ACR Introduces Imaging AI Practice Parameter and Assess-AI Framework
The American College of Radiology (ACR) has approved the first-ever ACR-SIIM Practice Parameter for Imaging Artificial Intelligence, establishing guidance for the implementation, monitoring, and governance of AI tools in radiology. The announcement was made during ACR 2026 in Washington, DC.
At the same time, the ACR Data Science Institute (DSI) published a new article in the Journal of the American College of Radiology (JACR) outlining the technical framework for Assess-AI, an AI quality registry and data service designed to support post-deployment monitoring of imaging AI systems.
According to the ACR, both initiatives are intended to support the safe, effective, and transparent use of clinical AI in imaging workflows.
Guidance for AI Deployment and Governance
The new practice parameter was developed in collaboration with the Society for Imaging Informatics in Medicine (SIIM) and applies to radiologists, technologists, medical physicists, IT professionals, administrators, and data scientists involved in imaging AI deployment.
The document outlines recommendations for:
- AI governance and oversight
- AI tool selection and acceptance testing
- Ongoing performance monitoring
- Model drift and safety evaluation
- HIPAA privacy and security compliance
- Continuous quality improvement workflows
“This first-of-its-kind ACR-SIIM Practice Parameter outlines steps that imaging facilities can follow to help implement, use, and continually update AI to successfully deploy these rapidly evolving technologies in clinical care — everything from selection, to monitoring, to continuous quality improvement,” said Tessa Cook, MD, PhD, FSIIM, FACR, chair of the practice parameter writing committee and incoming chair of the ACR Commission on Informatics.
The framework also supports participation in the ACR Recognized Center for Healthcare-AI (ARCH-AI) designation program.
Assess-AI Supports Post-Deployment Monitoring
The newly introduced Assess-AI platform focuses on monitoring imaging AI performance after clinical deployment.
According to the ACR, the system:
- Measures concordance between AI outputs and radiology reports
- Uses de-identified data integrated through ACR Connect
- Applies LLM-based prompting for surrogate label extraction
- Enables local review of discordant cases through ACR Forensics
“Assess-AI provides facilities with interactive analytics that show how AI tools perform across their practices over time, enabling them to take control and manage performance, product selection and risk. Site data can be compared to aggregated national performance benchmarks from other sites using AI for identical use cases,” said Christoph Wald, MD, PhD, MBA, FACR, vice chair of the ACR Board of Chancellors and chair of the ACR Commission on Informatics.
Assess-AI currently supports multiple radiology AI use cases, including:
- Intracranial hemorrhage
- Pulmonary embolism
- Pneumothorax
- Large vessel occlusion
- Breast density
- Cervical spine fracture
- Pleural effusion
- Obstructive hydrocephalus
Focus on Clinical AI Oversight
The ACR stated that Assess-AI is now part of the ACR National Radiology Data Registry portfolio and complements other AI-focused resources such as AI Central.
“Responsible use of AI in healthcare, particularly when dealing with critical patient data like medical imaging, is an ongoing process rather than a single event. It demands a dedicated team consistently applying processes supported by methods and technology,” said SIIM Board Chair Nabile Safdar, MD. “This practice parameter is based on the principles of AI science, interoperability standards, expert involvement, and methodological precision. This collaborative effort between SIIM and the ACR transforms these foundations into practical guidance that any imaging practice can implement to improve patient care now."
The ACR noted that it continues to work with SIIM, the U.S. Food and Drug Administration, and other stakeholders on the future development of radiology AI standards and governance.
