FDA Pathways for SaMD: 510(k) vs De Novo vs PMA

The regulatory pathway you choose determines your timeline, your budget, your evidence burden, and who gets to use your product code after you. For ContourCompanion, this decision shaped the entire project. Get it wrong and you're looking at six to twelve months of wasted effort building the wrong evidence package.

Three pathways exist for getting a SaMD to market in the US: 510(k), De Novo, and PMA. Each has a different standard of review, different costs, and different strategic implications. The right choice depends on one question: does a legally marketed predicate device already exist for what you're building?

The Three Pathways at a Glance

	510(k)	De Novo	PMA
When to use	Predicate exists, Class II	No predicate, low-to-moderate risk	High risk, Class III
Review time	90-180 days	150-300 days	180-360 days
FDA fee (FY2025)	$24,335 ($6,084 small business)	~$132,464 (~25% small business discount)	~$540,783
Evidence standard	Substantial equivalence	Safety and effectiveness	Rigorous clinical evidence
Annual registration	$9,280	$9,280	$9,280

These are FDA review times only. Real-world total elapsed time is longer. For AI/ML SaMD via 510(k), the median total time from submission to clearance is 142-151 days. A quarter of devices clear in under 90 days. The rest hit Additional Information (AI) requests that pause the review clock while you respond.

510(k): The Default Path for 97% of AI/ML SaMDs

97% of all FDA-authorized AI/ML medical devices came through the 510(k) pathway. The reason is straightforward: the bar is "substantial equivalence" to a predicate, not proof of safety and effectiveness from scratch.

Substantial equivalence means your device has the same intended use as the predicate and either identical technological characteristics or different characteristics that don't raise new questions of safety or effectiveness. For SaMD, the technology is almost always different (your algorithm isn't the same as the predicate's), so you demonstrate that your different approach produces equivalent or better clinical results.

Finding Your Predicate

This is where teams either save months or lose them. A predicate can be any device legally marketed before May 28, 1976, a device reclassified from Class III, a device cleared through a prior 510(k), or a device authorized via De Novo. The FDA's 510(k) database and Device Classification database are your starting points.

For ContourCompanion, predicate selection was critical. The autocontouring market has at least 15 FDA-cleared products: Mirada's DLCExpert (K181572), Limbus AI's Limbus Contour, Manteia's AccuContour (K191928), and others across the radiology panel (product code QKB). Having established predicates meant 510(k) was viable even though our IMDRF risk categorization was Category IV. The predicates shared our intended use (autocontouring OARs for radiation therapy planning), so the substantial equivalence argument was strong.

The trap: picking a predicate with a narrower intended use than yours. If your predicate was cleared for "displaying images for review" and your device "generates treatment recommendations," the intended uses don't match. FDA will reject the SE argument, and you've wasted the submission fee and months of preparation.

Use the Pre-Submission (Q-Sub) process to validate your predicate strategy before committing. More on this below.

What Goes in a 510(k) for SaMD

The eSTAR template became mandatory for all 510(k) submissions on October 1, 2023. It's an interactive PDF (requires Adobe Acrobat Pro) that walks you through every required section. Color-coded: red for missing, green for complete.

Key sections for SaMD:

1. Device description and intended use with predicate comparison
2. Software documentation per the June 2023 guidance. Two documentation levels (Basic and Enhanced) replaced the old three-tier "Level of Concern" system. Enhanced documentation is required when a software failure could present a hazardous situation with probable risk of death or serious injury
3. Performance testing (bench testing, analytical validation, clinical validation if applicable)
4. Risk analysis summary (the full ISO 14971 file stays in your Design History File, but the submission needs a Device Hazard Analysis)
5. Cybersecurity documentation per the September 2023 guidance: threat model, SBOM, vulnerability assessment, SPDF evidence
6. Electromagnetic compatibility (not typically applicable for standalone SaMD)

Real Cost of a 510(k)

The FDA fee is the smallest line item. A realistic budget for a SaMD 510(k):

Item	Cost Range
FDA user fee	$6,084 - $24,335
Annual establishment registration	$9,280
Regulatory consultant (if used)	$12,000 - $30,000
Performance testing	$50,000 - $100,000+
eSTAR preparation and assembly	Included in consultant or internal effort
Total	$75,000 - $250,000+

Testing is the largest variable. For AI/ML SaMD, clinical validation studies (dataset acquisition, annotation, statistical analysis) drive this number. A simple Class II device with bench testing only will land at the low end. A Category IV AI SaMD with clinical performance data across multiple anatomical sites will be at the high end or beyond.

