Artificial intelligence is being trained to help doctors diagnose glaucoma more consistently by analyzing the complex mix of patient data that specialists currently evaluate differently from one another. A fourth-year medical student at the Keck School of Medicine of USC is investigating whether AI tools can standardize glaucoma detection, potentially catching the disease earlier before it causes irreversible vision loss. What Makes Glaucoma So Hard to Diagnose? Glaucoma is a disease that damages the optic nerve, the bundle of nerve fibers that carries visual information from your eye to your brain. Unlike some eye conditions with obvious symptoms, glaucoma often progresses silently, which is why it's sometimes called the "silent thief of sight." By the time patients notice vision problems, significant damage may have already occurred. Diagnosing glaucoma requires doctors to synthesize multiple types of information. Ryan Shean, the USC medical student leading this research, explains the challenge: "There's so much information to synthesize, and different physicians synthesize it differently." Doctors must evaluate demographic factors, eye pressure measurements, vision sharpness, and structural and functional tests that reflect the health of the optic nerve. If the disease goes undetected, patients miss the opportunity for early treatments that can prevent vision from deteriorating. How Is AI Being Tested for Glaucoma Detection? Shean is working alongside Benjamin Xu, MD, PhD, associate professor of ophthalmology and chief of the Glaucoma Service at the USC Roski Eye Institute. Together, they're testing whether large language models and generative AI chatbots can improve diagnostic consistency. The researchers feed diagnostic criteria into both existing chatbots and custom algorithms trained using relevant medical information, then compare the machines' decisions with the consensus from human glaucoma specialists. The early results are encouraging. "So far, the performance from AI has approached that of specialists," Shean said. The algorithms show particular promise in identifying moderate to severe cases of glaucoma, where they could potentially help physicians screen for the disease more effectively. Steps to Implementing AI in Eye Care Safely - Accuracy First: Researchers are prioritizing accuracy over speed, ensuring AI systems are as reliable as possible before clinical use, since large language models can "hallucinate" and confidently relay false information. - Specialist Validation: All AI diagnostic decisions are compared against consensus from human glaucoma experts to verify performance and identify areas where the technology falls short. - Augmentation, Not Replacement: The goal is to create clinical decision-support tools that make physicians more effective, not to replace human doctors with machines. Shean emphasizes an important principle: "I want to make sure that this technology can be used to benefit physicians and improve patient care, and that the relationship between physician and patient is uncompromised." His own experience with a medical emergency as a teenager taught him the value of compassionate care, something no algorithm can replicate. What Are the Current Limitations? While the results are promising, significant challenges remain. Hallucinations are a major barrier to translating these findings into real clinical settings. "The large language models, and the specialized ones as well, need to be as accurate as possible," Shean noted. The sweet spot for current AI algorithms appears to be in identifying moderate to severe glaucoma cases. The technology may be less reliable for early-stage disease detection, which is ironically when intervention is most valuable. Researchers are continuing to refine the algorithms to improve performance across all disease stages. Shean's work has earned him a Medical Student Eye Research Fellowship from Research to Prevent Blindness, a competitive award that recognizes his exceptional promise in vision science research. "Research to Prevent Blindness has a distinguished history of identifying and supporting future leaders in vision science, and this highly competitive award reflects national recognition of Ryan's exceptional promise at an early stage in his career," said Dr. Xu. The research represents a broader shift in ophthalmology toward integrating technology with clinical expertise. Rather than viewing AI as a threat to the medical profession, researchers like Shean see it as a tool to standardize care and ensure that all patients receive the highest level of attention, regardless of which physician they see. As glaucoma continues to be a leading cause of irreversible blindness worldwide, innovations that improve early detection could have profound implications for preserving vision in millions of people.
