Moderator
Clara C. Chan, MD, FRCSC
Panelists
Marcony R. Santhiago, MD, PhD
Larissa Gouvea, MD
Viewing Papers
Expand a paper title to the right to view the paper abstract and authors. Use the video link to jump to that poster in the session.
Presenting Author
Yasser Rifay, MD
Purpose
This study evaluates the six-month outcomes of corneal wavefront-guided PRK (CW guided PRK) in the treatment of keratoconus across 467 eyes. The procedure aimed to enhance corneal stability and visual acuity by using a detailed and individualized treatment protocol.
Methods
The study included 467 eyes diagnosed with stable keratoconus. All patients underwent CW guided PRK using the Schwind Amaris laser system. The treatment planning and execution were facilitated by Phenix and ORK CAM software, ensuring precise ablation patterns tailored to each eye's specific topography and wavefront data. A simultaneous cross-linking procedure was performed immediately post-PRK to reinforce corneal stability. Key criteria for inclusion were a residual stromal thickness (RST) over 300 microns and an ablation depth not exceeding 50 microns.
Results
Over six months, significant improvements were observed in both corneal stability and visual acuity among the treated eyes. The majority of patients achieved a reduction in corneal astigmatism and an improvement in uncorrected visual acuity. The treatment maintained a high safety profile, with all procedures adhering to the required residual stromal thickness and ablation depth limitations. Post-operative keratometry values were generally maintained within the desired range, contributing to enhanced visual outcomes and patient satisfaction.
Conclusion
CW guided PRK is effective and safe for keratoconus, yielding significant improvements in visual acuity and corneal stability over six months, with personalized treatment addressing each patient's specific needs.
Presenting Author
Ryan S. Huang, MSc
Co-Authors
Marko Popovic MD, Clara Chan MD, FRCSC
Purpose
Keratoconus (KC) has been linked to atopy, connective tissue disorders, and mechanical trauma from repeated eye rubbing. However, evidence suggesting that certain medications may trigger KC remains limited to case reports. This study aims to investigate associations between drugs and reports of KC utilizing a large pharmacovigilance database.
Methods
Real-world pharmacovigilance data were sourced from the Food and Drug Administration Adverse Event Reporting System (FAERS) database between Q4 2003 and Q1 2024. Disproportionality analyses were performed using OpenVigil 2.1 (Kiel, Germany) data-mining software to compare the likelihood of KC being reported for a specific drug compared to all other drugs in the FAERS database. Reporting odds ratios (RORs) with 95% confidence intervals (CIs) were computed. Positive signals for adverse drug reactions (ADRs) were verified using the lower bound of the 95% CI for the information component (IC) of Bayesian confidence propagation neural network algorithms (threshold: IC025>0).
Results
A total of 189 cases of KC were reported in the FAERS database with an average age of 34.7�15.5 years. Most cases were male 109 (57.7%, n=109) and originated from the United States (39.2%, n=74/189). Progesterone (ROR=125.3, 95%CI=[75.1, 209.2]) and olopatadine (ROR=69.9, 95%CI=[37.0, 132.2]) showed the highest levels of disproportionate reporting of KC. Other notable drugs included isotretinoin (ROR=21.9, 95%CI=[11.9, 40.2), methylphenidate (ROR=45.1, 95%CI=[10.9, 35.0]), spironolactone (ROR=17.4, 95%CI=[10.1, 30.0]), estradiol (ROR=14.4, 95%CI=[7.6, 27.2]), aripiprazole (ROR=13.9, 95%CI=[8.3 23.1]), dupilumab (ROR=13.5, 95%CI=[9.3, 19.6]), and risperidone (ROR=8.7, 95%CI=[4.7, 15.9]).
Conclusion
Our study identified nine drugs associated with disproportionately high reports of KC. There is a pertinent need for increased clinical vigilance in monitoring for early signs of KC in patients prescribed these medications. Healthcare providers should be aware of the potential drug-related associations for KC, especially in high-risk groups.
Presenting Author
Marwan Ghabra, FRCSEd
Co-Authors
John Marshall PhD
Purpose
To evaluate the efficacy of deep, manually or femtosecond-assisted implantation of XENIA decellularized porcine corneal lenticule in treating patients with advanced and moderate keratoconus by reinforcing biomechanically compromised corneas. The study aims to assess the visual and topographic outcomes of this technique.
