Abstract Purpose: To determine the incidence and extent of axial length (AL) shortening after 12 months of repeated low-level red-light (RLRL) therapy in highly myopic eyes. In addition, to investigate the relationship between the AL shortening and changes in the posterior scleral curvature in the swept-source optical coherence tomographic (SS-OCT) images. Methods: Thirty participants aged 8 to 18 years with high myopia were studied. The spherical equivalent of refractive error (SER) and AL were measured, and SS-OCT images were recorded at baseline, and at 1, 3, 6, and 12 months after beginning RLRL therapy. Results: Twelve months of RLRL therapy led to an average reduction in myopic SER by 0.29 D (95% CI, 0.10 to 0.49 D; P < 0.05) and stabilization of AL, with a mean change of -0.03 mm (95% CI, -0.09 to 0.04 mm; P = 1.00). AL shortening >0.05 mm was observed in 53.8% of eyes. The average increase in the macular choroidal thickness was 26 µm (95% CI, 20 to 32 µm; P < 0.05), and an apparent change in posterior scleral curvature was observed in 5/26 (19.2%) eyes. One participant discontinued the RLRL therapy on day 2 because of photophobia. None of the other participants reported persistent photophobia or afterimage throughout the study, but one had 2 line reduction in BCVA without any OCT structural change. Conclusions: RLRL therapy was associated with AL stabilization in highly myopic eyes. Although apparent scleral curvature change was observed, it was a post hoc analysis in a small number of eyes and its association with choroidal thickness and the shortening of the AL needs further investigation.
Introduction: During the anti-vascular endothelial growth factor (anti-VEGF) loading phase for neovascular age-related macular degeneration (nAMD), the month-1 follow-up visit represents the earliest standardized imaging time point, yet whether month-1 optical coherence tomography (OCT) and OCT angiography (OCTA) improve prediction of short-term anatomical outcome beyond baseline OCT has not been formally tested. This study quantified the incremental predictive value of month-1 imaging using a nested-model framework that mirrors the sequential clinical workflow. Methods: This retrospective, single-center cohort included 124 treatment-naive nAMD eyes from 105 patients that completed three monthly anti-VEGF injections and had evaluable baseline, month-1, and month-3 imaging. Poor anatomical outcome was defined as residual intraretinal fluid and/or subretinal fluid at month 3. Nested generalized estimating equation logistic models accounted for inclusion of bilateral eyes and evaluated baseline OCT variables, the incremental contribution of month-1 structural OCT, and the additional contribution of month-1 OCTA-derived macular neovascularization (MNV) area. Discrimination was assessed using area under the receiver operating characteristic curve (AUC), with ΔAUC confidence intervals estimated by patient-level cluster bootstrap. Results: At month 3, 47 of 124 eyes (37.9%) had residual exudative activity. In the final model, month-1 central subfield thickness (OR 2.065 per 100-µm increase, 95% CI 1.107–3.853), residual month-1 intraretinal fluid (OR 14.623, 95% CI 3.305–64.703), residual month-1 subretinal fluid (OR 6.975, 95% CI 1.963–24.785), and month-1 MNV area (OR 1.130 per 1-mm² increase, 95% CI 1.030–1.240) were associated with poor month-3 anatomical outcome. AUC improved from 0.742 (95% CI 0.658–0.822) with baseline OCT alone to 0.889 (95% CI 0.828–0.940) after adding month-1 structural OCT (ΔAUC 0.147, 95% CI 0.072–0.226), and to 0.908 (95% CI 0.853–0.952) after adding month-1 MNV area. Baseline-to-month-1 quantitative changes were not associated with outcome. Conclusion: Month-1 structural OCT substantially improved prediction of month-3 residual exudative activity beyond baseline OCT alone, whereas OCTA-derived MNV area contributed modest additional discrimination. These findings identify the first post-treatment imaging visit as an informative point for short-term anatomical risk assessment, but external validation is required before clinical application.