De Novo: When You're Building Something New

De Novo exists for devices that are novel (no predicate) but low-to-moderate risk. The standard is higher than 510(k): you must demonstrate safety and effectiveness, not just equivalence. But the payoff is strategic.

A successful De Novo creates a new product classification and product code. Your device becomes the predicate for every future 510(k) submission in your category. Viz.ai used De Novo in February 2018 to create the category for AI-powered triage notification software. Every subsequent AI triage tool (and there are now dozens) used Viz.ai's clearance as their predicate. That's a meaningful competitive moat: you define the special controls that everyone after you must meet.

When De Novo Makes Sense for SaMD

• Your SaMD does something no cleared device does
• No existing product code covers your intended use
• The risk profile is low-to-moderate (Class I or II after classification)
• You want to own the product code and define the special controls

When De Novo Doesn't Make Sense

• A reasonable predicate exists and 510(k) would work. De Novo takes longer and costs 5x more in FDA fees alone
• Your device is clearly high-risk (Class III). De Novo can only create Class I or II classifications
• You can't afford the timeline. De Novo reviews take 150-300 days, and the evidence burden is heavier

De Novo Requirements Beyond 510(k)

Everything in a 510(k) submission applies (minus the predicate comparison), plus:

• Proposed classification: you recommend the product code, regulatory class, and panel
• Proposed special controls: you draft the conditions future devices in your category must meet
• Risk-benefit analysis: since there's no predicate to compare against, you explicitly argue that benefits outweigh residual risks
• More extensive clinical evidence: FDA typically requires prospective or large retrospective clinical validation

The eSTAR format became mandatory for De Novo submissions on October 1, 2025.

PMA: The Full Clinical Trial Path

PMA (Premarket Approval) is for Class III, high-risk devices. The evidence standard is "valid scientific evidence," which almost always means prospective clinical trials. The FDA fee alone is ~$540,783. Total cost to bring a PMA device to market typically runs $5M-$119M+ depending on the clinical trial requirements.

For SaMD, PMA is rare. Most software-only devices classify as Class II even when the clinical risk is significant, because the IMDRF risk categorization and FDA device classification don't always align. ContourCompanion was IMDRF Category IV but FDA Class II, cleared via 510(k). The existence of predicate devices in the radiology panel was the deciding factor, not the risk category.

PMA makes sense when:

• FDA has explicitly classified your device type as Class III
• No De Novo pathway is feasible (the risk is genuinely too high for Class II)
• You're working with an implantable or life-sustaining system where software failure has irreversible consequences

If you're reading this series and your SaMD requires PMA, you're likely already working with a specialized regulatory firm. The rest of this series focuses on the 510(k) and De Novo pathways that cover 99%+ of SaMD submissions.

The Pre-Submission: Your Free Insurance Policy

Before committing to any pathway, file a Q-Sub (Pre-Submission). It's free. No user fee. The only cost is 2-4 weeks of preparation time.

Here's the process:

1. Submit your Q-Sub package: device overview (3-5 pages), proposed classification, predicate strategy (if 510(k)), proposed testing plan, and 3-5 specific questions for FDA
2. FDA acceptance review (~2 weeks)
3. FDA internal review (60-70 days)
4. Written preliminary feedback shared 2-3 business days before the meeting
5. Meeting (teleconference or in-person, typically 1 hour)

The May 2025 Q-Sub guidance added Pre-Sub Supplements as a formal Q-Sub type, so you can follow up with additional questions after your initial meeting without starting over.

For ContourCompanion, the Pre-Sub confirmed our 510(k) pathway and predicate strategy before we built the performance testing package. That single meeting prevented us from spending months generating clinical evidence for the wrong regulatory standard.

When to use a Pre-Sub for SaMD:

• Novel technology or novel clinical claims
• Unclear classification or pathway
• AI/ML-specific concerns (training data adequacy, PCCP feasibility)
• Complex predicate strategy (multiple predicates, split predicates)
• Planning a De Novo and unsure about proposed special controls
• Cybersecurity architecture questions

Teams that skip the Pre-Sub to save 60-75 days routinely lose 6+ months to Additional Information requests during substantive review. The math is obvious.

