Radiology Partners Asks FDA to Clarify AI Vision Model Rules

Radiology Partners has formally petitioned the U.S. Food and Drug Administration to clarify how existing medical device regulations should apply to vision-language models used in diagnostic imaging.

The citizen petition, filed August 12 by the radiology group's technology division Mosaic Clinical Technologies, seeks guidance on whether commercially available AI models that analyze medical images require premarket clearance or approval. The move reflects growing uncertainty among healthcare organizations, developers, and clinicians about regulatory requirements for foundation models in clinical settings.

The regulatory gray zone

At the heart of the petition is a specific question: Does an AI model marketed with the expectation that healthcare institutions will fine-tune it on their own data constitute a medical device requiring FDA authorization?

Mike Peresie, president of Mosaic Clinical Technologies, told the FDA that differing interpretations of existing requirements have created uncertainty across the healthcare ecosystem. The petition asks the agency to clarify how regulatory principles apply to commercially distributed imaging vision-language models and to ensure consistent application of medical device standards.

Safety and validation concerns

Radiology Partners expressed concern that foundation models—large-scale AI systems trained to perform various downstream tasks—currently lack established validation standards. Without regulatory oversight, these models may be trained on biased or unrepresentative datasets and may lack transparency, explainability, and basic cybersecurity controls, according to the petition.

The group noted that several developers disclaim responsibility for their models' reliability and provide few warranties regarding performance or fitness for clinical use. This places substantial responsibility on radiologists and other downstream users to determine whether models are appropriate for their intended purposes.

"Models introduced without premarket review may not have demonstrated the baseline safety and effectiveness required for FDA authorization," Peresie wrote.

Why it matters

The petition highlights a critical tension in healthcare AI: Foundation models represent promising technological advances, but their rapid commercialization has outpaced regulatory frameworks. Without clear standards, healthcare organizations face inconsistent compliance expectations, and patients may be exposed to unvalidated diagnostic tools. Radiology Partners' request for clarity could influence how the FDA approaches oversight of generative AI across medical imaging and beyond.

Seeking consistency

The radiology group emphasized that clear expectations around validation, performance monitoring, transparency, and quality management would support patient safety while promoting responsible innovation. Consistent regulatory interpretation would benefit patients, healthcare providers, developers, and researchers, according to the petition.

Radiology Partners concluded by requesting FDA action to clarify stakeholder responsibilities in deploying diagnostic foundation models, noting that such guidance would provide greater consistency across the healthcare ecosystem.

These details were first reported by Radiology Business.