What FDA Clearance Data Reveals About Medical Imaging AI

The fastest way to understand medical image AI in 2026 is not to start with a vendor demo. Start with the FDA list. Through the end of 2025, the agency’s AI/ML-enabled medical device inventory contained 1,451 authorized devices, and radiology accounted for 1,104 of them, or about 76% of the total.[1][2] That is not a rounding error or a temporary lead. It is the center of gravity of regulated healthcare AI.

It is also not proof that every cleared imaging algorithm improves outcomes, saves radiologist time, or performs well after local deployment. FDA clearance, especially through the 510(k) pathway, is a regulatory determination built around substantial equivalence to a predicate device; it is not the same thing as independent clinical efficacy validation across health systems. The clearance data is therefore best read as a map of regulatory maturity and commercial activity, not as a substitute for evidence review.

Radiology’s lead is now structural

The FDA’s 2025 data makes radiology’s dominance difficult to wave away as a reporting artifact. The agency’s AI/ML-enabled device list reached 295 new clearances in 2025, a record annual total, and those clearances came from 221 manufacturers.[1] In the fourth quarter alone, radiology represented 55 of 72 clearances, and across 2025 it accounted for roughly 75% of all AI/ML device authorizations.[2]

There are practical reasons imaging arrived first. Radiology already runs on digital data, picture archiving and communication systems are established, and many imaging use cases can be framed as detection, triage, reconstruction, quantification, or workflow support around a defined input. None of that makes validation easy. It does explain why the FDA list is so heavily populated with imaging products while other clinical AI categories remain thinner.

The inclusion criteria matter. The FDA list includes AI/ML-enabled medical devices, which can include imaging hardware with embedded AI as well as standalone software. That means year-to-year device counts are not a clean proxy for the number of pure software algorithms entering hospitals. Still, for procurement and regulatory affairs teams, the operational fact remains: the authorized product universe they must sort through is overwhelmingly radiology-heavy.

For readers who want the broader authorization context outside imaging, the larger FDA surge is covered in AI medical technology authorizations. The narrower point here is that medical image AI is not merely participating in that surge. It is driving much of it.

Clearance volume creates a different kind of due diligence problem

A small market lets a committee ask whether a product category is real. A large cleared market forces a more annoying question: which of these cleared products is actually comparable to the problem the hospital is trying to solve?

The 2025 data points in two directions at once. On one side, the number of clearances, the median review time of 142 days, and the fact that 24% of clearances were completed in under 90 days all indicate a regulatory channel that is no longer novel in its basic operation.[1] On the other side, 183 of the 221 manufacturers with 2025 clearances had only a single clearance that year.[1] That is not a tidy market.

This is where clearance counts become less helpful by themselves. A procurement lead can see that a vendor has an FDA-cleared product. That does not immediately answer whether the product has evidence in a comparable patient population, whether it integrates with existing imaging infrastructure, whether radiologists will accept the alerting behavior, or whether the manufacturer can support upgrades, audits, cybersecurity reviews, and post-market performance monitoring.

The clinical evidence gap deserves its own review, and for radiology it is examined more directly in FDA-cleared radiology AI and the clinical evidence landscape. The regulatory list can tell a buyer what has crossed an FDA authorization threshold. It cannot tell the buyer whether the deployment will work in their emergency department at 2 a.m., on their scanners, with their case mix, and under their staffing model.

The leaderboard is real, but it is not the whole market

The radiology AI clearance leaderboard is led by familiar imaging companies: GE HealthCare with 120 radiology AI clearances, Siemens Healthineers with 89, Philips with 50, Canon with 45, United Imaging with 38, Aidoc with 31, and DeepHealth with 28.[1] Those figures refer to radiology AI clearances, not total AI/ML devices across every specialty.

That distinction matters because the leaderboard can be misread as a simple proxy for product quality or clinical impact. It is not. A deep clearance portfolio may indicate regulatory experience, installed-base leverage, and the ability to manage multiple device submissions. It may also reflect a broad hardware and software footprint rather than superiority in a specific use case. A smaller company with one clearance may still have a product that fits a narrow operational problem better than an incumbent suite.

For purchasing committees, the useful comparison is rarely “Which company has the most clearances?” It is closer to: which cleared indication matches the clinical workflow, which version is being sold, what evidence supports that use, what data was used for validation, how the product changes after installation, and what contractual obligations attach to monitoring those changes.

A company landscape can help organize that first pass, particularly when it is built around use cases rather than brand visibility. A complementary vendor-oriented view is available in AI medical imaging companies in 2026. The FDA list should sit beside that kind of framework, not replace it.

