Modern diabetic retinopathy screening is scaling fast

Diabetic retinopathy (DR) screening has traditionally relied on trained human graders reviewing retinal photographs — a process that is accurate, but labor-intensive and increasingly difficult to scale.

As screening volumes rise and staffing pressures grow, artificial intelligence (AI) has emerged as a promising solution. A large UK-based study provides important insight into how well AI performs when paired with modern retinal imaging systems.

Why this study matters

The study, published in the British Journal of Ophthalmology, evaluated the diagnostic accuracy of an AI-enabled grading algorithm (EyeArt) when analyzing retinal images from two different imaging systems:

Traditional "standard" color fundus cameras used in national diabetic screening programs
The iCare EIDON true-color, wide-field confocal scanning system

The key question: Does AI maintain screening-level safety and accuracy when used with newer imaging technology?

Study design at a glance

More than 1,250 patients attending routine diabetic eye screening were included.

Each patient underwent imaging with both standard fundus cameras and the EIDON platform.

Human graders reviewed the standard fundus images according to national screening protocols, and these grades served as the reference standard. The AI algorithm then analyzed both image sets independently.

Key results: sensitivity where it matters most

For screening programs, sensitivity — especially for vision-threatening disease — is critical. Missing advanced diabetic retinopathy can delay referral and treatment.

Any diabetic retinopathy: ~92% sensitivity on both EIDON and standard fundus images
Vision-threatening diabetic retinopathy: 99–100% sensitivity
Proliferative diabetic retinopathy: 100% sensitivity across both imaging systems

In practical terms, the AI algorithm demonstrated nearly identical performance on EIDON images compared to traditional fundus photographs, exceeding typical sensitivity thresholds recommended for screening tools.

What about missed cases?

One case of vision-threatening disease was missed by the AI when analyzing EIDON images but was identified by human graders.

Importantly, no proliferative cases were missed, and all vision-threatening cases were detected by AI on standard fundus images.

Clinical takeaway: AI is best viewed as a screening and triage tool, not a replacement for clinical judgment.

Implications for clinical practice

iCare EIDON images are suitable for AI-based diabetic retinopathy screening
AI performance is comparable to traditional screening camera workflows
High-risk, referral-level disease is reliably identified

For clinics and health systems facing growing screening demand, this supports more scalable models of care without compromising patient safety.

Looking ahead

The authors note that future work should focus on improving AI specificity — particularly in confidently identifying patients with no retinopathy — and understanding subtle lesion detection differences between imaging systems.

Bottom line: AI-assisted grading on iCare EIDON retinal images performs at a screening-safe level comparable to traditional fundus photography, supporting its role in modern diabetic eye screening programs.

Interested in iCare retinal imaging for your practice? Contact Blue River Medical to learn more.

AI Grading of Diabetic Retinopathy: What a Major Screening Study Means for Modern Retinal Imaging