ACR Council Approves First Practice Parameter for Imaging AI

The American College of Radiology Council voted Tuesday to approve what it calls a groundbreaking framework for assessing artificial intelligence in medical imaging. The decision came during ACR's annual meeting in Washington, D.C., running through May 6, 2026. Imaging leaders formally OK'd the new Practice Parameter for Imaging Artificial Intelligence, developed jointly with the Society for Imaging Informatics in Medicine.

This isn't just another policy document gathering dust on a shelf. The practice parameter applies to radiologists, technologists, physicists, informaticists, administrators, and anyone deploying AI tools in imaging workflows. It outlines concrete steps facilities can follow to implement, utilize, and continually update AI tools. (Most hospitals are drowning in AI vendors right now, so this is actually useful.)

According to the official ACR press release, the parameter explains how practices can choose AI tools, evaluate them before deployment, watch performance over time, and protect patient privacy. The document details specific operational steps: setting up a governance group, maintaining an inventory of AI tools in use, running local acceptance testing, and monitoring real-world model performance.

SIIM Board Chair Nabile Safdar, MD, emphasized that responsible AI use in healthcare is an ongoing process rather than a single event. It demands a dedicated team consistently applying processes supported by methods and technology. The parameters are based on principles of AI science, interoperability standards, expert involvement, and what Safdar called "methodological precision."

Independent reporting from Radiology Business corroborates the timeline and scope of the changes. The trade publication confirms the Council's approval occurred during the annual meeting and notes the collaborative nature of the effort between ACR and SIIM.

Simultaneously, the ACR Data Science Institute published a landmark article in the Journal of the American College of Radiology detailing the technical framework for Assess-AI. This marks the world's first AI quality registry and data service, designed to help practices monitor and improve AI performance. The registry currently supports multiple imaging AI use cases: intracranial hemorrhage, pulmonary embolism, pneumothorax, large vessel occlusion, bone age, cervical spine fracture, breast density, pneumoperitoneum, tube malposition, pleural effusion, brain mass effect, and obstructive hydrocephalus.

Think about the physical reality here. A radiologist opens a CT scan on their workstation. An AI tool flags a potential pulmonary embolism. The radiologist clicks through the interface, reviews the highlighted areas, and makes a final determination. Assess-AI monitors those AI results and collects contextual information: anonymized patient demographics, exam metadata, and output from radiology reports. Most legacy radiology systems were never built to help sites ensure clinical AI tools perform as expected.

ACR Vice Chair Christoph Wald, MD, PhD, MBA, FACR, explained that Assess-AI provides facilities with interactive analytics showing how AI tools perform across their practices over time. Site data can be compared to aggregated national performance benchmarks from other sites using AI for identical use cases. This innovative data solution can also help AI model developers improve future model versions in collaboration with customers.

Practices implementing AI responsibly can earn the ACR Recognized Center for Healthcare-AI (ARCH-AI) designation. This is the first international AI facility quality assurance program. Both ARCH-AI and the new practice parameter guide sites through selection, monitoring, and continuous quality improvement. Facilities join a learning community of sites supporting each other in this journey.

Tessa Cook, MD, PhD, FSIIM, FACR, chair of the practice parameter writing committee and incoming chair of the ACR Commission on Informatics, called this first-of-its-kind parameter a vital step toward safe, effective, and transparent acceleration of radiology AI adoption. The goal is helping radiologists and allied professionals provide the highest-quality patient care.

ACR's portfolio of AI products includes AI Central, a tool that helps facilities make informed AI solution purchasing decisions. Woojin Kim, MD, chief medical officer at ACR DSI, said the organization looks forward to expanding ACR and DSI facilitated offerings. These provide tangible, real-world approaches to address AI challenges that radiologists increasingly face.

The physical interaction matters. A technologist loads a patient's imaging study. The AI processes it in seconds. The radiologist reviews the output. If the AI drifts—if its performance degrades over time—someone needs to know. Assess-AI monitors that drift. It catches when a model trained on one population starts underperforming on another. That's the difference between a tool that helps and a tool that creates liability.

ACR, SIIM, and allied stakeholders are working with leadership at the U.S. Food and Drug Administration and Congress. They aim to serve a central role in the future of radiology AI. The question is whether regulators will adopt these frameworks as standards or leave them as voluntary guidelines.

Whether hospitals actually implement these parameters remains the real question. The framework exists. The registry is operational. But adoption requires resources, time, and institutional will. Some facilities will embrace it. Others will treat it as another compliance checkbox. The difference between those two approaches will determine whether this framework changes patient care or just adds another layer of bureaucracy.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn