Vendor-Neutral Radiology SaMD: FDA Clearances and Predicates
For regulatory and clinical teams evaluating AI-based radiology software, a foundational question is whether FDA clearance requires the software sponsor to control the upstream image-acquisition hardware. The answer has direct implications for market access strategy, predicate selection, substantial equivalence arguments, and the scope of clinical validation studies — particularly as pure-play software companies increasingly position their products as hardware-agnostic.
This analysis examines FDA 510(k) and De Novo clearances in the radiology SaMD space, focusing on product codes LLZ, QIH, QAS, and QFM, to identify cleared devices where the sponsoring entity is not the manufacturer of the CT, MR, X-ray, or mammography system generating the input images. The review draws on FDA Product Classification records and 510(k) summaries from January 2023 forward, and separately identifies imaging OEM sponsors to establish a clear contrast with vendor-neutral software-only entrants.
Want to ask Rhizome your own regulatory questions? Try it for free.
Vendor-neutral radiology SaMD: FDA-cleared software that runs on images the sponsor didn't acquire
The short answer
In diagnostic radiology, the "sponsor does not own the acquisition device" model is not an edge case: it is the dominant architecture for AI/image-analysis SaMD. These products are cleared as software that ingests DICOM images produced by whatever scanner the customer already owns. FDA reviews them as vendor-agnostic, and sponsors validate performance across multiple scanner OEMs precisely because they do not control the hardware. The clearest concentration is in the triage/notification and prioritization codes (QAS, QFM), which are almost entirely populated by independent software companies (Aidoc, Viz.ai, iSchemaView/RapidAI, Qure.ai, Brainomix, Annalise/Harrison-AI, Avicenna.AI, VUNO), none of which manufacture CT or X-ray hardware.
Why the product codes matter
All four codes sit in radiology (medical specialty RA) and are Class II. The regulation text makes the decoupling from hardware explicit:
- QAS and QFM (both under 21 CFR 892.2080) are defined as image-processing devices that act on "medical images acquired from radiological signal acquisition systems," aiding prioritization and triage without returning diagnostic image findings 34. The definition presumes the acquisition system is a separate device.
- QIH (21 CFR 892.2050) covers automated radiological image-processing software implementing non-adaptive machine-learning algorithms trained on clinical or artificial data; performance depends on the training data, not on any bundled scanner 2.
- LLZ (21 CFR 892.2050) is the long-standing "System, Image Processing, Radiological" code that predates the AI-specific codes and captures general PACS/visualization/post-processing software 1.
Scope in the sampled window (unique K-numbers, decisions since 2023-01-01): LLZ ~231 6, QIH ~248 255, QAS ~46 503, and QFM ~18 237 clearances.
Confirmed examples: software sponsors that do not own the acquisition hardware
Each of the following was read from its 510(k) summary. In every case the sponsor is a software company, the input is DICOM from third-party scanners, and there is no evidence the sponsor manufactures the acquisition hardware.