Mature and fragmented can both be true

The usual language around medical image AI tends to split into two unhelpful camps. One treats the FDA count as if it proves clinical inevitability. The other treats each cleared tool as if it were still an untested research prototype. The clearance data supports neither shortcut.

Regulatorily, this is a mature category. In 2025, 62% of AI/ML-enabled device clearances were classified as software as a medical device, and 63% had a diagnostic purpose.[1] That mix shows that the FDA is not only seeing AI as embedded device functionality or administrative support. A substantial portion of the pipeline is software intended to participate in diagnostic work.

Commercially, the market is large enough to attract broad investment. Published market estimates place the AI medical imaging market around $2.2 billion to $2.5 billion in 2026, depending on the source and methodology.[3][4] That range should be read as market context, not as a precise measurement of FDA-cleared product revenue. It still helps explain why so many manufacturers are entering the category.

Operationally, fragmentation is the tax buyers pay for that growth. A radiology administrator facing dozens of vendors cannot rely on FDA status as the sorting mechanism once every serious candidate has some form of clearance. The shortlist has to move from authorization status to indication fit, evidence fit, implementation burden, governance, service model, and lifecycle control.

PCCPs change what “cleared” needs to mean in procurement

The most important regulatory shift in the 2025 data is not another high clearance count. It is the early appearance of Pre-specified Change Control Plans. In 2025, 10.2% of clearances included PCCPs, allowing certain iterative post-market updates to proceed under a pre-specified plan rather than requiring every eligible change to be treated as a wholly new regulatory event.[1]

For health systems, PCCPs make the old checklist question too thin. “Is it FDA cleared?” remains necessary, but it is no longer sufficient when the product may change under a cleared change-management plan. The next questions are more specific: what modifications were pre-specified, what boundaries limit those modifications, what performance testing is required before an update, how will the hospital be notified, and who decides whether local revalidation is needed?

• Ask whether the marketed version matches the cleared version and documentation.
• Request the scope of any PCCP and identify which future changes are covered.
• Define how updates will be communicated to radiology, IT, compliance, and clinical governance teams.
• Specify whether the hospital can pause, test, or reject updates before they affect clinical workflow.
• Tie post-deployment monitoring to measurable performance, safety, and bias checks rather than vendor assurances alone.

This is not a theoretical governance preference. A continuously updated product changes the burden on the buyer. Compliance teams need traceability. Radiologists need to know when a tool behaves differently. IT needs release management. Quality and safety teams need a monitoring plan that survives beyond go-live. For bias and performance monitoring after deployment, a practical companion framework is available in medical imaging AI bias evaluation.

Other specialties show how uneven the category still is

Radiology’s dominance should not be mistaken for uniform AI adoption across medicine. In the FDA AI/ML-enabled device data, cardiology represented 8.8% of devices and neurology 4.7%.[1] Those are meaningful categories, but they are nowhere near radiology’s 76% share.

That specialty gap matters for strategy. A health system evaluating imaging AI may be operating in a crowded vendor field with many cleared alternatives. A team evaluating AI in another specialty may face a thinner regulatory record, fewer comparable products, and a different evidence base. The procurement posture should change accordingly.

For specialty-by-specialty evidence comparisons, the better starting points are AI diagnostics evidence by specialty and AI in healthcare by specialty. The FDA list can show where regulated products are accumulating. It cannot, by itself, equalize the maturity of evidence across cardiology, neurology, radiology, and other fields.

What the FDA list can and cannot do for buyers

The FDA AI/ML-enabled device list is an unusually useful procurement starting point because it imposes a factual floor. It can identify whether a product has been authorized, which specialty it falls into, who the manufacturer is, and how the authorized device is described. It can also reveal market structure: radiology concentration, annual clearance pace, manufacturer spread, and the arrival of PCCPs.

Its weakness is just as important. It does not rank clinical usefulness. It does not tell a hospital whether a tool will reduce report turnaround time, avoid alert fatigue, improve diagnostic accuracy in the local population, or maintain performance after an update. It does not convert a crowded category into a qualified shortlist.

Medical image AI is no longer early in regulatory terms. The harder problem now belongs to the people buying, governing, and monitoring it. Abundance has made simple clearance counting inadequate; fragmentation has made vendor comparison harder; and PCCPs mean the evaluation has to cover the product lifecycle, not just the product label at purchase.

References

1. 2025 FDA AI/ML Medical Device Authorizations Year in Review. Innolitics.
2. Radiology Dominates FDA AI List. The Imaging Wire. March 2026.
3. AI In Medical Imaging Market Size, Share & Trends Analysis Report. Grand View Research.
4. Artificial Intelligence in Medical Imaging Market. Coherent Market Insights.