ACR Unveils First AI Practice Parameter and Assess‑AI Registry

*TL;DR: The American College of Radiology (ACR) approved its first AI practice parameter and introduced the Assess‑AI registry, giving radiology sites a concrete framework and national benchmarks for monitoring AI tools.

Context The ACR Council adopted the ACR‑SIIM practice parameter at its 2026 meeting in Washington, DC. The parameter, authored by a multidisciplinary committee, outlines how imaging facilities should select, test, deploy, and continuously evaluate AI algorithms. Simultaneously, the ACR Data Science Institute published the technical design of Assess‑AI, the first national quality registry for imaging AI, in the Journal of the American College of Radiology.

Key Facts - The practice parameter provides a step‑by‑step guide for physicians, technologists, physicists, IT staff, data scientists, and administrators to implement AI safely and transparently. It mandates an AI governance group, an inventory of tools, local acceptance testing, ongoing performance monitoring, and compliance with HIPAA privacy rules. - Facilities that follow the protocol can earn the ACR Recognized Center for Healthcare‑AI (ARCH‑AI) designation, joining a global learning community. - Assess‑AI integrates de‑identified exam data, patient demographics, and radiology report‑derived surrogate labels via ACR Connect. It uses large‑language‑model prompting to extract these labels automatically. - Users receive interactive analytics that track AI output over time and compare site performance against national benchmarks for identical use cases such as intracranial hemorrhage detection or breast density assessment. - The registry supports a broad menu of AI applications, including pulmonary embolism, bone age, cervical spine fracture, and tube malposition, among others. - According to Tessa Cook, MD, PhD, the parameter “gives imaging facilities a step‑by‑step guide for implementing, using, and continuously improving AI in clinical care.” Christoph Wald, MD, PhD, MBA, adds that Assess‑AI “allows sites to compare their data with national benchmarks for identical AI use cases.”

What It Means The combined rollout creates the first standardized pathway for radiology departments to adopt AI with measurable quality control. By linking real‑world performance data to a national benchmark, sites can identify model drift—declines in accuracy over time—and intervene before patient care is affected. The closed‑loop workflow, which includes local case review through ACR Forensics, turns performance gaps into actionable improvements.

For AI developers, the registry offers a feedback loop to refine algorithms based on aggregated clinical outcomes. For patients, the initiative promises greater consistency in AI‑assisted diagnoses across hospitals.

Looking Ahead Watch for the first ARCH‑AI certifications and for early studies that correlate Assess‑AI benchmark participation with diagnostic accuracy improvements.