FDA Software Routes: 510(k), De Novo, and Exemptions
- Software that only transfers, stores, or converts medical device data without clinical interpretation is not a device.
- Imaging software fails the Clinical Decision Support exemption because it analyzes a medical image, failing Criterion 1.
- Being 510(k) exempt only waives premarket notification, not ongoing FDA General Controls like adverse event reporting.
- A 510(k) requires proving substantial equivalence to a predicate device with the same intended use and characteristics.
- De Novo classification applies to novel moderate-risk software with no predicate and requires primary clinical evidence.
The United States Food and Drug Administration (FDA) regulates medical software through four distinct routes. Your code either qualifies as non-device software, falls into an exempt classification, requires a 510(k) clearance, or needs a De Novo classification. Software that only moves or stores data without clinical interpretation sits outside device regulations. Software that analyzes medical images or drives clinical decisions requires premarket clearance.
This article provides engineering orientation, not legal advice, and the FDA or a qualified regulatory consultant determines the classification for a specific product. It explains how FDA pathways evaluate software functions, how 510(k) clearance differs from De Novo classification, and why medical imaging software cannot claim decision support exemptions.
Which FDA pathway applies to your medical software?
The FDA regulates medical devices according to patient risk across three tiers:
- Class I: Low risk. Most Class I devices are exempt from premarket review.
- Class II: Moderate risk. Most Class II devices require a 510(k) clearance before commercial release.
- Class III: High risk. Devices that sustain life require Premarket Approval (PMA), the most stringent review tier.
Software as a Medical Device (SaMD) rarely enters Class III unless it controls life-support hardware. Most clinical algorithms, cloud diagnostic viewers, and artificial intelligence tools sit in Class II.
Understanding where your product sits prevents two expensive errors. It stops you from spending money on unnecessary submissions for basic plumbing, and prevents you from shipping an unapproved diagnostic algorithm.
Non-device software under Section 520(o) and MDDS
Before exploring device pathways, check whether your code is a medical device at all. In 2016, Section 3060 of the 21st Century Cures Act amended Section 520 of the Federal Food, Drug, and Cosmetic Act (FD&C Act). It added Section 520(o), which excludes specific software functions from the definition of a device.
Section 520(o)(1)(D) covers Medical Device Data Systems (MDDS). Under FDA guidance on Medical Device Data Systems, software intended solely for electronic transfer, storage, format conversion, or display of medical device data is not a device.
An ingestion service that receives DICOM instances over TLS, extracts routing tags, and stores raw objects in cloud storage is non-device software. It transfers and stores data without altering clinical contents. Lossless format conversion for transmission also stays non-device. In contrast, lossy compression that alters pixel values can cross into medical device territory.
If your product only ingests, archives, and delivers imaging studies to other systems, you do not need an FDA submission. You build software under standard commercial software practices. You can learn more about drawing this line in our guide on whether software qualifies as a medical device.
Why can’t medical imaging software use the CDS exemption?
Engineering teams often read about Clinical Decision Support (CDS) software and assume their diagnostic algorithm is exempt from FDA review. Section 520(o)(1)(E) of the FD&C Act excludes certain decision support tools from the device definition. Many founders believe that if software assists a radiologist, the FDA will not regulate it.
This assumption fails for medical imaging. The statute sets a strict four-part test. Software qualifies as Non-Device CDS only if it meets all four criteria:
- Criterion 1: It does not acquire, process, or analyze a medical image or signal.
- Criterion 2: It displays, analyzes, or prints medical information about a patient or clinical guidelines.
- Criterion 3: It provides recommendations to a health care professional about prevention, diagnosis, or treatment.
- Criterion 4: It enables the health care professional to independently review the basis for the recommendations.
Criterion 1 excludes medical images immediately. If your software ingests, parses, enhances, or analyzes a DICOM image, it fails Criterion 1. It cannot qualify for the statutory CDS exclusion.
The FDA reinforced this boundary in its Clinical Decision Support Software guidance. Computer-aided detection (CADe), diagnosis (CADx), and triage (CADt) tools evaluate image patterns. Because they process image pixels to detect lesions or flag urgent cases, the FDA regulates them as medical devices. If your algorithm touches imaging pixels to generate clinical outputs, plan for a formal device pathway.
510(k) exemptions: what does exempt actually mean?
Some software functions meet the definition of a medical device, but the FDA exempts them from premarket notification. The FDA grants Class I and Class II device exemptions when premarket review is not needed to ensure safety and effectiveness.
