This Week in Ophthalmology — Jun 6, 2026

Welcome to This Week in Ophthalmology. This week we're covering 9 notable papers spanning the expanding role of artificial intelligence in diagnostics, new strategies in glaucoma risk assessment, and evolving therapeutics for the anterior and posterior segment. Let's dive in.

This week, artificial intelligence and advanced imaging take center stage, offering new ways to diagnose disease and predict future risk. First, a study in JAMA Ophthalmology demonstrates the power of AI in public health, using a smartphone application for proactive self-screening of ocular surface malignancies [6]. Researchers developed and validated a deep learning model called CaptureTumor, deployed through a mobile app with guided photography instructions. A media campaign reached over 250,000 individuals in China, prompting 614 to complete at-home screening. The app provided immediate risk stratification, and impressively, it led to the histopathological confirmation of 20 malignancies. Nineteen of these were new diagnoses, and none of the patients required enucleation, suggesting the power of early detection. The smartphone-based system achieved an area under the curve of 0.905, nearly matching the performance of a model trained on high-quality slit-lamp images. This study provides a powerful proof-of-concept for a scalable, accessible model for screening rare but serious ocular diseases. Moving from population screening to individual risk prediction, a paper in The British Journal of Ophthalmology uses AI to identify children with high myopia who are most likely to develop pathologic changes [1]. In a prospective study following 375 children with high myopia for a median of 15 years, researchers found that 17% went on to develop pathologic myopia. An AI model analyzing fundus photographs, OCT metrics, and ocular biometry identified several key predictors. These included retinal microvascular changes, such as lower vessel density and reduced fractal dimension, which increased the odds of progression by nearly tenfold and sixfold, respectively. Other factors like a thinner subfoveal choroid and faster rates of axial elongation and choroidal thinning in early life were also significant predictors. The combined AI model achieved excellent predictive accuracy, with an area under the curve of 0.98, and has been implemented in a web-based tool to help clinicians stratify risk and tailor surveillance for these vulnerable patients. The predictive power of imaging extends beyond the eye itself. A study in JAMA Ophthalmology shows that retinal OCT in very preterm infants can forecast neurodevelopmental outcomes two years later [7]. In a cohort of 72 infants born at less than 32 weeks' gestation, researchers measured retinal layer thickness at 36 weeks' postmenstrual age. They found that a greater retinal nerve fiber layer, or RNFL, thickness was significantly associated with better outcomes at age two. Specifically, for every 10-micron increase in RNFL thickness, motor scores on the Bayley Scales were 7.5 points higher and cognitive scores were 3.7 points higher. A thicker RNFL was also linked to lower autism risk scores and fewer internalizing behavior problems. Adding RNFL thickness to a predictive model that included factors like gestational age significantly improved its ability to predict motor and cognitive scores. This highlights the retina as a potential, noninvasive window to brain development, which, if confirmed, could help guide early intervention for high-risk infants. But as AI's capabilities grow, so do concerns about its misuse. A report in Ophthalmology investigated the presence of text generated by large language models in the ophthalmic literature [4]. By analyzing over 27,000 articles, researchers found a marked increase in the use of stylistic words characteristic of AI-generated text since the release of ChatGPT in 2022. Using a validated AI-detection tool on a sample of full articles published by 2025, they found that over 25% of research articles and over 21% of commentary articles had scores suggesting a high likelihood of AI-generated content. Among these high-likelihood publications, a concerning 22% of sentences in research articles and 90% in commentary articles were flagged as likely written by AI. Critically, none of the publications with these outlier scores disclosed the use of AI. This study serves as a crucial warning about the potential for undisclosed AI use to jeopardize authorship integrity and the long-term reliability of scientific literature, underscoring the need for clear disclosure policies.

Turning to glaucoma, two large database studies offer new ways to think about risk and management. For our many patients classified as open-angle glaucoma suspects, a common clinical question is how often to bring them back for follow-up. A study in the American Journal of Ophthalmology provides a data-driven framework to help answer this [3]. Analyzing data from over 83,000 glaucoma suspects in a large United States database, the study found an overall annual conversion rate to primary open-angle glaucoma of 6.1 percent, with the rate being highest in the first year at 9.4 percent. The researchers then stratified patients by risk. The lowest-risk group—those under 50 and not on treatment—had an annual conversion rate of just 2 percent. In contrast, the highest-risk group—those over 70 and on treatment—had an annual conversion rate of nearly 17 percent. Based on these findings, the authors propose a risk-stratified monitoring strategy. For a very low-risk patient, to reach a 5 percent per-visit conversion risk threshold, the follow-up interval could be extended to as long as 6 years, while a high-risk patient would require follow-up every 7 months. This provides a pragmatic evidence base for personalizing surveillance, potentially reducing visit burden for low-risk patients while ensuring timely detection for those at high risk. Another large database study, this one in Investigative Ophthalmology & Visual Science, uncovers a surprising potential link between a common osteoporosis medication and glaucoma risk [8]. Researchers used a multinational database to compare over 47,000 patients with osteoporosis treated with the RANKL inhibitor denosumab to a propensity-score matched group treated with bisphosphonates. Over 5 years, they found that patients taking denosumab had a significantly lower risk of developing both ocular hypertension and primary open-angle glaucoma. Specifically, the risk of developing ocular hypertension was reduced by 28%, and the risk of developing glaucoma was reduced by 37% compared to the bisphosphonate group. The findings were consistent across multiple sensitivity analyses and patient subgroups. While this is a retrospective study and cannot prove causation, it highlights the RANKL pathway as a potentially important area for future investigation in glaucoma pathophysiology and prevention.

