Radiology
236 cleared products across 10 device types
Most cleared healthcare AI reads pictures. This is where the scans live, and it is by far the busiest field on the list.
FDA-Cleared AI Directory
Every AI product the FDA has cleared for use in care, sorted by medical field and by what each one is allowed to do. Every row carries the clearance number so you can check it yourself.
A vendor tells you their AI is FDA approved. You want to know if that is true, and you want to know it before the next meeting, not after the contract.
This is that list. It is rebuilt straight from the FDA's own clearance records, so the product name, the company, the date and the clearance number all come from the regulator rather than from a brochure.
The newest additions, straight from the FDA record. The clearance number is the one you quote back to a salesperson.
| Product | Company | Cleared | Type | FDA record |
|---|---|---|---|---|
| CADDIE | Odin Medical Limited | September 12, 2026 | Gastrointestinal Lesion Detection Software | K260133 |
| Rayvolve AZchest | AZmed | August 28, 2026 | AI Fracture Detection Software | K261378 |
| Tempus ECG-PH | Tempus AI, Inc. | August 21, 2026 | Pulmonary Hypertension AI Notification | K253699 |
| Transpara (2.1.0) | Screenpoint Medical B.V. | August 14, 2026 | AI Cancer Detection Software for Radiology | K260898 |
| CadAI-B Dx | BeamWorks, Inc. | August 12, 2026 | AI Cancer Detection Software for Radiology | K260086 |
| EW10-EC02 Endoscopy Support Program | Fujifilm Corporation | August 7, 2026 | Gastrointestinal Lesion Detection Software | K260567 |
| CAD - Lower GI (EC07-01) | Xiamen Inno Medical Technology Co., Ltd. | August 5, 2026 | Gastrointestinal Lesion Detection Software | K252853 |
| See-Mode Augmented Reporting Tool, Breast (SMART-B) | See-Mode Technologies Pty, Ltd. | July 28, 2026 | AI Cancer Detection Software for Radiology | K260303 |
| ThinkSono Guidance | ThinkSono GmbH | July 15, 2026 | AI-Guided Image Acquisition | K260338 |
| Rapid Vessel Occlusion (Rapid VO) | iSchemaview | July 13, 2026 | AI Triage and Notification Software for Radiology | K261085 |
| CXRDetectAI | Neurocareai, Inc. (Dba Savelife.Ai) | July 8, 2026 | AI Lesion Prioritization Software for Radiology | K253751 |
| Koios DS (3.8) | Koios Medical, Inc. | July 2, 2026 | AI Diagnostic Software for Suspicious Lesions | K253884 |
236 cleared products across 10 device types
Most cleared healthcare AI reads pictures. This is where the scans live, and it is by far the busiest field on the list.
15 cleared products across 8 device types
Heart and vessel AI that watches for one named condition and raises a hand. Each product here is cleared for one specific warning.
5 cleared products across 4 device types
Software that reads digitized slides. Small in number, and the first products cleared to help predict how a cancer will behave.
8 cleared products across 1 device types
Eye screening AI, including the first products cleared to give a result without a specialist reading the image.
6 cleared products across 1 device types
A short list, led by software cleared to help diagnose autism in young children.
23 cleared products across 1 device types
Colonoscopy software that flags possible lesions live, while the scope is still moving.
Almost every AI product in healthcare is cleared, not approved. The two words sound alike and mean very different things.
Think of it like moving into an apartment building. Approval is the city inspecting your new building from the foundation up. Clearance is the city saying your unit looks just like the one next door, which already passed, so you can move in. Clearance is a real bar. It is a lower one, and it is the bar almost all of this software clears.
So when a salesperson says approved, the honest translation is usually cleared. Ask which one. Then ask for the number, because every clearance has one.
You do not need to read the whole directory. Most visits are one of these three questions:
This is the part that costs hospitals money, so it is worth being blunt.
A clearance says the FDA agreed this product is close enough to something already on the market. It does not say the product works well on your patients, your scanners, or your workflow. It does not promise the accuracy you saw in the pitch deck. It does not mean the model behaves the same way two years from now, after the company has retrained it.
That is why the useful next step after this list is the evidence, not more marketing. Every device type page ends with the questions worth asking, and the company profiles on this site carry the same buyer brief.
The FDA does not publish a tidy list called AI. It publishes product codes, and it writes the name of each code itself. So the rule here is simple and you can check it: a product code is included when the FDA's own name for that code says the software does the work. That means computer-assisted, computer aided, machine learning, artificial intelligence, software algorithm, diagnosis aid, or autonomous.
That rule is deliberately narrow. It leaves out general image-viewing software that may happen to contain a model, because the FDA name does not say so and guessing would make this list worse. It also catches a handful of product codes that match the wording but are clearly not clinical AI, and those are named below rather than quietly dropped.
The whole directory is rebuilt from the openFDA 510(k) and device classification endpoints on a schedule. Nothing here is typed in by hand, which is the only way a list like this stays right.
| FDA code | Why it is not here |
|---|---|
| NOF | Dental restoration milling equipment, not a clinical decision function. |
| ONA | Surgical hair harvesting robot, not a diagnostic or triage function. |
| ODX | Dialysis blood leak alarm, a hardware sensor rather than a trained model. |
| PDR | Sedation delivery hardware, not a diagnostic or triage function. |
Clearance data comes from the FDA's public 510(k) and device classification records through openFDA and is refreshed on a schedule. A clearance is a regulatory decision about a product, not a recommendation, and it does not measure how well the product works in your own setting. Always confirm the current status and the exact cleared indication on the FDA record before you buy.