FDA’s PCCP rule for AI radiology: what patients and clinicians should know

FDA’s June 17, 2026 final order clarifies how certain AI tools used in radiology can be updated over time under a regulatory mechanism called a Predetermined Change Control Plan (PCCP). The practical point for patients, caregivers, and clinical teams is that “AI updates” are not treated as one-size-fits-all: FDA’s framework emphasizes evidence, risk management, and labeling so users understand what changed and when performance may not match expectations.

What FDA decided on June 17, 2026

On June 17, 2026, FDA published a final order in the Federal Register that codifies a regulatory classification for a specific type of radiology software—radiological machine learning–based quantitative imaging software with a predetermined change control plan.

Under this framework, the device type is placed in Class II with special controls. Those special controls describe what manufacturers must include to support FDA’s “reasonable assurance” standard, including requirements related to evidence, labeling, and risk management tied to the kinds of changes covered by a PCCP.

What a PCCP means in real life (and what it doesn’t)

A PCCP is meant to let certain software updates happen in a controlled way—but only when changes stay within the predetermined plan described and supported by the manufacturer’s evidence package. The goal is to reduce the risk that software changes could:

Alter outputs in ways that may not meet the intended performance for the intended clinical use.
Lead to misuse or misunderstanding (for example, when users don’t realize performance may vary after an update, or when local workflow/input conditions differ from what was supported).

What the “special controls” require for evidence and risk management

FDA’s special controls for this PCCP-based radiology software type are designed around the idea that the manufacturer must show—up front—how it will verify and validate both:

The base software performance, and
The planned modifications that will occur under the PCCP.

In general terms, the evidence expectations include:

Design verification and validation, including documentation of the software’s intended inputs/outputs and performance measurement approaches (including testing intended to reflect clinically important groups).
Planned modifications described under the PCCP, with methods for developing, verifying, and validating those modifications in line with the plan’s performance requirements.
Risk management for the planned changes, including identifying how the planned modifications could introduce new or different risks and the mitigation steps tied to those risks.

Labeling requirements: the “what to check” list

FDA’s order places substantial emphasis on labeling that helps users interpret the tool appropriately—especially if the software changes over time. Labeling is expected to cover, in substance:

Intended use and the patient population for which the software was validated.
User information about the role of the software in the workflow (and the expertise needed for safe use).
Inputs and outputs, and the compatible imaging contexts/hardware and protocols relevant to intended use.
Performance summaries, including how performance may vary depending on relevant factors and clinically important conditions.
Failure/performance-limitation conditions—i.e., scenarios where the software may not perform as expected—described in a way that supports safer interpretation of results.
PCCP-related information, including that the device has a PCCP, descriptions/summaries of modifications implemented, and version history, along with how users are expected to be informed about PCCP changes.

When those elements are done well, labeling can help teams answer a straightforward question: Did the software change, and if so, what performance expectations apply now?

How clinicians can think about this in real life

Even with good regulatory design controls, real-world performance can vary with local conditions—like image acquisition practices, equipment, workflow, and patient mix. Peer-reviewed radiology literature has emphasized that transparency about validation approaches and the intended role of AI is important for safe use.

Meanwhile, real-world monitoring efforts (such as those described by radiology organizations focused on AI quality) reflect the same safety idea from a different angle: clinical teams should be able to track stability over time, including factors like software version and workflow-relevant inputs, so they can recognize when outcomes might drift from expectations.

Questions patients and caregivers can ask

If you’re discussing AI-supported imaging with your care team, you can ask practical, safety-focused questions such as:

What is the exact AI tool/software name and its intended role?
Does the tool use a PCCP? If yes, where can you find version history and update-related information (for example, in practice documentation or labeling)?
What performance limits or failure conditions are described in the labeling?
How do clinicians verify outputs in context, especially after an update?

Questions clinicians and teams can ask

For clinicians-adjacent readers—radiology leaders, governance teams, PACS/EHR stakeholders—the PCCP labeling framework suggests concrete checkpoints:

Does the labeling clearly define intended use, patient population, and the relevant imaging conditions?
Are performance limitations/failure scenarios explicit enough to support safer interpretation in your local workflow?
Is there a clear process to track software versions and ensure teams know when updates occurred?
Does local oversight include verification steps that account for labeled limitations and likely input shifts over time?

What remains uncertain

The PCCP and special-controls framework focuses on what manufacturers must demonstrate and document—so users get clearer information and risk controls tied to planned modifications. But it does not guarantee identical performance for every patient in every setting. Real-world results can still vary with factors such as imaging protocols/equipment, workflow, and how closely patients match labeled validation groups.

Bottom line

FDA’s June 17, 2026 action codifies a Class II special-controls pathway for certain AI radiology quantitative imaging software that uses an FDA-authorized PCCP. For everyday decision-making, the key is the combination of evidence, labeling (including PCCP/version information and performance limitations), and clinician verification.

A reasonable next step for families is to ask how AI results are interpreted in context and where version/update information and limitations are documented. For clinical teams, align local workflow with labeling and maintain a practical approach to monitoring performance over time.

Sources

Editorial note: Weence articles are researched from cited public-health, medical, regulatory, journal, and reputable news sources and may be drafted with AI assistance. They are checked for source support, clarity, and safety guardrails before publication.

This article is for general informational purposes only and is not medical advice. Research findings can be early or incomplete, and health guidance can change. Always talk with a qualified healthcare professional about personal symptoms, diagnosis, medications, vaccines, screenings, or treatment decisions. If you think you may have a medical emergency, call emergency services right away.