Finally, we'll look at innovations in surgical and medical treatments for the cornea and retina. A comprehensive review in Progress in Retinal and Eye Research proposes a new framework for managing corneal ectatic diseases like keratoconus [9]. The authors argue for a paradigm shift away from traditional full-thickness tissue replacement towards more selective, function-restoring procedures. They summarize the evidence for techniques like Bowman layer transplantation, corneal allogenic intrastromal ring segments, and stromal lenticule addition keratoplasty. To guide clinical decision-making, they propose a three-tier 'Stabilize-Reshape-Replace' framework. This pragmatic approach prioritizes stabilizing the cornea first, often with cross-linking; then reshaping it to improve optics, using techniques like stromal addition; and only resorting to tissue replacement as a final step. This conceptual model reflects a broader trend toward minimally invasive, biologically integrated surgery. Building on this theme of tissue-sparing innovation, a pilot study in Cornea introduces a novel device for repairing corneal perforations [5]. Instead of relying on cyanoacrylate tissue adhesive, which can be prone to dislodgement, researchers developed a cufflink-shaped plug made from glycerin-preserved donor corneal tissue. An eye bank technician was able to reliably produce these plugs using a modified microkeratome system. In ex vivo testing on corneoscleral rims, the plugs trended toward higher burst pressures than standard glue for both 1- and 2-millimeter perforations. More importantly, when tested in five whole human globes, the plugs maintained a seal at pressures well above the physiological range, with a minimum burst pressure of 112 millimeters of mercury, and none of them dislodged. This feasibility study suggests these plugs could offer a more mechanically stable, sutureless option for in-office perforation repair, pending further animal and human trials. Shifting to the posterior segment, the SAGA trial, published in Ophthalmology Retina, evaluated an oral agent, gildeuretinol acetate, for geographic atrophy [2]. This 24-month randomized trial compared daily oral gildeuretinol, a modified form of vitamin A, to placebo in 198 participants. The primary endpoint was the rate of GA growth over 24 months. The study did not meet its primary endpoint; the gildeuretinol group had a mean growth rate of 1.62 millimeters squared per year compared to 1.87 in the placebo group, a 13.4% relative reduction that was not statistically significant. However, several prespecified secondary and exploratory analyses showed nominally significant results favoring gildeuretinol. These included a 15% reduction in GA growth between months 6 and 24, and better preservation of low-luminance visual acuity, with about 4 fewer letters lost in the treatment group. Patient-reported outcomes on visual function questionnaires also favored gildeuretinol. While the primary endpoint was negative, these secondary signals suggest the drug may warrant further evaluation in larger, adequately powered trials.

If you only have time for one paper this week, make it the study on risk-stratifying glaucoma suspects from the American Journal of Ophthalmology [3]. It provides a strong evidence base to support personalizing follow-up schedules, which could significantly improve clinic efficiency and patient convenience without compromising safety for low-risk individuals.

Here are the key takeaways from this week in Ophthalmology: First, for low-risk glaucoma suspects, such as untreated patients under age 50, the annual conversion rate to glaucoma is low, around 2%. This supports a data-driven approach to extending follow-up intervals, potentially improving clinical efficiency. Second, in a large database study, the osteoporosis drug denosumab, a RANKL inhibitor, was associated with a significantly lower risk of developing both ocular hypertension and glaucoma compared to bisphosphonates. This suggests a novel biological pathway that merits further investigation for glaucoma prevention. Third, AI-powered smartphone applications are showing real-world success in population-level screening. A new app successfully detected new cases of ocular surface malignancy, demonstrating a scalable model to improve early diagnosis for rare but serious conditions. Fourth, the retina continues to prove itself as a window to systemic health. In very preterm infants, RNFL thickness measured by OCT at 36 weeks postmenstrual age is a strong predictor of motor and cognitive neurodevelopmental outcomes at age two. Finally, be aware that undisclosed use of AI to write academic papers is increasingly common in ophthalmology literature. This raises significant concerns about scientific integrity and highlights the need for editorial oversight and clear disclosure policies.

That's your roundup for This Week in Ophthalmology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.