Ophthalmology · Autonomous diabetic retinopathy screening
EyeArt
Eyenuk
EyeArt is the autonomous diabetic-retinopathy screener that cleared FDA on both the more-than-mild and vision-threatening endpoints — broader than the predicate that created its product code — and carries a three-jurisdiction footprint across the US, EU and Canada, where its licence actually predates the US clearance. The evidence runs deeper than most: a manufacturer-led pivotal trial plus a 30,405-episode independent NHS study. That study also sets the honest counterweight — real-world specificity of 68 per cent against 95.7 per cent sensitivity.
Performance Metrics
Clinical Evidence
The evidence base is unusually deep for an autonomous screener and mixes manufacturer-led and independent work. The pivotal study (Ipp et al., JAMA Network Open 2021; NCT03112005; n=893 analysed, 1,786 eyes; manufacturer-authored) evaluated EyeArt v2.1.0 against a Wisconsin Fundus Photograph Reading Center reference standard graded to the [[etdrs]] scale. On the undilated imaging protocol it reported [[sensitivity]] 95.5% (95% CI 92.4–98.5) and [[specificity]] 85.0% (82.6–87.4) for more-than-mild disease, and sensitivity 95.1% (90.1–100) with specificity 89.0% (87.0–91.1) for vision-threatening disease — clearing the pre-specified superiority thresholds for both. The strongest independent signal comes from the English diabetic eye-screening programme: Heydon et al. (Br J Ophthalmol 2021) ran EyeArt across 30,405 consecutive screening episodes at three programmes including Moorfields, reporting 95.7% sensitivity (94.8–96.5) for referable retinopathy. The counterweight is in the same paper — specificity for no-retinopathy was 68%, a false-positive rate near a third, the price a sensitivity-weighted screener pays and a real workload consideration at population scale. A manufacturer real-world analysis (Bhaskaranand et al., Diabetes Technol Ther 2019) reported 91.3% sensitivity and 91.1% specificity for referable disease across 850,908 images from 101,710 visits on v2.0. Beyond diabetic retinopathy itself, an independent cross-sectional study (Ophthalmol Ther 2024, v2.1.0, 298 eyes) found the system detected active diabetic macular oedema against SD-OCT with 82.6% sensitivity and 84.5% specificity — a reminder that oedema is the harder call. A small independent series (J Diabetes Sci Technol 2022, 260 patients) reported perfect patient-level sensitivity and negative predictive value but a positive predictive value of only 19.2%, consistent with the over-referral profile. A 2026 systematic review and meta-analysis (Ophthalmologica, 17 studies) put the summary AUC at 0.932 (0.885–0.985). No randomised controlled trial of the device has been published.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Ipp E, et al. (Eyenuk / investigators; NCT03112005) | ProspectiveProspective | 893 | 95.5% mtmDR / 95.1% vtDR | 85.0% mtmDR / 89.0% vtDR | — | JAMA Netw Open 2021; 1,786 eyes; v2.1.0; undilated protocol; manufacturer-authored |
| Heydon P, et al. (independent; English DESPs incl. Moorfields) | ProspectiveProspective | 30,405 | 95.7% (referable) | 68% (no retinopathy) | — | Br J Ophthalmol 2021; 30,405 consecutive screening episodes; independent |
| Bhaskaranand M, Ramachandra C, Bhat S, et al. (Eyenuk) | RetrospectiveRetrospective | 101,710 | 91.3% (referable) | 91.1% | — | Diabetes Technol Ther 2019; 850,908 images / 101,710 visits; v2.0; manufacturer-authored |
| Independent tertiary medical-retina cohort (Ophthalmol Ther) | RetrospectiveRetrospective | 298 | 82.6% (active DME vs SD-OCT) | 84.5% | — | Ophthalmol Ther 2024; 298 eyes; v2.1.0; cross-sectional; independent |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow EyeArt reaches a verdict.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
Two undilated fundus images are captured per eye.
A healthcare provider — often a nurse or technician, not an eye-care specialist — acquires two colour fundus images per eye with a compatible non-mydriatic camera. EyeArt is cleared for use with more than one camera manufacturer: Canon CR-2 AF and CR-2 Plus AF, and, from the v2.2.0 clearance (K223357), the Topcon NW400.
Imaging is undilated by default; dilation is reserved for eyes the system cannot grade.
Input
Colour fundus images
Inference
Per-eye
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
K223357
Class II (21 CFR 886.1100, product code PIB)
Class IIb
100705
Class 2
Safety Record
No recall, field safety notice or regulator safety communication specific to EyeArt was found in public reporting as of July 2026; FDA MAUDE and the enforcement database could not be queried directly from this environment (403), so this rests on search-surfaced sources rather than a direct register read. The design over-refers rather than under-refers — the 68% real-world specificity (Br J Ophthalmol 2021) is the systematic cost, borne as follow-up workload, not as missed disease. A negative EyeArt result is bounded to the cleared indication: adults with diabetes not previously diagnosed with more-than-mild retinopathy, screened for the labelled endpoints. Ungradable images are routed to re-acquisition rather than forced to a verdict.
Intended Use & Indications
EyeArt is indicated for use by healthcare providers to automatically detect more-than-mild diabetic retinopathy and vision-threatening diabetic retinopathy (severe non-proliferative diabetic retinopathy, proliferative diabetic retinopathy and/or diabetic macular oedema) in eyes of adults diagnosed with diabetes who have not previously been diagnosed with more-than-mild diabetic retinopathy. The provider acquires two colour fundus images per eye with a compatible non-mydriatic camera; the system analyses them and issues an autonomous per-eye result — negative, positive, or ungradable — with no requirement for the acquiring clinician or any other physician to interpret the photographs. The US label spans two 510(k) clearances: K200667 (v2.1.0, both endpoints) and K223357 (v2.2.0), which added the Topcon NW400 camera, on-camera image-quality feedback and a dilate-and-retry protocol for ungradable eyes.