Methods
Seven previously Cross-linked eyes of 7 KC patients were included in this case series study. A 9-mm posterior-stromal pocket (100 ?m from the corneal endothelium) was manually created using a calibrated and guarded knife or femtosecond-assisted laser. A decellularized, cross-linked porcine corneal lenticule (44 +/-5 microns thickness and 7.5-8.0 mm diameter) was inserted into the pocket. The visual, pachymetric, topographic, tomographic, and corneal higher-order aberrations (HOAs) outcomes were evaluated preoperatively and at least 16 months post-operatively.
Results
At 12 months postoperatively, the mean anterior keratometries (maximum keratometry, flat K, steep K and average K) values significantly decreased (p < 0.05, for all). The mean anterior corneal elevation and posterior corneal elevation values significantly decreased from 31.4 � 18.1�m and 47.2 � 13.13 �m, respectively to 5 � 2.67 �m and 8.6 � 11.88 �m, respectively (p = 0.0149 and 0.0015, respectively). Although not reaching statistically significance, the mean central corneal thickness values increased from 429 � 22.19 �m preoperatively to 462.8 � 21.89 �m postoperatively (p = 0.1496).
Conclusion
GHABRA technique of posterior corneal stromal XENIA implantation with pocket creation via manual delamination/ femtosecond laser in combination with CXL is safe and results in significant improvements.This novel technique for implants may become standard surgical treatment for keratoconus that addresses the disease and its visual consequences.
Presenting Author
Kangjun Li, MD
Purpose
To investigate the capability of ChatGPT for forecasting the conversion from keratoconus suspect patients (KCs) to keratoconus (KC).
Methods
The clinical data, corneal tomographic, biomechanics, optical density, and confocal microscopy image parameters were selected 2 year before KC development and queries were developed subsequently by converting digitized parameters into textual format based on both eyes of all patients. ChatGPT 4.0 and 4o models were used and compared for all subjects in forecasting conversion from KCs to KC based on various objective metrics. The ChatGPT�s �memory� function was used and the predictive accuracy, recall rate, weighted F1 score, and AUC were evaluated for both models.
Results
ChatGPT4.0 demonstrated an accuracy of 80%, AUC of 0.71, sensitivity of 66%, specificity of 82%, and weighted F1 score of 0.82 in predicting conversion to KC 2 year before onset. ChatGPT4o provided an accuracy of 65%, AUC of 0.63, sensitivity of 61%, specificity of 68%, and weighted F1 score of 0.69 in predicting conversion to KC 2 year before onset.
Conclusion
ChatGPT4 shows superior performance in predicting KC development 2 year before onset, significantly outperforming GPT4o. LLMs hold significant potential for KC predictions. The continued refinement of LLMs, integrating long-term memory and data input for clinical applications, shows great promise and deserve further investigation.
Presenting Author
Suzanne Kirk, MD
Co-Authors
William Wiley MD, Abinaya Thenappan MD, Sudhinder Koushik MD, Mitch Ibach OD, John Berdahl MD
Purpose
To evaluate visual acuity outcomes and corneal tomography changes in patients with keratoconus undergoing corneal tissue addition keratoplasty (CTAK).
Methods
A retrospective analysis of 20 eyes from patients who underwent CTAK surgery for keratoconus. The procedure involved creating a femtosecond laser channel in the host cornea, followed by the implantation of custom-designed preserved corneal tissue. Outcomes measured include uncorrected distance visual acuity (UDVA) at 1 day, 1 week, and 1 month postoperatively. Additional outcomes include changes in inferior corneal steepening as assessed by corneal tomography. This study has been expanded to include additional data for more meaningful results.
Results
Initial analysis of 31 eyes shows preop mean UDVA at 0.94 LogMAR (20/175) at distance, improving to 0.70 LogMAR (20/100) at week 1 and 0.63 LogMAR (20/86) at month 1. Mean pre-operative CDVA was 0.35 LogMAR (20/45) and improved to 0.16 LogMAR (20/29) by post-op month 1. Pre-operative mean K on corneal tomography averaged 54.62 D, improving to 52.31 D by 1 month postoperatively. Statistical significance will be evaluated.
