and 46.44 ± 2.93 preoperatively to 46.18 ± 3.65 and 45.54 ± 2.78 at 12 months postoperatively, for C-CXL and A-CXL, respectively ( P = .003 for both groups). The mean Kmax decreased from 52.46 ± 4.82 and 51.50 ± 3.87 preoperatively to 51.30 ± 4.42 and 50.30 ± 3.52 postoperatively, for the C-CXL and A-CXL, respectively ( P < .001 for both groups). There was no difference in the simK and Kmax changes between the C-CXL and A-CXL groups ( P = .814 and P = .913), visual acuity, and refraction between the 2 groups ( P > .05). CONCLUSIONS: A-CXL with a 20-minute riboflavin/HPMC soaking produced deeper DL and similar corneal haze, topographic, refractive, and visual results to C-CXL with a 30-minute riboflavin/dextran soaking.

Asymmetric All-Femtosecond Laser-Cut Corneal Allogenic Intrastromal Ring Segments.

PURPOSE: To describe a novel technique for cutting asymmetric allogenic segments using the femtosecond laser for the management of cases of keratoconus with non-coinciding astigmatism and coma axes. METHODS: Four eyes of 2 patients with irregular keratoconus and asymmetric allogenic segments were included. Visual, refractive, tomographic, and aberrometric outcomes, and optical coherence tomography (OCT) sections were measured preoperatively and 6 months postoperatively. Evaluations included slit-lamp examination, manifest refraction, uncorrected (UDVA) and corrected (CDVA) distance visual acuity, and simulated and maximum anterior keratometry (Kmax) using anterior segment OCT. RESULTS: Spherical and cylindrical refractive errors decreased from -2.38 ± 2.96 and -2.94 ± 2.16 to -1.81 ± 2.77 (P = .04) and -1.75 ± 2.07 (P = .01) diopters (D), respectively, 6 months postoperatively. There was an average gain of three lines of CDVA. Kmax decreased from 50.02 ± 1.99 to 47.89 ± 3.05 D (P= .03) and coma from 1.05 ± 0.21 to 0.21 ± 0.19 D (P = .01). CONCLUSIONS: Asymmetric femtosecond laser-cut allogenic segments allow a higher level of customization based on size, shape, and arc length, in contrast to the limited range of available synthetic asymmetrical segments. [J Refract Surg. 2023;39(12):856-862.].

Corneal Allogenic Intrastromal Ring Segments (CAIRS) for Corneal Ectasia: A Comprehensive Segmental Tomography Evaluation.

PURPOSE: To evaluate the visual, refractive, and tomographic results of patients with corneal ectasia treated with corneal allogenic intrastromal ring segments (CAIRS) insertion without concomitant corneal cross-linking. METHODS: Fifty-two eyes from 39 patients with stable corneal ectasia and unsatisfactory visual acuity with contact lenses were included. All patients underwent CAIRS insertion with no concomitant corneal procedure at the American University of Beirut Medical Center between September 2019 and July 2022. Visual, refractive, topographic, aberrometric, epithelial, stromal, and segment thickness data were measured relative to baseline at 1 week, 1 month, and at least 3 months postoperatively. Evaluations included slit-lamp examination, manifest refraction, uncorrected (UDVA) and corrected (CDVA) distance visual acuity, and tomography using anterior segment optical coherence tomography. RESULTS: Mean follow-up time was 6.9 ± 5.2 months. UDVA and CDVA improved from 0.97 ± 0.47 and 0.56 ± 0.19 preoperatively to 0.52 ± 0.21 (P < .001) and 0.23 ± 0.19 (P < .001) 3 months postoperatively. Manifest refraction spherical equivalent and cylinder improved from -6.71 ± 6.51 and -4.02 ± 2.24 diopters (D) preoperatively to -3.78 ± 4.07 D (P < .001) and -2.35 ± 1.98 D (P < .001) 3 months postoperatively, respectively. Maximum anterior keratometry and vertical coma decreased from 58.09 ± 7.92 D and 1.56 ± 1.09 µm to 52.48 ± 6.69 D (P < .001) and 0.43 ± 0.77 µm, respectively (P < .001). Corneal epithelium thickened proximal to the allogenic segment by 7.25 µm (P < .001), whereas stromal elevation at the cone decreased from 38.61 ± 18.5 to 23.82 ± 13.4 µm, respectively (P < .001). No major complications were observed and only 1 eye lost one line of CDVA. CONCLUSIONS: Treatment of corneal ectasia with CAIRS improved visual, refractive, topographic, and tomographic parameters. Epithelial thickening central to CAIRS, along with anterior stromal flattening is postulated to contribute to tomographic flattening and regularization. [J Refract Surg. 2023;39(11):767-776.].