An exemption from 510(k) does not mean an exemption from FDA oversight. It only excuses you from submitting a 510(k) application and waiting for clearance before marketing the software. If your software is a 510(k)-exempt device, you must still comply with FDA General Controls:
- Establishment registration: Annual facility registration under 21 CFR 807.
- Device listing: Listing your software product in the FDA registry.
- Medical device reporting: Reporting adverse events under 21 CFR 803.
- Quality Management System Regulation (QMSR): Maintaining a quality system under 21 CFR 820.
In medical imaging, product code LLZ under regulation 21 CFR 892.2050 covers image management and processing systems. Viewing software that provides manual viewing, window-level adjustments, and basic manual measurements is 510(k)-exempt.
Read the limitations of exemptions in 21 CFR 892.9. An exemption disappears if your software introduces a new intended use or novel technology. If you add automated measurement tools or AI segmentation to an exempt viewer, the software loses its exemption. You must submit a 510(k).
The 510(k) pathway: demonstrating substantial equivalence
The 510(k) premarket notification is the primary clearance pathway for medical software in the United States. It applies to Class II devices that have a comparable device already on the market.
The core principle of a 510(k) is Substantial Equivalence (SE). You do not prove safety and effectiveness from first principles. Instead, you prove that your software is as safe and effective as a legally marketed predicate device.
Under Section 513(i) of the FD&C Act, your software is substantially equivalent if it meets two standards. It must share the same intended use as the predicate. It must also have the same technological characteristics, or differences that do not raise new safety and effectiveness questions.
Finding the right predicate device is the primary technical task in planning a 510(k). Search the public FDA 510(k) records for cleared products with similar clinical claims. For example, if you build software that flags intracranial hemorrhage on head CT scans, look for predicates under product code QFM. Review the public 510(k) summary for each predicate to see the performance benchmarks and testing standards that the FDA accepted.
Since October 2023, the FDA requires all 510(k) submissions to use the electronic Submission Template And Resource (eSTAR). The template requires software architecture descriptions, cybersecurity documentation, verification testing, and traceability matrices. The standard FDA review clock for a 510(k) is 90 days, though total time usually takes four to six months including sponsor responses.
The De Novo request: classifying novel software
What happens if your software is a moderate-risk medical device, but nobody has cleared a similar product before? Under statutory rules, any new device without a predicate defaults to Class III. That requires a costly Premarket Approval (PMA) application with prospective clinical trials.
The De Novo classification request solves this problem. Enacted under Section 513(f)(2) of the FD&C Act, the De Novo pathway provides a route for novel software that presents low-to-moderate risk. If general controls alone or combined with special controls ensure safety and effectiveness, you can request a De Novo classification into Class I or Class II.
The De Novo process requires primary clinical evidence. Because you have no predicate to compare against, you must prove the clinical validity of your software directly. You run formal reader studies, multi-center retrospective validations, and algorithmic stress tests.
A granted De Novo request grants marketing clearance and establishes a new classification regulation with special controls. Those special controls define testing, labeling, and performance requirements for all future devices of that type. Your software becomes the predicate device for subsequent 510(k) submissions from competitors. The review target is 150 FDA days, but expect nine to twelve calendar months in practice.
How to choose your regulatory path early
Choosing your regulatory pathway should happen during your initial software architecture phase. Changing your intended use late in development forces you to rewrite documentation or run new validation studies.
Follow these four steps to identify your route:
- Define your exact intended use: State precisely what clinical data your software consumes, what output it produces, and who acts on that output. Does it only transport DICOM objects, or does it highlight suspicious lesions?
- Check for statutory exclusions: If your code strictly moves, formats, or stores data without clinical interpretation, document it as non-device MDDS under Section 520(o)(1)(D). Keeping your ingestion infrastructure decoupled from analytical modules preserves this boundary.
- Search FDA classification records: Look up your intended use in the FDA Product Classification portal. If you find cleared devices with identical intended uses and similar technology, identify your candidate predicates for a 510(k).
- Evaluate the need for De Novo: If your algorithm introduces an entirely new clinical claim or analyzes data in a novel way, plan for a De Novo request.
Separating your core transport pipeline from clinical analysis modules keeps infrastructure code outside medical device regulations. When you connect hospital scanners to your platform, your ingestion layer operates as communication plumbing, while your analytical software follows formal medical device controls.
To see how modern imaging backends decouple network connectivity from downstream processing, explore how to configure DICOM endpoints and receive STOW-RS deliveries. Reviewing software risk management under ISO 14971 will also help you structure verification protocols for your upcoming FDA submission.
Dicomly builds infrastructure for healthcare software teams that receive DICOM studies from hospitals. Why we built Dicomly →