Conclusion
Preliminary results suggest that CTAK offers patients with keratoconus a promising new surgical option, demonstrating rapid visual recovery and reduced inferior corneal steepening, which could enhance functional vision.
Presenting Author
Anitha Venugopal, MBBS, DNB
Co-Authors
Meenakshi Ravindran DO, DNB, Nambi Nallasamy MD
Purpose
To analyse the asymmetry and severity of keratoconus across different age groups of individuals with Down Syndrome (DS), utilizing ocular tomographic parameters to evaluate the distribution and progression of the disease.
Methods
This cross-sectional study included 96 individuals with DS; aged 3 to 40 years (mean 15.73 � 7.33 years). The cohort consisted of 55 males (57.3%) and 41 females (42.7%). Ocular tomography was performed on both the right (52.1%) and left eyes (47.9%). The severity of keratoconus was classified using the Amsler-Krumeich grading system.
Results
Keratoconus varied within the cohort: 6.2% had advanced keratoconus (ADKC), 2.1% forme fruste keratoconus (FFKC), 18.7% keratoconus (KC), 40.6% keratoconus suspects (KCS), and 32.3% showed no keratoconus. Scissor reflex was observed in 5.2% of cases. Mean astigmatism was 2.37D � 2.34, with median visual acuity at 0.39 Log MAR (IQR: 0.18-0.60). Staging showed 32.2% at Stage I, 42.7% at Stage II, 11.5% at Stage III, and 13.5% at Stage IV. Corneal thickness and keratometry readings significantly varied, showing increased disease severity and ocular asymmetry with progression. Notable worsening tomographic metrics included ART Max, I-S value, and higher order aberrations (HOA).
Conclusion
The study reveals variability in keratoconus severity among individuals with Down syndrome, emphasizing the need for early management to prevent visual impairment. Challenges in visual assessments, particularly in advanced cases, underscore the importance of ocular tomography for managing keratoconus in this vulnerable group.
Presenting Author
Marcony R. Santhiago, MD, PhD
Co-Authors
Larissa Stival MD, PhD, Juliana Santos MD, Claudia Morgado MD
Purpose
To investigate the relationship of inflammatory biomarkers with corneal epithelial quantifiable metrics in patients with keratoconus and in healthy eyes.
Methods
This prospective observational comparative study included 100 eyes of 100 patients, 48 eyes of 48 patients with KC, and 52 healthy eyes of 52 healthy controls. The concentrations of tear cytokines were investigated in both groups: interleukin (IL) 1B, IL6, IL8, IL10, IL12p70 and TNF? were obtained by capillary flow and measured using flow cytometer. Cortisol concentrations were determined from the most proximal hair segment as an index of cumulative secretion and measured by liquid chromatography mass spectrometry. Epithelial variables were obtained with optical coherence tomography (OCT). Pearson correlation (r) was used to measure linear dependence between two different variables.
Results
Eyes with keratoconus presented statistically significantly higher levels of IL1b (p=0.02), IL6 (p <0.0001), il8="" (p=""><0.0001), and="" tnf?="" (p=""><0.0001) and="" hair="" cortisol="" concentration="" (p="0.01)" compared="" to="" healthy="" controls.="" there="" was="" a="" significant="" correlation="" between="" il6="" and="" measurement="" epithelium="" min-max="" [pearson="-0.59" (-0.69,="" -0.47);=""><0.0001] and="" epithelial="" standard="" deviation="" (std="" dev)="" [pearson="+0.56" (0.44,="" 0.67);=""><0.0001]. there="" was="" a="" significant="" correlation="" between="" hair="" cortisol="" concentration="" and="" epithelium="" min-max="" [pearson="-0.27" (-0.42,="" -0.1);=""><0.0001] and="" epithelium="" std="" dev="" groups="" [pearson="+0.2" (0.03,="" 0.36);="" p="">
Conclusion
The higher concentration of inflammatory markers (IL6 and hair cortisol) in eyes with keratoconus present a significant correlation with OCT metrics identifying epithelial variability, such as Epithelial Min-Max and Std Dev. These findings demonstrate that epithelial changes detectable with OCT are sensitive to this inflammatory process.