Extended Dehydration of Corneal Allogenic Intrastromal Ring Segments to Facilitate Insertion: The Corneal Jerky Technique.

PURPOSE: The aim of this study was to describe a new technique to facilitate the insertion of corneal allogenic intrastromal ring segments. METHODS: A single-segment corneal allogenic intrastromal ring segment (CAIRS) was trephined from donor corneas and allowed to markedly dehydrate for 75 minutes before the start of the procedure with a room humidity of 35% to 45%. The duration of the insertion step and the intrastromal segment size at 1 week as measured by optical coherence tomography were compared with previously performed single-segment CAIRS procedures using the conventional technique. RESULTS: A total of 41 eyes of 36 patients underwent 1-segment CAIRS insertion of the same trephination size (750 µm). Fifteen eyes underwent the conventional insertion procedure, and 26 eyes had a dehydrated segment inserted. The time taken to insert the CAIRS analyzed by surgical video recording starting after the femtosecond tunnel creation and initiation of the insertion to the segment ironing step was 282 ± 103 and 97 ± 23 seconds for the conventional and the dehydrated segment technique, respectively ( P < 0.001). Anterior segment optical coherence tomography performed 1 week postoperatively revealed similar segment thickness and width of 471.3 ± 54.1 µm and 1285.1 ± 191.0 µm for the conventional allogenic segments and 483.4 ± 58.3 µm and 1227.2 ± 165.2 µm for the dehydrated segments ( P = 0.515 and 0.314, respectively). CONCLUSIONS: Markedly dehydrated corneal allogenic segments are easier and faster to insert than the nondehydrated ones while maintaining similar sizes intrastromally. This dehydration technique makes the procedure similar to the one with synthetic segments and hence reduces the learning curve.

Therapeutic visual rehabilitation in a patient with high hyperopia and flat cornea years after radial keratotomy.

A 51-year-old man was referred for refractive surgery evaluation. Spectacle dependence and poor visual quality in both eyes was his chief complaint. He cannot tolerate contact lenses. Corrected distance visual acuity (CDVA) was 20/40 in both eyes. Manifest refraction was +5.25 -2.25 @ 90 (20/40) in the right eye and +6.25 -2.25 @ 105 (20/40) in the left eye. The patient had a history of radial keratotomy (RK) almost 30 years ago in both eyes and at the slitlamp presented 8 RK incisions, proportionally spaced between one another. All incisions were closed, and there were no relevant signs of scarring. The patient denied any history of ocular trauma, systemic disease, or medications. Corneal topography with different technologies revealed an irregular pattern with marked central flattening in both eyes, with some points below 30 diopters (D) (Supplemental Figures 1 and 2, available at http://links.lww.com/JRS/A862 and http://links.lww.com/JRS/A863, respectively). There were no signs of cataract, and fundus examination was normal. Optical coherence tomography (OCT) of the right eye revealed a more homogeneous thickness pattern, little variation between the thinnest and thickest areas, and adequate transparency (Figure 1JOURNAL/jcrs/04.03/02158034-202306000-00018/figure1/v/2023-05-31T172126Z/r/image-tiff). In the left eye, there is wide variability between the thinnest and thickest stromal points, with annular thinning and central thickening (Figure 2JOURNAL/jcrs/04.03/02158034-202306000-00018/figure2/v/2023-05-31T172126Z/r/image-tiff). Both eyes show marked epithelial irregularity. Considering this patient's current ocular status, how would you reach visual rehabilitation? Because he is contact lens intolerant, would you consider surface ablation, for example, photorefractive keratectomy (PRK) with mitomycin-C (MMC)? If that were the case, would you think of an optimized or a topography-guided (TG) treatment? Would you immediately consider a corneal transplant option? Would you instead consider a more conservative approach? Which one and why?

Deep Learning-Based Estimation of Implantable Collamer Lens Vault Using Optical Coherence Tomography.

