Optometry has exactly one FDA-cleared, Medicare-reimbursed autonomous AI test, and it isn't for glaucoma

A sales rep demoing an "AI diagnostic" tool for glaucoma or macular degeneration will often use the same language as a rep selling autonomous diabetic retinopathy screening: detects, flags, diagnoses, catches what the naked eye misses. The language sounds interchangeable. The regulatory reality behind it is not. One of those conditions has a specific FDA-cleared pathway for AI to issue a screening result without a clinician reading the image first, plus a dedicated CPT code that pays for exactly that. The others don't, and as of this writing, none of them do. For a practice owner trying to separate a genuine clinical tool from confident marketing, that single fact does more work than any feature comparison.

Autonomous and assistive are not the same product category

The distinction that matters is whether the software makes the screening call or whether a clinician does. An autonomous AI system analyzes a retinal image and issues a result, refer or don't refer, without a clinician independently interpreting that image first. A person still acts on a positive result and every patient still sees a doctor, but the screening decision itself is the software's, made and documented before any clinician looks at the image. An assistive system does something narrower: it measures, scores, or highlights a feature in an image, an optic nerve cup-to-disc ratio, a suspicious drusen pattern, a retinal nerve fiber layer thinning trend, and hands that output to a clinician who makes the actual diagnostic call. Nearly every AI tool built into the OCT machines, fundus cameras, and visual field analyzers already sitting in optometry practices today falls into the second category.

The reason this distinction is worth learning, rather than treating as a technicality, is that only one of the two categories has a track record of FDA clearance and Medicare billing in optometry. Knowing which one you're being sold is the fastest way to evaluate a vendor claim without needing a clinical trial background to do it. A tool shouldn't get treated as clinically validated on the strength of a vendor's own word for it: a specific study or a specific FDA clearance number backs the claim, or it doesn't, and a vendor's use of the word "diagnostic" doesn't settle which.

Diabetic retinopathy is the one condition with a cleared, billable pathway

In 2018, the FDA granted De Novo authorization to IDx-DR, since rebranded LumineticsCore (Digital Diagnostics Inc.), as the first autonomous AI-based diagnostic system cleared in any specialty of US medicine, not just ophthalmology (FDA De Novo authorization, 2018, widely reported in FDA and industry coverage at the time). Eyenuk's EyeArt received 510(k) clearance roughly two years later on that predicate pathway, cleared to detect both more-than-mild and vision-threatening diabetic retinopathy; company-adjacent reporting describes it as in use across hundreds of clinics and cumulatively screening more than 230,000 diabetic patients, a figure that should be read as a vendor-adjacent cumulative claim rather than an independently audited installed base (Eyewire+, Ophthalmology Times, 2024-2025 coverage). More recently, AEYE Health's AEYE-DS became the first autonomous AI cleared for a handheld camera, paired with Optomed's Aurora device rather than a tabletop unit, with FDA clinical trial data reporting sensitivity of 92-93% and specificity of 89-94% for detecting more-than-mild diabetic retinopathy (PR Newswire and Optomed announcements, 2024, referenced again in trade coverage into January 2026; treat the sensitivity and specificity figures as company-reported trial results, not an independent replication).

The billing side moved in step with the clinical clearances. CPT code 92229, "imaging of retina for detection or monitoring of disease, point-of-care autonomous analysis and report, unilateral or bilateral," took effect January 1, 2021 (AMA CPT code set; CMS Physician Fee Schedule rulemaking). It was, at the time, the first CPT code that let AI software generate a billable diagnostic report without requiring a clinician's independent interpretation as the billed service. For 2026, CMS sets the national Medicare Physician Fee Schedule rate for 92229 at $46.76; actual payment varies by geographic locality, site of service, and whether a given commercial payer has adopted the Medicare rate or set its own (CMS 2026 Physician Fee Schedule; secondary fee-schedule aggregators corroborate the national figure).

So what this means for you: if a vendor's pitch for a diabetic retinopathy screening tool references FDA clearance and a billing pathway by name, that claim is checkable in minutes against public FDA and CMS records, not something you have to take on trust.

Glaucoma AI is real, and it is assistive, not autonomous

Glaucoma has not seen the same regulatory arc. As of this writing, no autonomous AI-enabled system for glaucoma screening has received US FDA approval or clearance, even though a substantial and growing body of research explores AI's potential in the disease, and numerous AI-enabled software features are already commercialized inside instruments used for quantifying retinal images and visual fields (trade and clinical review coverage, including Review of Ophthalmology's ongoing glaucoma AI coverage; EyeWiki's AI in Ophthalmology overview). Research in this area also flags a real limitation, not just a regulatory gap: both AI systems and optometrists show suboptimal performance diagnosing glaucoma from fundus photographs alone, which is one reason a cleared autonomous product hasn't emerged the way it did for diabetic retinopathy, a condition with clearer, more standardized image-based diagnostic criteria.