Presenting Author
Marcony R. Santhiago, MD, PhD
Co-Authors
Felipe Taguchi MD, Claudia Morgado MD, Nicole Larivoir MD, Lucas Orlandi de Oliveira PhD, Larissa Stival MD, PhD
Purpose
To test artificial intelligence unsupervised algorithms to sort examinations from an unlabeled corneal epithelium map patterns database into different useful diagnostic clusters, without human intervention.
Methods
Corneal epithelium maps were captured using OCT, all examinations were anonymized and exported using the PachymetryWide report. Images were cropped to include only the epithelium thickness map and applied as input for a convolutional neural network (CNN) with pre-trained weights for general image classification tasks (VGG-16). The artificial intelligence (AI) method CNN was used for feature extraction, generating a vector for each examination, which was analyzed by a dimensionality reduction algorithm, Principal Component Analysis, flattening the data to a 100 dimensions vector. An AI unsupervised learning clustering algorithm was then employed to partition the data set into clusters.
Results
The applied AI model was able to group patterns into 4 cluster groups. A total of 912 images were divided into four groups with 90, 310, 393, and 119 images. The sum of squared errors (SSE) and the Silhouette score were determined. Silhouette AI derived number was 0.23038744926452637. Furthermore, a cluster plot was also created by plotting the three most representative components (explained variances of 27.15%, 18.86%, and 8.95%) from the 100 dimensions (explained variance of 93.34%) in order to study the spatial distribution of these clusters in the data set.
Conclusion
Unsupervised machine learning has revealed an efficient method for clustering large unlabeled databases of epithelial thickness maps into groups of useful patterns. These findings allow differentiation repeated normal from non-normal patterns, significantly increasing diagnostic capabilities beyond what is recognized by human analysis.
Presenting Author
Alexander Kumar, BSc
Co-Authors
Tushar Dave MD, Kashif Baig MD, MBA, Saama Sabeti FRCSC, MPH, MD
Purpose
To develop and validate a novel clinical decision support tool by training machine learning algorithms to predict best corrected visual acuity based on Pentacam corneal tomography data of patients with keratoconus.
Methods
This retrospective study includes 903 keratoconus patients who underwent Pentacam imaging and BCVA testing at Precision Cornea Centre between February 2020 and May 2024. Eyes with clinically diagnosed keratoconus were included, while those with visually significant corneal abnormalities, retinal pathologies or prior corneal surgery were excluded. Machine learning models will be trained and validated to perform classification analysis, predicting BCVA from Pentacam data. The primary outcome will be the predictive performance (sensitivity, specificity, accuracy, and AUC) of the final algorithm, which will be integrated into a web-based clinical decision support tool.
Results
The best-performing model utilized ResNet-18 to extract features from multiple map types, combined with Pentacam summary data and a novel symmetry quantification metric, before regression with the XGBoost algorithm. This model predicted the BCVA with scleral lenses within a mean absolute error (MAE) of 0.0476 logMAR and an R� value of 0.3869 on the test dataset. Maps representing anterior chamber depth, elevation relative to best fit sphere, and pachymetry created features with the highest predictive importance.
Conclusion
This model identified novel relationships between Pentacam data and BCVA to predict patients� vision potential with scleral lenses. Accurate prediction of BCVA in keratoconus patients provides clinicians, especially those without access to scleral lens testing, with valuable data to enhance the risk-benefit analysis for surgical interventions.
Presenting Author
Barbara Dutra, MD, MS
Co-Authors
Bassel Hammoud MS, MD, Giuliano Scarcelli PhD, William Dupps MS, PhD, MD, James Randleman MD
Purpose
To compare regional epithelial thickness between normal controls, eyes with subclinical KC, and eyes with early keratoconus to evaluate the utility of automated epithelial thickness metrics to identify subclinical KC.
Methods
Retrospective analysis of 200 eyes from 200 patients., including 100 control eyes from 100 patients with bilaterally normal corneal topography/tomography and slit-lamp examination (controls), 50 eyes from 50 patients with subclinical keratoconus (SKC) in the evaluated eye, and 50 eyes from 50 patients with manifest keratoconus. Only one eye per patient was used for analysis in all the groups. Epithelial mapping was performed using anterior segment optical coherence tomography (AS-OCT) imaging (Avanti RTVue XR version 2018.1.1.63; Optovue).
