Imagine stepping into a supermarket, train station, or retail center and having eyes screened for glaucoma in seconds—no appointment necessary. This goal may soon become a reality, thanks to the AI-based Glaucoma Screening (AI-GS) network, built by a research team led by Nakazawa and Associate Professor Parmanand Sharma at Tohoku University’s Graduate School of Medicine. The findings were published in npj | Digital Medicine.
Overview of the AI-Based Glaucoma Screening (AI-GS) network. The process begins with an image of the retina captured by a fundus camera. The AI-GS network employs lightweight AI models to analyze the image, producing outputs such as segmented images, quantitative values of glaucoma-related features, and their presence in the image. Based on a comprehensive analysis, the system generates a final recommendation: "Normal" or "Consult an ophthalmologist.” Image Credit: Sharma et al.
Glaucoma is Japan’s and the world’s leading cause of irreversible blindness. Early identification is crucial because the condition advances silently, gradually reducing one’s peripheral range of vision.
Patients frequently do not notice this loss of vision at first, which means that substantial and irreversible damage can occur before they even consider scheduling a doctor’s visit. As a result, many instances go misdiagnosed because of the scarcity of ophthalmologists and the difficulties of conducting mass screenings, particularly in resource-constrained areas.
This is why we developed a new, quick, portable testing method. It analyzes multiple key indicators of glaucoma, integrates the findings, and determines the presence of the disease with unprecedented precision.
Toru Nakazawa, Professor, Tohoku University
The AI-GS network was evaluated on a dataset of 8,000 fundus images of the back of the eye (where glaucomatous damage develops), and it achieved an amazing 93.52% sensitivity at 95% specificity, which is similar to expert ophthalmologists. Unlike typical AI models, this method is effective at diagnosing early-stage glaucoma, even when fundus abnormalities are subtle and difficult to detect.
A fundamental barrier in AI-driven healthcare is its lack of interpretability—the so-called “black box” dilemma, in which it is unclear what processes the AI took to reach a decision. AI-GS tackles this problem by assigning numerical values to each diagnostic feature, allowing ophthalmologists to understand and verify the decision-making process. This transparency builds confidence and allows for smooth inclusion into clinical practice.
Size was also a key consideration for making actual implementation as simple as feasible. The AI-GS network is designed to be portable and efficient, with a size of only 110 MB. It uses little computational power and returns diagnostic results in under a second.
AI-GS brings expert-level glaucoma screening to your pocket, complementing specialist evaluations. It can be run on a mobile device and used in all sorts of public places because of its portability. You can run screenings at train stations or even remote regions that otherwise have limited access to ophthalmologists.
Parmanand Sharma, Associate Professor, Tohoku University
“This AI technology bridges a critical gap in glaucoma detection by making specialist-level diagnostics accessible to underserved communities. By enabling early detection on a large scale, we have the potential to prevent blindness for millions worldwide,” Nakazawa added.
With its high accuracy, AI explainability, and lightweight architecture, the AI-GS network is a significant step forward in AI-driven ophthalmology, taking glaucoma screening out of hospitals and into the community. The widespread application of this method could transform glaucoma care, ensuring that no patient goes untreated due to a lack of access to specialists.
Journal Reference:
Sharma, P. et. al. (2025) A hybrid multi model artificial intelligence approach for glaucoma screening using fundus images. npj | Digital Medicine. doi.org/10.1038/s41746-025-01473-w