AI can identify treatment gaps for patients with macular disease, study shows

Using artificial intelligence (AI) in eye clinics could reduce both undertreatment and overtreatment of neovascular or ‘wet’ age-related macular degeneration (AMD). AMD is the leading cause of irreversible vision loss in older people and accounts for around 1 in 10 ophthalmology appointments in the NHS.

A new study published in Eye demonstrates that AI can accurately assess disease activity in wet AMD, potentially transforming how the NHS monitors and manages the condition. The technology could combat the twin challenges of workforce shortages in ophthalmology and increasing patient numbers, which have led to concerns over avoidable sight loss due to appointment delays. 

An estimated 520,000 people have wet AMD in the UK, with 40,000 new cases diagnosed each year and a treatment cost to the NHS of between £3,300 and £9,600 per patient each year. 

The research, led by Jeffry Hogg and Pearse Keane at Moorfields Eye Hospital NHS Foundation Trust, UCL Institute of Ophthalmology and University of Birmingham, compared AI-enabled assessments of wet AMD disease activity against real-world clinical assessments at two major NHS eye centres, Moorfields Eye Hospital and Newcastle Eye Centre.

During the study, 521 pairs of retinal images were curated from 468 patients attending consecutive routine appointments for wet AMD. The images, taken using optical coherence tomography (OCT), show fluid changes that indicate if the disease is active and requires treatment. 

When clinicians view a patient’s OCT image without the assistance of AI, they do not see the coloured segmentation of the retina shown here. In the OCT image [above right], the AI system has accurately identified parts of the retina that indicate health and disease. [Above left] AI has identified a peaking of the neurosensory retina (green) as it is pulled by the poster hyaloid face (cyan), indicating a condition called vitreomacular traction. AI has also identified subretinal hyper-reflective material (brown) and fibrovascular pigment epithelial detachment (red), both indicating dry Age-related Macular Degeneration (AMD). This complex example demonstrates the potential of AI-supported assessment for more accurate care management.

Across the 521 cases in the study, nearly 6 in 10 of cases thought to have active disease in real-world care did not, which dropped to just over 4 in 10 cases from independent AI analysis. This suggests AI could substantially lower unnecessary treatments and increase appointment availability for patients at risk of losing their sight. 

In real-world care nearly 2 in 10 of cases thought to be stable actually had active disease, which dropped to less than 1 in 10 from independent AI analysis. This indicates that not only could AI free up more appointments, but it could improve treatment decisions for patients at those appointments.

Lead author Jeffry Hogg, a clinical researcher and assistant professor at University of Birmingham

The research team conducted extensive error analysis and found no evidence of biased performance across age, sex, or ethnic groups. However, they identified some limitations with image quality that affected AI performance, highlighting the continued importance of human clinical oversight. The authors noted that results between Moorfields and Newcastle were consistent, which was promising for future implementation.

Lead author Jeffry Hogg said: Macular disease is one of the single biggest demands on ophthalmology services, and patients lose vision if services cannot keep up with demands. With 75% of departments reporting insufficient consultant staffing, we need innovative solutions that can help deliver better care more efficiently. Our research demonstrates how AI could help to reduce clinical demand, without compromising visual outcomes for NHS patients.” 

Senior author Pearse Keane said: "This is the first published comparison of autonomous AI-enabled treatment monitoring with standard NHS care across multiple centres. It provides an evidence base for how we might safely harness AI to meet the growing demands on NHS eye services. Encouragingly, patients and healthcare professionals we engaged with told us they would find AI contributions to treatment decisions acceptable if appropriately validated.”

Senior author Pearse Keane is Professor of Artificial Medical Intelligence at UCL Institute of Ophthalmology, a consultant ophthalmologist at Moorfields Eye Hospital, and Director of the INSIGHT Health Data Research Hub.

While the results are promising, the researchers note that current AI medical devices are approved only for decision support rather than autonomous use. The study concludes that future AI systems incorporating specific decision thresholds might enable more efficient pathways, allowing some stable patients to have fewer clinic visits without sacrificing safety.

Moorfields data for the study was curated through the INSIGHT Health Data Research Hub, an NHS-led initiative that comprises more than 30 million routinely collected ophthalmic images linked to clinical records. This data substantially improved ethnic diversity compared to using data only from Newcastle Eye Centre, and demonstrated the AI system's performance across different OCT imaging equipment.

The study was funded by the National Institute of Health Research.

---

For more information on eye health data available through INSIGHT for healthcare research, please visit our Researcher section: INSIGHT Data

To enquire about accessing eye health data through INSIGHT, please get in touch: Contact