Results
The study included 200 eyes (117 right, 83 left) from 200 patients, including 100 normal eyes from 100 patients with bilaterally normal exams (Controls), 50 eyes from 50 patients with SKC, and 50 KC eyes. The mean of corneal epithelial thickness in the 9-mm diameter in 25 sectors, including a central 2-mm zone and paracentral (2 to 5 mm), mid-peripheral (5 to 7 mm), and peripheral (7 to 9 mm). The results of the Kruskal-Wallis analysis of variance demonstrated there were no differences between normal control groups in any OCT parameter for total, epithelial, or stromal data (p>0.05 for all). There were weak, nonsignificant correlations found in regional epithelial thickness between N and KC.
Conclusion
Although it is generally accepted that the rate of change in corneal curvature could affect the local variability in epithelial thickness, this pattern is not useful as a primary metric to identify subclinical keratoconus. Further, epithelial remodeling does not appear to be consistent in early keratoconus.
Presenting Author
Esteban Peralta, MD
Co-Authors
Katherine Peters MD, Cason Robbins MD, Kevin Jackson MD, Lloyd Williams MS, MD, PhD, Anthony Kuo MD
Purpose
The prevalence of keratoconus is increasing worldwide, and improved access to KCN screening is imperative as many treatments for keratoconus (KCN) are limited to the early stages of disease. This study aims to validate SmartKC (Microsoft), a low-cost, smartphone-based portable corneal topographer in the screening of KCN.
Methods
SmartKC uses a 3-D printed placido disc, LEDs, a smartphone camera, and an open-source image processing pipeline to construct axial and tangential topographical heatmaps and quantitative curvature estimates (SimKs) of the anterior corneal surface. This is a prospective study of 60 patients with KCN and normal controls seen at the Duke Eye Center between Oct 2023 and August 2024. Patients with known corneal comorbidities or prior ocular surgery were excluded. All eyes underwent imaging with SmartKC, Oculus Pentacam (Scheimpflug tomography), and Zeiss Atlas (computerized corneal topography). Correlation of SimK values and other relevant parameters between devices were measured.
Results
88 patients (178 eyes) were enrolled. 68/178 Placido ring images were processed by the SmartKC open-source software. Axial and tangential maps were generated along with estimated sim-K1 and sim-K2 values. SmartKC images demonstrated low sensitivity and specificity in detecting keratoconus.
Conclusion
At its current stage of development, the SmartKC prototype has not been validated as a screening device for keratoconus. Further research and development are needed to enhance the performance of this portable corneal topographer before it can be reliably deployed in clinical settings.
Presenting Author
Pooja Khamar, MD, PhD
Co-Authors
Rohit Shetty FRCS
Purpose
Keratoconus (KC) is a progressive inflammatory disorder causing corneal thinning and distortion. This study investigates the efficacy of immunomodulators (IMLs) in stabilizing KC and delaying the need for surgical intervention.
Methods
This is a prospective, randomized, controlled study that included patients with mild to moderate KC showing documented progression. Group 1 (165 eyes) received placebo treatment, while Group 2 (170 eyes, no allergy or eye rubbing) received immunomodulators(IML) - trehalose 3% eye drops combined with sodium hyaluronate. Corneal topography, epithelial mapping, and stromal elevation were assessed using MS-39 (CSO). Collagen and Bowman's layer imaging was performed using polarization-sensitive optical coherence tomography (PSOCT) before and 8 months after treatment.
Results
Group 1 (placebo, 129 of 165 eyes) showed continued keratoconus progression, with Kmax increasing from 52.12�5.8 to 53.01�6.1, along with collagen disorganization and Bowman's layer thinning on PSOCT. Significant reduction in corneal and epithelial thickness were observed in Group 1. Group 2 (Trehalose based treatment, 154 of 170 eyes) showed a significant Kmax reduction (52.28�6 to 51.25�6, p<0.05) and="" improved="" collagen="" orientation="" with="" no="" bowman's="" thinning.="" although="" corneal="" thickness="" remained="" unchanged,="" group="" 2="" maintained="" better="" disease="" stability="" compared="" to="" group="">
Conclusion
IML treatment shows promise in managing inflammation and stabilizing early-stage KC, potentially delaying the need for surgical intervention. While corneal curvature and thickness changes were minimal, the treatment appears effective in maintaining disease stability.