PURPOSE: To develop and validate a deep learning neural network for automated measurement of implantable collamer lens (ICL) vault using anterior segment optical coherence tomography (AS-OCT). DESIGN: Cross-sectional retrospective study. METHODS: A total of 2647 AS-OCT scans were used from 139 eyes of 82 subjects who underwent ICL surgery in 3 different centers. Using transfer learning, a deep learning network was trained and validated for estimating the ICL vault on OCT. A trained operator separately reviewed all OCT scans and measured the central vault using a built-in caliper tool. The model was then separately tested on 191 scans. A Bland-Altman plot was constructed and the mean absolute percentage error (MAPE), mean absolute error (MAE), root mean squared error (RMSE), Pearson correlation coefficient (r), and determination coefficient (R2) were calculated to evaluate the strength and validity of the model. RESULTS: On the test set, the model achieved a MAPE of 3.42%, an MAE of 15.82 µm, a RMSE of 18.85 µm, a Pearson correlation coefficient r of +0.98 (P < .00001), and a coefficient of determination R2 of +0.96. There was no significant difference between the vaults of the test set labeled by the technician vs those estimated by the model: 478 ± 95 µm vs 475 ± 97 µm, respectively, P = .064). CONCLUSIONS: Using transfer learning, our deep learning neural network was able to accurately compute the ICL vault from AS-OCT scans, overcoming the limitations of an imbalanced data set and limited training data. Such an algorithm can assist the postoperative assessment in ICL surgery.

Effect of fluence levels on prophylactic corneal cross-linking for laser in situ keratomileusis and transepithelial photorefractive keratectomy.

PURPOSE: The purpose of this study is to assess the effect of various fluence levels on prophylactic corneal cross-linking (CXL) combined with femtosecond laser in situ keratomileusis (FS-LASIK-Xtra) or transepithelial photorefractive keratectomy (TransPRK-Xtra) on biomechanics, demarcation line (DL), and stromal haze. METHODS: Prospective analysis where two prophylactic CXL protocols (lower/higher fluence [LF/HF]: 30 mw/cm2 , 60/80 s, 1.8/2.4 J/cm2 ) were performed as part of either an FS-LASIK-Xtra or TransPRK-Xtra procedure. Data were collected preoperatively and at 1 week and 1, 3, and 6 months postoperatively. Main outcome measures were (1) dynamic corneal response parameters and the stress-strain index (SSI) from Corvis, (2) actual DL depth (ADL), and (3) stromal haze on OCT images analysed by a machine learning algorithm. RESULTS: Eighty-six eyes from 86 patients underwent FS-LASIK-Xtra-HF (21 eyes), FS-LASIK-Xtra-LF (21 eyes), TransPRK-Xtra-HF (23 eyes), and TransPRK-Xtra-LF (21 eyes). SSI increased similarly by around 15% in all groups 6 months postoperatively (p = 0.155). All other corneal biomechanical parameters were statistically significant worsening postoperatively, but the change was similar in all groups. At 1 month postoperatively, there was no statistical difference in mean ADL among four groups (p = 0.613), mean stromal haze was similar between the two FS-LASIK-Xtra groups, but higher in the TransPRK-Xtra-HF group compared with the TransPRK-Xtra-LF group. CONCLUSIONS: FS-LASIK-Xtra and TransPRK-Xtra lead to a similar ADL and improve SSI equally. Lower fluence prophylactic CXL might be recommended as it achieves similar mean ADL with potentially less induced stromal haze, especially in TransPRK. The clinical relevance and applicability of such protocols remains to be assessed.

Demarcation Line Depth in Epithelium-Off Corneal Cross-Linking Performed at the Slit Lamp.

We aimed to evaluate the depth of the demarcation line following accelerated epithelium-off corneal cross-linking (A-CXL) performed at the slit lamp with the patient sitting in an upright position. Twenty-three eyes from twenty patients, undergoing epi-off A-CXL (9 mW/cm2 for 10 min) using a CXL device at the slit lamp in the upright position. Demarcation line depth was assessed at 1 month after the procedure using anterior segment optical coherence tomography (AS-OCT) and specialized software. Surgery was uneventful in all cases. The average postoperative demarcation line depth achieved was 189.4 µm (standard deviation: 58.67 µm). The demarcation line depth achieved with patients sitting upright, receiving CXL at the slit lamp, is similar to published data on CXL performed in the supine position, suggesting that demarcation line depth is not dependent on patient orientation during CXL.