Purpose: To examine the pupillary light reflex (PLR) in Immunoglobulin G4-related ophthalmic disease (IgG4-ROD) compared with healthy controls using virtual reality–based infrared pupillometry. Design: Cross-sectional matched case-control study. Participants: Twenty-five eyes from 25 biopsy-proven IgG4-ROD patients without a history of optic neuropathy and 50 eyes from 50 age-, sex-, and smoking-matched healthy subjects were included. Methods: Participants underwent PLR assessment using a virtual reality head-mounted display with integrated infrared pupillometry. Pupillary responses to three progressively increasing white-light stimuli were recorded at 120 Hz. Twenty-seven PLR parameters were extracted across response latency, constriction, and redilation phases. Group differences were analyzed using paired t-tests, multinomial logistic regression, and backward stepwise variable selection. Machine learning classifiers were evaluated using accuracy and area under the receiver operating characteristic curve (AUC). Main Outcome Measures Pupil constriction latency, constriction amplitude, velocity, redilation parameters, and machine learning classification performance metrics. Results: Pupil constriction latency at the third (highest intensity) stimulus (LAT3rd) was significantly prolonged in IgG4-ROD eyes compared with controls (P < 0.05), while other PLR parameters were comparable between groups. LAT3rd remained a significant differentiating feature on multinomial logistic regression and backward stepwise selection. Machine learning models using all features showed good performance, with Random Forest achieving an accuracy/AUC of 0.875/0.900. When using bootstrapped LAT3rd-only data, K-nearest neighbors, decision tree, and random forest models also demonstrated good classification performance. Conclusion: IgG4-ROD eyes exhibit prolonged pupil constriction latency, suggesting subclinical involvement of the pupillary light reflex pathway. Virtual reality–based infrared pupillometry may serve as a useful non-invasive tool for detecting neural or autonomic dysfunction in IgG4-ROD.
Introduction: To systematically rate the quality of online educational websites focused on keratoconus. Methods: Google, Yahoo and Bing were queried for the keywords “Keratoconus”, “Keratoconus treatment”, “Keratoconus surgery”. For each keyword, the first 50 websites were assessed for quality of information using the Modified Ensuring Quality Information for Patient (mEQIP) scale. English websites intended for non-professional public were included while forums, blogs and scientific articles were excluded. Websites with a score above the 75th percentile were labelled as high-score, while those below were classified as low-score. Statistically significant values were considered if p < 0.05. Results: Overall, 119 eligible websites were included and classified in 5 groups based on their source: healthcare portals (n=36, 30.25%), practitioners (n=48, 40.34%), hospitals (n=21, 17.65%), professional societies (n=13, 10.92%) and encyclopedias (n=1, 0.84%). The mean mEQIP score was 20.03 ŷ 4.13. Websites with a score of 23 or higher were labelled as high score, websites below 23 were classified as low score. High-score websites (27.73%) provided more comprehensive details about treatments, medical procedures, benefits, and risks, and were supported by scientific references. Conversely, low-score websites (72.27%) often lacked essential information and quantitative data. There was no significant difference in the quality of information for webpages published before and after COVID-19 pandemic (P = .79), or between different types of sources (P = .80). Conclusions: Overall, this study indicates moderate quality of online information regarding keratoconus and underscores the need for improvements of online health resources.
BACKGROUND:Knowledge graph (KG) is an artificial intelligence technique that provides a structured representation of medical entities and their relationships, thereby facilitating integration of heterogeneous information, knowledge discovery, and intelligent reasoning. The process of KG construction includes knowledge acquisition, knowledge extraction, knowledge fusion, knowledge inference, KG visualization, and KG evaluation. In ophthalmology, KGs have demonstrated significant potential in advancing disease understanding, supporting clinical decision-making, assisting in ophthalmic image analysis, and facilitating clinical intelligent question-answering. SUMMARY:This paper reviews the methodologies for constructing medical KGs and highlights their applications in ophthalmology, with particular emphasis on the integration of ophthalmic KGs with medical imaging and large language models. Furthermore, it discusses existing challenges - ranging from privacy and regulatory constraints to high construction and maintenance costs, limited fusion of imaging and multimodal data, insufficient coverage of rare eye diseases and insufficient application in education and basic research - aiming to provide insights that promote deeper research and clinical translation of ophthalmology KGs. KEY MESSAGES:The integration of KGs with multimodal technologies represents a research frontier in the transformation of ophthalmic clinical practice, enabling more structured and precise diagnostic reasoning. Future research should focus on the construction of domain-specific KGs for rare ocular diseases and spearhead the development of multimodal knowledge frameworks to fill existing knowledge gaps, thereby ultimately enhancing both patient clinical outcomes and medical education.
