AI (artificial intelligence) that is trained to recognize red flags in retinal images and clinical data can predict if and when people at high risk of glaucoma, commonly referred to as ‘glaucoma suspects’, may actually develop it, according to research published online. goes into British Journal of Ophthalmology.
Subject to further refinement with larger numbers of people, it could prove a helpful diagnostic aid for doctors, the researchers concluded.
Recent advances in AI have prompted the design of algorithms to better detect glaucoma progression. But no one has yet drawn on clinical features to predict disease progression in people at high risk, the researchers noted.
Glaucoma is one of the leading causes of blindness worldwide. But it’s particularly difficult for doctors to know when and if people with suspicious symptoms of primary optic nerve damage, but without the cardinal diagnostic feature of abnormally high internal pressure in the eye—intraocular pressure, or IOP for short—could be glaucoma. And risk losing their vision, they explain.
Aiming to use AI to bridge this gap, researchers reviewed clinical data on 12,458 eyes with suspicious primary symptoms.
Of these, they focused on 210 eyes that progressed to glaucoma and 105 eyes that did not, all of which were monitored every 6–12 months for at least 7 years.
They then used red flag signs on retinal images taken during the observation period and created a set of ‘predictive’ combinations of 15 key clinical features, which were then fed into 3 machine learning classifiers – an algorithm that automatically orders or categorises data.
Clinical characteristics included age, gender, IOP, corneal thickness, retinal nerve layer thickness, blood pressure, and weight (BMI).
All three algorithms performed well and were able to consistently predict glaucoma progression and when with a high degree of accuracy: 91–99%.
The 3 most important predictive clinical characteristics are baseline IOP, diastolic blood pressure—the second number in the blood pressure reading that measures arterial pressure between heartbeats—and mean retinal nerve fiber layer thickness.
The mean age of the participants at the start of the observation period was 55, with a range of 33 to 76. Baseline age did not emerge as a key predictive factor, but the average age of those who progressed to glaucoma was significantly younger than those who did not, the researchers note.
They acknowledge various limitations of their findings. For example, the results of the AI training were based on relatively little data, and only those with normal IOP who were not given any glaucoma treatment during the observation period were included in the study.
“The present results, thus, only show that the constructed model works well for a limited range of patients,” they warn.
Yet they conclude: “Our results are suggestive [deep learning] Models trained on both eye images and clinical data have the potential to predict disease progression [glaucoma suspect] patients
“We believe that with additional training and testing on a larger dataset, our [deep learning] Models can be made better, and with such models, clinicians will be better equipped to predict individuals [glaucoma suspect] Patients’ respective disease courses.”
They add: “Prediction of disease course on an individual-patient basis will help clinicians to present appropriate management options for patients such as duration of follow-up, initiation of IOP-lowering treatment (or not) and target IOP levels.”
Ha, A. etc. (2023). Deep-learning-based prediction of glaucoma conversion in normotensive glaucoma suspects. British Journal of Ophthalmology. doi.org/10.1136/bjo-2022-323167.