Is your cloud software a medical device?

  • Under EU MDR and FDA rules, software is a medical device only if it has its own medical purpose, such as diagnosis or treatment.
  • MDCG 2019-11 states that communication systems moving or storing data without interpreting it are not medical devices.
  • A single module, like an alarm generated from physiological data, can qualify as a medical device even inside an otherwise non-device system.
  • Classification runs feature by feature, not company by company or product by product.
  • A DICOM pipeline that validates transfers and writes bytes without reading pixel data reads like a communication system, not a device.

Under EU law, software is a medical device only if it has a medical purpose of its own — diagnosis, prevention, monitoring, or treatment. Under FDA rules, the test is close to the same: software “intended to be used for one or more medical purposes” that doesn’t depend on dedicated hardware. Software that only moves, stores, or displays data without interpreting it usually falls outside both definitions. Software that analyzes, scores, or flags something clinical usually falls inside.

This article explains how both frameworks reach that line. It is engineering-level orientation, not legal advice. A notified body in the EU, or the FDA in the US, makes the real determination for a specific product — this article can only explain the framework they use.

The EU test: Article 2(1) and “intended purpose”

Regulation (EU) 2017/745 (MDR) defines a medical device by what it’s meant to do, not by what it’s built from. The MDCG 2019-11 guidance — the official EU guidance on qualifying software under the MDR — puts it plainly:

“Software must have a medical purpose on its own to be qualified as [Medical Device Software]. […] In order to be qualified as [Medical Device Software], the product must first fulfil the definition of software according to this guidance and the definition of a medical device according to Article 2(1) of Regulation (EU) 2017/745.”

The same guidance draws a sharp line between two kinds of software behavior. Moving or storing information, without interpreting it, doesn’t create a medical purpose on its own:

“Communication systems are normally based on software for general purposes, and do not fall within the definition of a medical device.”

But a module added on top, one that reads clinical meaning into that data, can qualify on its own even inside a system that is otherwise just plumbing:

“A software module generating alarms based on the monitoring and analysis of patient specific physiological parameters is qualified as a medical device (MDSW).”

That second sentence is the one worth reading twice. Qualification in the MDR is not all-or-nothing for a whole platform. A single module — one alert, one score, one automated interpretation — can be a device even when the pipeline around it is not.

The US test: Software as a Medical Device (SaMD)

The FDA’s own page on Software as a Medical Device adopts the definition from the International Medical Device Regulators Forum (IMDRF):

“Software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device.”

This mirrors the EU logic. The word doing the work is “intended” — your own stated purpose, labeling, and marketing claims decide the question as much as the code itself. Software that you market as “just a data pipe” and that genuinely never interprets clinical content sits outside SaMD. The moment you add or claim a diagnostic, monitoring, or treatment function, the same code can cross into it.

Where a DICOM data pipeline usually lands

Most infrastructure that moves DICOM studies from a scanner or hospital PACS into cloud storage does one job. It receives an association, validates the transfer, and writes the bytes somewhere durable. It doesn’t read pixel data. It doesn’t score a study. It doesn’t decide whether a finding is urgent. Under both the EU and US tests above, that description reads like a communication system, not a device — the same category MDCG 2019-11 names directly.

That reading changes the moment you add a feature that touches clinical meaning. Auto-routing “urgent” studies by a modality field, flagging incomplete series as failed, or extracting measurements from pixel data are all plausible features. Each one moves that specific feature closer to a medical purpose, even if the transport layer around it stays plain transport. Classification runs feature by feature, not company by company.

MDCG 2019-11 lists the kind of image-handling functions that do carry a medical purpose on their own, once they go beyond plain transport:

“software which searches medical images for findings that support a clinical hypothesis as to the diagnosis or evolution of therapy […] software which locally amplifies the contrast of the finding on an image display, performs contrast stretching, edge enhancement, gray scale manipulation, smoothing and/or sharpening.”

None of that list is about moving bytes. It’s about changing what a viewer or a clinician sees, in a way meant to support a clinical decision. A viewer feature that only reformats an image for display, with no claim about supporting diagnosis, sits in a greyer area. The guidance covers that case separately, for the same reason a specific determination needs a specific product in front of it, not a general list.

The MDR classification ladder, once you’re in scope

If a module does qualify as Medical Device Software, MDR Annex VIII, Rule 11, sets its default class. The rule text, quoted directly from the regulation:

“Software intended to provide information which is used to take decisions with diagnosis or therapeutic purposes is classified as class IIa, except if such decisions have an impact that may cause: — death or an irreversible deterioration of a person’s state of health, in which case it is in class III; or — a serious deterioration of a person’s state of health or a surgical intervention, in which case it is classified as class IIb. Software intended to monitor physiological processes is classified as class IIa, except if it is intended for monitoring of vital physiological parameters […] in which case it is classified as class IIb. All other software is classified as class I.”

MDCG 2019-11 splits this single rule into three named sub-rules: 11a (decisions with a diagnosis or therapeutic purpose), 11b (monitoring physiological processes), and 11c (everything else). Each sub-rule applies separately, not as one combined test. Software that feeds a diagnosis or treatment decision defaults to class IIa under 11a, and only escalates to IIb or III if a wrong decision could seriously harm someone. Software that monitors physiological processes follows its own, similar ladder under 11b. Everything else — including a pipeline with no diagnostic or monitoring purpose at all — lands in class I by default under 11c, the lightest MDR class, with the least paperwork.

This matters for a data pipeline specifically. Qualifying under 11c, because none of your features carry a diagnostic or monitoring purpose, is a very different outcome from qualifying under 11a because one feature does. Both count as “medical device software” in MDR terms. One carries far more process overhead than the other.

Why this decision is worth making early

Classification decides your compliance workload. It sets which quality-system clauses apply, whether a notified body must review anything before you can sell, and how much clinical evidence you need on file. Getting pulled into class IIa or higher, when your actual feature set doesn’t need it, means carrying certification cost for no safety benefit. Missing a genuine class IIa feature means shipping without the controls regulators expect for that risk. Both mistakes are avoidable. Ask the qualification question — does this specific feature have a medical purpose? — before you ship it, not after an auditor asks.

If you’re unsure where a specific feature lands, that’s a question for a regulatory consultant or notified body. No general guide, including this one, can answer it for your product.

For the DICOM transport concepts this decision usually turns on, see DICOM C-STORE vs DICOMweb STOW-RS and what is DICOM Modality Worklist (MWL)?.

Dicomly builds infrastructure for healthcare software teams that receive DICOM studies from hospitals. Why we built Dicomly →

Sources

  1. Regulation (EU) 2017/745 on Medical Devices (MDR) (eur-lex.europa.eu)
  2. MDCG 2019-11 Guidance on Qualification and Classification of Software under MDR (health.ec.europa.eu)
  3. FDA — Software as a Medical Device (SaMD) (www.fda.gov)

Start receiving DICOM today.

Early access. No credit card. First endpoint free.