Australian researchers develop AI test that diagnoses glaucoma in just 10 seconds
Glaucoma is a surprisingly common disease that can have serious consequences if left untreated. Recognizing the importance of early detection, a team of engineers and ophthalmologists from Australia has developed a new AI-based approach to diagnosing glaucoma that can deliver results in just 10 seconds.
Glaucoma is a condition in which pressure inside the eye increases, which in turn can put pressure on the optic nerve. If the optic nerve is exposed to this over a long period, it can lead to irreversible damage and vision loss. The disease can be treated with medication, but it first has to be detected.
Diagnosing glaucoma usually involves a 30-minute test with an ophthalmologist or a visit to a specialized optometry clinic. However, not everyone has access to such specialists.
Hoping to find a simpler way to diagnose glaucoma, Dinesh Kumar, a professor at the Royal Melbourne Institute of Technology (RMIT), and his colleagues sought to develop a new approach based on artificial intelligence and the fact that glaucoma changes the way a person's pupil reacts to light.
The scientists assembled a small group of study participants: 13 with glaucoma, 13 healthy people as a control group, and another 13 younger people. The volunteers' eyes were imaged at 60 frames per second while they focused on a point under ambient lighting, and a machine learning algorithm was then applied to the data to detect minute changes in pupil dilation in the glaucoma group compared with the healthy control group. Based on this work, the researchers developed a glaucoma diagnostic model.
"The complexity of participants' pupils was analyzed using Higuchi fractal dimension, permutation entropy and conditional entropy. We found a statistically significant difference in pupil complexity between glaucoma patients and the control group. The difference was more pronounced when using the fractal dimension measure. These results support the possibility of using pupil complexity for glaucoma screening with commercial devices," the study says.
“Our software can measure how the pupil adapts to ambient light and capture minor changes in pupil size. Existing AI-based glaucoma tests require the patient to remain completely still for 10 minutes. Our technology does the job in 10 seconds without compromising accuracy”
This diagnostic model could help millions of people around the world who are at risk of developing glaucoma, especially those over 60, as they face a higher risk.
“The research points to a non-contact, easy-to-use and inexpensive test that can be carried out in ordinary clinics. It could also help enable community-wide screening programs, reaching people who would otherwise not seek treatment until it was too late”
Kumar said that although this study used a specialized eye-tracker, he expects that one day the approach could be used through a simple smartphone app, and the team plans to explore this possibility in the near future. The researchers are also interested in conducting clinical trials of their approach in 2022.
"Our pioneering study examined the difference in the complexity of hippus (rhythmic constrictions and dilations of the pupil) between glaucoma and healthy eyes and proposed a framework for glaucoma detection that has the potential for automated and unsupervised recording and analysis of data. This study shows the potential of pupil complexity as a biomarker for glaucoma, which can be recorded with an inexpensive eye-tracker, and pupil recordings can be analyzed automatically. The fractal dimension of pupil signals from glaucoma patients was significantly higher than in the control group of healthy peers and young people. The same results were obtained using permutation entropy and conditional entropy. This shows that pupil data can help in detecting glaucoma," the scientists said.
However, the study has several serious limitations. First, it is a pilot study in which the number of participants is insufficient to examine differences across many factors. Second, the work did not investigate the correlation between pupil complexity and the severity of glaucoma. In real life, changes in pupil complexity may reflect other eye diseases.
Further development of their work will involve studying the effects of disease progression and visual acuity, differences between sexes and ethnic groups, and identifying the effect of other eye diseases and concomitant factors. There is also a need to optimize the proposed framework for a broader range of participants.