Age-related macular degeneration sits in a similar place: AI-assisted identification exists inside tools practices already use for pattern recognition on OCT scans, but as an assistive layer, not an autonomous, independently billable diagnostic decision. The practical result for an independent optometry practice is that "AI for glaucoma" and "AI for AMD" today mean decision support bundled into equipment you may already own or be evaluating for other reasons, priced and billed as part of that equipment, not a standalone diagnostic product with its own FDA clearance and CPT code the way diabetic retinopathy screening has.

So what this means for you: if a vendor markets standalone "AI glaucoma diagnosis" as functionally equivalent to autonomous diabetic retinopathy screening, whether in a spec sheet or a sales call, that framing is currently ahead of what any cleared product in the US actually does. Ask directly whether the tool is FDA-cleared as autonomous or offered as clinician decision support, and ask for the clearance number if the answer is autonomous.

Why the gap between the two categories matters for a small practice's economics

For a two- or three-location independent practice, the diabetic retinopathy pathway is a rare case where regulatory clearance, a specific CPT code, and a national Medicare rate all line up on the same tool. That combination lowers the evaluation burden considerably: the clinical claim has already been tested against an FDA bar, and the billing question has an actual code and a published national rate to check a payer contract against, rather than requiring the practice to build a business case from a vendor's own ROI marketing. A practice with a meaningful diabetic patient population, whether from direct referrals, comanagement with primary care, or a broader medically-oriented scope, can model the volume and reimbursement fairly precisely before committing to a device.

That precision doesn't exist yet for glaucoma or AMD AI, and pretending it does by treating every AI feature in an equipment quote the same way risks building a business case on a claim the product can't actually support. This is also where the American Optometric Association's more skeptical public posture is relevant context, not a contradiction: the AOA has urged the FTC and FDA to examine direct-to-patient automated eye testing technology and has updated its own policy to state that such technology should never replace doctor-led diagnostic care (AOA public statements, 2025-2026). That concern is aimed primarily at consumer-facing, at-home vision testing products bypassing a clinician entirely, a different category from clinician-supervised, in-office autonomous screening devices like the ones covered above, but it reflects the same underlying caution this article is built on: know exactly what a piece of software is cleared to do before treating its output as a diagnosis.

So what this means for you: build the model for diabetic retinopathy screening first if you're going to build one at all this year, because it's the one condition where the regulatory and billing groundwork is actually done. Treat glaucoma and AMD AI as a feature you're evaluating inside broader equipment decisions, not as a standalone revenue line yet.

What to actually check before signing anything

Four questions separate a genuine autonomous, billable AI product from an assistive feature wearing diagnostic language. First, ask for the specific FDA clearance or De Novo authorization number and look it up directly on the FDA's own database rather than accepting a vendor's summary of it. Second, ask whether the system is cleared as autonomous, meaning it issues its own screening result, or assistive, meaning a clinician's interpretation remains the billed service. Third, if autonomous use and a CPT code are both claimed, confirm the specific code, 92229 for point-of-care autonomous retinal analysis being the only one currently in this category in optometry, and check your own payer mix against the CMS national rate rather than assuming full reimbursement. Fourth, ask what the FDA clearance actually covers: a clearance for diabetic retinopathy detection does not extend to glaucoma, AMD, or any other condition the same camera or software might also flag, regardless of how the marketing material groups those features together.

None of this replaces your own clinical judgment about a specific device, and nothing here is clinical or regulatory advice for a specific purchase decision; confirm current FDA clearance status and payer coverage directly before committing budget. What it does do is give you a fast, checkable filter for separating the one part of optometry's AI market with real regulatory and billing infrastructure behind it from the much larger part still operating on assistive features and forward-looking marketing.

The call

If diabetic retinopathy screening fits your patient population, the case for evaluating an autonomous AI system is stronger than for almost any other AI purchase decision available to an independent optometry practice right now: FDA clearance exists, a specific CPT code exists, and a national Medicare rate exists, all three checkable in public records before a sales call ever happens. For every other eye disease currently marketed with AI, treat the pitch as an assistive feature evaluation, not a diagnostic AI purchase, until a cleared autonomous product for that condition actually exists.

For a scored, independent read on where diagnostic AI fits your specific patient mix and payer contracts, the AI Opportunity and Growth Assessment covers exactly this kind of vendor-claim verification. Start with a free 20-minute discovery call.

See also: our CARE Framework comparison of AI scheduling tools built for optometry, why clinical variation was never the real barrier to multi-location AI standardization, and a general five-question checklist for evaluating any AI tool's vendor claims before you buy.

Related: Weave vs Solutionreach vs RevenueWell: only 2 fit optometry · Optometry lost 10 points to consolidation · 5 questions to ask before buying any AI tool