Ophthalmology · Autonomous diabetic-retinopathy screening
AEYE-DS
AEYE Health, Inc.
The third autonomous diabetic-retinopathy AI cleared in the United States, and the first to be cleared on a handheld camera as well as a desktop one. Its pivotal study reported 92.98% sensitivity and 91.36% specificity for more-than-mild retinopathy from a single image per eye, with most patients screened without pupil dilation. What separates it from its two predecessors is not the headline accuracy, which is comparable, but the evidence trail behind it: the pivotal diagnostic-accuracy data live in the FDA clearance summaries rather than in a standalone peer-reviewed pivotal paper, and the company's published work to date covers an adjacent macular-edema model and patient acceptability rather than the cleared retinopathy device.
Performance Metrics
Clinical Evidence
AEYE-DS entered a category its two predecessors had already defined. IDx-DR (now LumineticsCore) was granted De Novo authorization in 2018 on the strength of a prospective pivotal trial published in npj Digital Medicine; EyeArt followed in 2020 with its own peer-reviewed pivotal evidence. AEYE-DS is the third such device, and it arrived with a specific pitch — a single non-mydriatic image per eye, rather than the two-field-per-eye protocols the earlier systems were validated on. Its pivotal study (NCT04612868, run across US primary-care sites with the Topcon NW400) reported 92.98% sensitivity (95% CI 83.00–98.05) and 91.36% specificity (95% CI 88.19–93.91) against a reading-center reference standard, with imageability above 99% and 88% of subjects screened without pupil dilation. Those figures sit in the same band as the earlier devices and support the single-image claim. The honest qualification is where that evidence lives. The pivotal diagnostic-accuracy results are documented in the FDA 510(k) clearance summaries and the manufacturer's own reporting, not in a dedicated peer-reviewed pivotal publication of the kind that underpins LumineticsCore and EyeArt. The two peer-reviewed papers carrying AEYE Health's name address neighbouring questions rather than the cleared retinopathy device: a deep-learning model for diabetic macular edema validated on the public EyePACS dataset (Ophthalmology Science 2025; image-level AUC 0.954), which is a research model adjacent to — not a validation of — the cleared mtmDR product; and a prospective patient-acceptability cohort at Johns Hopkins using the Aurora AEYE configuration (Clinical Ophthalmology 2026), which measures patient perspectives, not diagnostic accuracy. For a device whose accuracy is its central claim, the absence of an independent, peer-reviewed pivotal report is the material gap a procurement reviewer should weigh, and the reason this entry stops short of the top editorial mark.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| AEYE Health, Inc. — pivotal study (NCT04612868), reported in FDA 510(k) K221183 clinical summary | ProspectiveProspective | 469 | 92.98% (95% CI 83.00–98.05) | 91.36% (95% CI 88.19–93.91) | — | FDA 510(k) K221183 summary + AEYE Health report, 2022; ≈470 subjects across US primary-care sites, Topcon NW400, single image per eye; imageability >99%, 413 (88%) imaged without dilation. Not published as a standalone peer-reviewed pivotal paper. |
| AEYE Health, Inc. — handheld validation, reported in FDA 510(k) K240058 summary | ProspectiveProspective | 679 | — | — | — | FDA 510(k) K240058 summary, 2024; two prospective, multicentre, single-arm, blinded studies (n=317 and n=362) using the Optomed Aurora handheld camera, reference standard ETDRS grading by an independent reading centre. First autonomous DR clearance on a handheld camera; manufacturer-reported ~91.9% sensitivity / 93.6% specificity. |
| Bressler I, Aviv R, Margalit D, Cohen GY, Ianchulev T, Savant SV, Ramsey DJ, Dvey-Aharon Z (AEYE Health-supported) | RetrospectiveRetrospective | 15,892 | — | — | 0.954 image-level (0.962 patient-level) | Ophthalmology Science, 2025; deep-learning diabetic-macular-edema model developed and validated on the public EyePACS dataset (32,049 images from 15,892 patients). A research model adjacent to the cleared mtmDR product — not a validation of the cleared AEYE-DS retinopathy device. |
| Rustam Z, Xie Y, Amezcua Moreno J, Tran D, Zhu G, Yu SE, Sandrosyan A, Cai CX (Wilmer Eye Institute, Johns Hopkins) | ProspectiveProspective | 256 | — | — | — | Clinical Ophthalmology (Dovepress), 2026; prospective cohort using the Aurora AEYE configuration (AEYE-DS) at an urban US medical centre, measuring patient acceptability of autonomous screening — not a diagnostic-accuracy study. Sample size approximate pending direct source read. |
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 AEYE-DS 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
One non-mydriatic image per eye, from a paired camera.
A health-care provider captures a single fundus image per eye with a paired, separately regulated camera — the Topcon NW400 desktop unit under the original clearance, or the Optomed Aurora handheld under the 2024 clearance. The workflow is deliberately economical: one image per eye rather than the multi-field protocols the earlier autonomous systems were validated on, and, for most patients, no pupil dilation. In the pivotal study, imageability exceeded 99% and 88% of subjects were screened without dilation.
Input
Fundus image (one per eye)
Inference
Autonomous, per-patient
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
K240058
Class II
Safety Record
No recalls, field safety notices, or FDA safety communications attributable to AEYE-DS were found in indexed public sources as of August 2026 — read as "none found", not an exhaustive audit, since the FDA MAUDE and recall databases could not be queried directly from this environment. The residual risks are the familiar ones for an autonomous screening tool: a no-refer result is a screening output, not a normal-eye finding, and false negatives place the burden of catching missed disease on the programme's re-screening interval and the patient's onward follow-up. As with every device in this category, the labelled use is restricted to adults with diabetes who have not previously been diagnosed with diabetic retinopathy; it is not validated for surveillance of established disease.
Intended Use & Indications
AEYE-DS (AEYE Diagnostic Screening) is a retinal diagnostic software device indicated for use by health-care providers to automatically detect more-than-mild diabetic retinopathy (mtmDR) in adults with diabetes who have not been previously diagnosed with diabetic retinopathy. A provider captures a fundus image with a paired, separately regulated camera; AEYE-DS assesses image quality and returns an autonomous refer / no-refer result without a clinician reading the image. The original clearance (K221183, 2022) pairs the software with the Topcon NW400 desktop camera; the follow-on clearance (K240058, 2024) adds the Optomed Aurora handheld camera and sets the labelled population at adults aged 22 and older. In both configurations the software is the regulated diagnostic device and the camera is image-capture hardware. The intended setting is point-of-care screening in primary care and similar non-eye-care environments, where the autonomous result is a screening trigger for onward referral, not a substitute for a dilated eye examination or an ophthalmologist's diagnosis.