| Device (K-number) | Sponsor | Modality | Acquisition source in the record | Named third-party scanner OEMs |
|---|---|---|---|---|
| BriefCase-Triage (K261317, QAS) | Aidoc Medical, Ltd. 549 | Head CTA / mCTA 552 | DICOM-compliant CT scanner images; no sponsor-built scanner 553 | DICOM-compliant CT, 64-slice or higher 553 |
| Viz AAA (K223443, QFM) | Viz.ai, Inc. 572 | CT/CTA abdomen 569 | Images from DICOM-compliant imaging devices, forwarded to Viz.ai server 567 | GE, Siemens, Toshiba (incl. Discovery 610, Optima CT660, Revolution EVO/Maxima, SOMATOM Definition AS/Flash/Perspective, Sensation 40/64, Aquilion ONE) 567 |
| Rapid CTA 360 (K251151, QAS) | iSchemaView, Inc. 621 | Head/neck CTA 621 | DICOM input from multiple sites/sources; no sponsor scanner 620621 | GE, Toshiba, Siemens, Philips 620 |
| qER-CTA (K251610, QAS) | Qure.ai Technologies 562 | Head CTA 559 | DICOM input; no sponsor-built acquisition hardware 562 | None named; DICOM input only 559 |
| Brainomix 360 Triage Stroke (K251983, QAS) | Brainomix Limited 598 | Non-contrast head CT 604 | DICOM routing of acquired NCCT; no sponsor scanner 603604 | Siemens, GE, Philips 604 |
| Annalise Enterprise (K253818, QAS) | Harrison-AI Medical Pty Ltd 610 | Non-contrast brain CT 617 | DICOM CT from image/order-management systems; five US sites 613 | GE Healthcare, NeuroLogica, Siemens, Toshiba (test set) 609 |
| CINA-VCF (K240612, QFM) | Avicenna.AI 594 | Chest/abdomen CT 595 | Images from radiological signal-acquisition systems / DICOM nodes; software on off-the-shelf server 589596 | Siemens, Philips, GE, Canon/Toshiba (training data) 589 |
| VUNO Med-Chest X-ray Triage (K241439, QFM) | Vuno Inc. 583 | Chest X-ray 579 | Vendor-agnostic; runs after acquisition on other radiological equipment 579 | GE, Philips, Fujifilm, Canon, Samsung, Siemens, Konica Minolta, Carestream (training/test) 579 |
The recurring pattern is important for a regulatory read: sponsors describe the device as accepting "any DICOM-compliant" input and then report validation performance across a named set of scanner OEMs. That combination is how the sponsor demonstrates the software generalizes across acquisition hardware it neither owns nor controls. Viz AAA and VUNO are the most explicit, listing broad multi-vendor scanner sets 567579; Aidoc and Qure.ai lean on the generic "DICOM-compliant" framing with fewer named models 553559.
The contrast: when the sponsor does own the hardware
Not every clearance in these codes fits the pattern. The LLZ and QIH codes also collect post-processing and visualization software submitted by the imaging OEMs themselves, where the sponsor does make the scanner:
- GE Healthcare, "View" (K253639, LLZ) 22
- Siemens Medical Solutions / Siemens Healthineers, syngo.MR Applications (K253495) 23 and AI-Rad Companion Prostate MR (K252608) 29
- Canon Medical Informatics, Vitrea CT TAVR Planning (K260324, QIH) 262
- Fujifilm, Synapse PACS (K261713) 255 and Synapse 3D Base Tools (K254189) 270
For these, the software and the acquisition system come from the same corporate manufacturer, so the "sponsor does not own the device the data is generated with" framing does not apply. The distinction to carry into any competitive or predicate analysis: the triage/prioritization codes (QAS, QFM) are dominated by independent software vendors, while the general image-processing codes (LLZ, QIH) are a mix of independent SaMD developers and OEM post-processing suites.
What this means for a regulatory strategy
- If the target profile is "software only, runs on the customer's existing fleet," QAS and QFM are the natural home, and the predicate landscape is almost entirely vendor-neutral SaMD. Expect FDA to want multi-scanner, multi-site validation to support a generalizability claim, because the sponsor cannot constrain the acquisition device 567579609.
- The intended-use language in these clearances consistently ties compatibility to DICOM conformance rather than to a specific scanner, which is what lets the same clearance be marketed against any OEM's installed base 553596.
- LLZ remains a broad, older catch-all; a device landing there may be competing against both independent SaMD and OEM-bundled post-processing, so predicate selection should account for who manufactured the acquisition hardware in each cited predicate 122255.
Limitations
This is a scoped sample from 2023 forward across four product codes, not an exhaustive registry pull, and clearance counts reflect unique K-numbers in that window rather than the full history of each code. "Does not own the acquisition hardware" was confirmed by reading eight representative 510(k) summaries; a definitive determination for any single device requires reading that device's own summary and, where needed, its labeling. Named scanner OEMs frequently appear as training/test data sources rather than as a formal compatibility list, so they evidence multi-vendor validation, not a guarantee of clinical compatibility with every listed model. A reader who needs a complete, de-duplicated inventory by sponsor, or the corresponding EU MDR / UKCA status of these same products, would want to ask Rhizome those as direct follow-ups.