Presenting Author
Rohit Shetty, FRCS
Co-Authors
Pooja Khamar MD, PhD, Abhijit Roy PhD
Purpose
This study aims to assess whether thin corneas (<480 �m)="" are="" indicative="" of="" keratoconus="" by="" analyzing="" corneal="" tomography="" and="" collagen="" distribution.="" the="" objective="" is="" to="" determine="" if="" thin="" corneas="" always="" correlate="" with="" keratoconus="" and="" to="" evaluate="" the="" potential="" of="" advanced="" imaging="" techniques="" in="" diagnosing="" this="">
Methods
150 eyes from 75 patients, with a healthy axial and tangential curvature with absence of Posterior elevation but thin corneas (<480 �m)="" deemed="" unsuitable="" for="" lasik/smile,="" were="" studied.="" imaging="" techniques="" including="" polarization-sensitive="" optical="" coherence="" tomography="" (ps-oct),="" corvis="" st="" for="" corneal="" biomechanics,="" and="" pentacam="" for="" corneal="" topography="" were="" utilized.="" an="" ai="" model="" was="" employed="" to="" evaluate="" preoperative="" corneal="" layers="" and="" collagen="" distribution,="" with="" scores="" ranging="" from="" 0="" (healthy)="" to="" 1="" (clinical="">
Results
Patients were divided into two groups based on preoperative tomography-based indices (TBI): Group 1 (0.45�0.01) and Group 2 (0.85�0.14, P<0.001). despite="" higher="" risk="" indicators="" in="" group="" 2,="" ai="" scores="" for="" both="" groups="" remained="" below="" the="" keratoconus="" threshold=""><0.5). this="" suggests="" that="" thin="" corneas="" are="" not="" always="" indicative="" of="" keratoconus,="" as="" assessed="" by="" the="" ai="">
Conclusion
The study highlights that thin corneas do not necessarily indicate keratoconus and a crosslinking may be avoided. The AI model based on PS-OCT and other imaging techniques shows potential for accurately assessing corneal health and diagnosing keratoconus in thin corneas.
Presenting Author
Nambi Nallasamy, MD
Co-Authors
Binh Duong Giap PhD, Jefferson Lustre BA, Keely Likosky BSc, Joshua Ong MD, Anitha Venugopal MBBS, DNB
Purpose
The incidence of keratoconus (KCN) in patients with Down syndrome (DS) is relatively high, and eye rubbing has been suggested as a contributing factor to the progression of KCN. This study aims to develop and validate an AI-powered system to detect eye-rubbing behavior using sensor data collected from wearable devices.
Methods
Motion sensor data, including 3-axis rotation, gravity, acceleration, and quaternion, was recorded from 10 subjects performing five non-eye-rubbing (NER) activities and eye-rubbing (ER) behaviors over 30 seconds at a frequency of 100 Hz. The data was collected using wearable devices on both the dominant and non-dominant wrists. A total of 1,680 samples, each with twelve channels, was obtained by applying a sliding window of 250 timepoints on each recorded timeseries of 3,000 timepoints. A deep neural network (DNN) comprising two 1D convolutional layers followed by two fully connected layers was developed to classify the samples into NER and ER classes.
Results
The DNN model was trained on 80% of the dataset (7 subjects with 840 windows) for 100 epochs with a batch size of 16, using the Adam optimizer with default parameters and an initial learning rate of 0.001. The trained model was evaluated on the remaining 20% of the dataset (2 subjects with 228 windows) using 5-fold cross-validation. It achieved high classification performance, with an average AUC of 96.23% (�3.48%) and an F1-score of 93.55% (�3.53%).
Conclusion
The proposed AI-powered analysis system effectively detected eye-rubbing behaviors using 12-channel motion sensor data from a wrist-worn wearable device. This system could serve as a valuable tool for large-scale analysis of risk factors for keratoconus development and progression in patients with Down syndrome.
To log in, click the teal "Login" button in the upper right-hand corner of this page. If you are logged in but still do not have access, please check your 2025 Annual Meeting registration.
