Four articles. Four minutes. | Issue #16
Artificial Intelligence: Machine-learning cup-to-disc ratio outperforms human grading for glaucoma detection in a population-based cohort
Laser: Repeat SLT was less durable than initial SLT over long-term follow-up
Surgery: SIBS microshunt showed higher 2-year success than gel microstent or trabeculectomy in a single-surgeon cohort
Epidemiology: 1 in 17 screened adults aged 77-89 had undetected open-angle glaucoma, mostly early to moderate
A Comparison of Machine Learning and Human Graders for Glaucoma Diagnosis from Fundus Images for Population Screening
Methods
The study included 6,304 participants with fundus photographs graded by trained human readers and an externally trained machine-learning model. Vertical cup-to-disc ratio estimates were compared for their ability to identify specialist-confirmed glaucoma or glaucoma-suspect status.
Results
Machine-estimated cup-to-disc ratio outperformed human grading, with Area Under the Receiver Operating Characteristic Curve (AUROC) of 90% versus 81% for right eyes and 88% versus 79% for left eyes. At 95% specificity, machine grading detected 50.5% of affected right eyes versus 39.0% for human grading. Human graders assigned a cup-disc ratio above 0.8 to only 2 glaucomatous eyes, versus 87 by the model, consistent with compression of the upper cup-disc range.
Conclusion
In this population-based UK cohort, machine-graded cup-disc ratio detected specialist-confirmed glaucoma more accurately than trained human graders. The cross-sectional design, two-dimensional photographs, and nearly all-White cohort limit immediate clinical implementation.
Our Angle
Anyone who routinely evaluates optic nerves knows how subjective and variable cup-to-disc ratio estimation can be, even among experienced clinicians. This study suggests that machine-learning analysis may extract more consistent and clinically useful information from a standard fundus photograph than visual grading alone. That is particularly relevant in screening and teleophthalmology, where OCT and visual fields may not be immediately available and the first decision is simply who needs a more complete glaucoma evaluation.
I would see the near-term role of this technology as triage rather than diagnosis. Even at 95% specificity, the model detected only about half of glaucoma or suspect eyes, so a normal AI result would not be reassuring enough to exclude disease. Its value may be in efficiently identifying higher-risk patients for clinical examination with gonioscopy, OCT and VF.
Before this becomes part of routine screening, we need prospective validation in more diverse populations and real-world workflows. The clinically meaningful endpoints are not simply AUROC or agreement with graders, but how many meaningful cases are detected earlier, how many unnecessary referrals are generated, and whether this approach ultimately improves access and reduces vision loss at an acceptable cost.
Laser
Long-term Outcomes of Selective Laser Trabeculoplasty on Intraocular Pressure Control and Visual Field Progression
Design: Single-center retrospective case-control study
Journal: American Journal of Ophthalmology, August 2026
Authors: Lee et al.
Among 500 eyes followed for a mean of 9.6 years after selective laser trabeculoplasty, most as an adjunct to topical therapy, median time to treatment failure was 4.6 years after initial treatment and 2.7 years after repeat treatment. Failure was more likely in eyes with prior glaucoma surgery (HR 2.33), moderate-to-advanced disease, and repeat sessions, and less likely with higher per-pulse energy. Among 302 eyes analyzable for visual field progression, 11% progressed at 5 years, though eyes receiving another IOP-lowering procedure before year 5 were excluded, which may have lowered this estimate. The authors caution that the energy-progression association likely reflects harder-to-control disease rather than a harmful laser effect, and note that the single-center retrospective design and physician-titrated dosing limit causal interpretation.
Surgery
Intermediate Outcomes of SIBS Microshunt, Gelatin Microstent and Trabeculectomy: A Comparative Retrospective Cohort Study
Design: Single-center, single-surgeon retrospective cohort study
Journal: Ophthalmology Glaucoma, July 2026
Authors: Berkache et al.
In a single-surgeon retrospective study of 142 eyes, the SIBS microshunt (PreserFlo), 45 µm gelatin microstent (Xen), and trabeculectomy were compared at 2 years, all with mitomycin C. Complete success (IOP 6 to 17 mmHg with at least 20% reduction and no medications) was 72.1% for the microshunt, 50.0% for trabeculectomy, and 38.1% for the microstent, despite prior filtering surgery in 41.9% of microshunt eyes versus under 3% in either comparator. Needling was required in 2.3% of microshunt eyes versus roughly 30% after the other procedures, while reoperation was most frequent after the microstent (15.9% versus under 3%). These results may inform counseling and postoperative planning, although the non-randomized baseline differences, single-surgeon technique, and 2-year follow-up limit comparative conclusions.
Epidemiology
Prevalence and severity of undetected open-angle glaucoma in persons 77-89 years old
Design: Population-based cross-sectional screening study
Journal: Ophthalmology Glaucoma, July 2026
Authors: Oskarsdottir et al.
In a population-based screening of 602 Malmö residents aged 77 to 89, previously undetected open-angle glaucoma was found in 5.98% of participants included in the prevalence analysis (35 patients, 95% CI 4.2% to 8.2%), with 74% of cases unilateral. Visual field loss was early in 66% and moderate in 20%, while 14% had advanced or severe damage and no patient had end-stage disease. Median IOP among newly identified patients was 16 mmHg (IQR 14 to 19), well below the 25 mmHg screening threshold, and tonometry alone would have missed most cases. The authors frame these findings as groundwork for evaluating screening strategies in this age group, although the 30.8% attendance rate and mainly European-derived single-city population limit generalizability.
Artelo Biosciences enrolls the first patient in the phase 2 DREAM study of an oral cannabinoid receptor agonist designed to lower intraocular pressure in glaucoma and ocular hypertension.
FDA classifies a recall of nearly 12 million cartons of Rohto Cooling Eye Drops as Class II, citing lack of sterility assurance across eight over-the-counter products, with no illnesses reported.
Biocon announces the U.S. commercial launch of Yesafili, an FDA-approved interchangeable aflibercept biosimilar, for neovascular AMD, diabetic macular edema, diabetic retinopathy, and macular edema following retinal vein occlusion.
ForSight Robotics completes 26 fully robotic-assisted cataract surgeries on its JASPER platform with no conversions to manual surgery, using standard anesthesia protocols and same-day discharge.
The drug atropine is named after Atropos, who in Greek mythology was:
A. The goddess of healing
B. The Fate who spun the thread of life
C. The Fate who cut the thread of life
D. The muse of medicine
See answer at bottom of newsletter.
Get the next issue in your inbox
The Open Angle is a weekly glaucoma research digest for busy eye care professionals.
Four articles worth knowing, in under four minutes.
Edited by Jella An, MD, MBA and Jason Dossantos, MD.
The Open Angle is an educational editorial product. It is not medical advice. Readers should review original sources before changing practice.
Trivia Answer: C