How We Made the Decision for ContourCompanion

The decision tree was straightforward once we mapped it:

1. Does a predicate exist? Yes. Multiple cleared autocontouring products in product code QKB (radiology panel). Mirada, Limbus, Manteia, and others had established the category.
2. Same intended use? Yes. Autocontouring Organs at Risk for radiation therapy treatment planning.
3. Can we demonstrate substantial equivalence? Yes, through clinical performance data (Dice Similarity Coefficients, Hausdorff distances) showing equivalent or superior contour accuracy.
4. Is 510(k) appropriate despite Category IV IMDRF risk? Yes. FDA classifies by product code, not IMDRF category. QKB is Class II. The IMDRF categorization drove our risk management rigor, not our regulatory pathway.

The decision was 510(k). The Pre-Sub confirmed it. Total FDA review time was well within the typical range for AI/ML SaMD.

If we'd been building the first autocontouring tool (no predicates existed), the path would have been De Novo. We'd have defined the product code, proposed the special controls, and accepted the longer timeline and higher fee. That's exactly what Viz.ai did for AI triage, and the strategic value of owning the classification justified the cost.

Common Mistakes That Add Six Months

Picking a predicate with mismatched intended use. The intended use must be the same, not similar. "Aids in detection" is not the same as "diagnoses." FDA will catch this at the RTA (Refuse to Accept) screening on day 15 and send your submission back.

Skip the Pre-Sub and you're leaving free money on the table. Already covered above, but it bears repeating: free feedback from the people who will review your submission.

Underestimating the software documentation. The June 2023 guidance replaced the old "Level of Concern" framework with a two-tier system (Basic vs Enhanced). Enhanced documentation requirements are substantial: complete software requirements specification, architecture design chart, traceability analysis, software testing with full test protocols and results, unresolved anomalies list. If your software failure could cause death or serious injury, you're writing Enhanced documentation.

The cybersecurity package catches teams off guard too. Since September 2023, FDA has refused to accept 510(k) submissions that lack cybersecurity documentation. SBOM, threat model, vulnerability assessment, and a plan for post-market security monitoring are all required. This isn't a section you can bolt on at the end.

Finally, budget for AI requests. About half of AI/ML SaMD submissions receive an Additional Information request around day 60 of review. The review clock stops while you respond. Budget 30-60 additional days for this in your project plan.

The pathway decision is the foundation. Everything else in this series builds on it: the IEC 62304 software lifecycle, the ISO 14971 risk management file(coming March 20), the cloud architecture, the clinical evidence package. Each of those deliverables scales with the pathway you've chosen.

Frequently Asked Questions

What is the difference between 510(k) and De Novo?

A 510(k) requires a legally marketed predicate device and demonstrates your device is substantially equivalent. A De Novo is for novel devices with no predicate but low-to-moderate risk, requiring you to demonstrate safety and effectiveness from scratch. De Novo costs roughly 5x more in FDA fees and takes 150-300 days versus 90-180 for 510(k), but the strategic payoff is that your device becomes the predicate for every future 510(k) in your category.

How much does a 510(k) submission cost?

The FDA user fee for a 510(k) ranges from $6,084 (small business) to $24,335 (FY2025). But the fee is the smallest line item. A realistic total budget for a SaMD 510(k) is $75,000-$250,000+, covering performance testing ($50,000-$100,000+), regulatory consulting ($12,000-$30,000), and annual establishment registration ($9,280). AI/ML SaMD with clinical validation studies across multiple sites will land at the high end.

When should I use a Pre-Submission (Q-Sub)?

File a Pre-Submission whenever you have a novel technology, unclear classification, complex predicate strategy, or AI/ML-specific concerns. Q-Subs are free and give you written FDA feedback in 60-75 days. Teams that skip the Pre-Sub to save time routinely lose 6+ months to Additional Information requests during substantive review. For ContourCompanion, the Pre-Sub confirmed our pathway and predicate strategy before we committed engineering resources.

Can a Category IV SaMD be cleared through 510(k)?

Yes. IMDRF risk categories don't map directly to FDA device classes. ContourCompanion was IMDRF Category IV (highest risk tier) but FDA Class II, cleared through 510(k). The existence of predicate devices in the same product code (QKB, radiology panel) made 510(k) viable. The IMDRF categorization drove the rigor of our risk management and clinical evidence, not the regulatory pathway.

References

• FDA 510(k) Program
• FDA De Novo Classification Request
• FDA PMA Program
• FDA eSTAR Template
• FDA Q-Sub Guidance (May 2025)
• FDA MDUFA Fee Schedule FY2025
• FDA AI/ML-Enabled Medical Devices
• Content of Premarket Submissions for Device Software Functions (June 2023)
• Cybersecurity Guidance for Medical Devices (September 2023)