Guiding the Future: AI in Radiology and the New ESR Surveillance Standards

ESR guidelines issue consensus recommendations that mandate comprehensive post‑market surveillance of AI medical devices to detect performance drift and protect diagnostic accuracy in routine radiology.

The guidance defines continuous monitoring requirements and coordinates reporting duties between providers and deployers to meet MDR and the EU AI Act. It also calls for interoperable PMS platforms and periodic summary reporting to harmonize regulatory expectations across institutions — changes that have immediate operational implications for departmental workflows. Departments therefore must translate these requirements into concrete monitoring and governance actions.

Departments now face expanded oversight beyond traditional manufacturer responsibilities. Deployer‑level activities such as synthetic imaging validation and proactive screening‑overdiagnosis mitigation, previously sporadic, are expected components of an institutional PMS strategy.

Evidence‑first, prioritized post‑market surveillance activities include routine performance monitoring against baseline metrics with uncertainty intervals to flag drift; thresholded alerts tied to defined action plans to shorten time‑to‑corrective action; real‑world clinical validation (PMCF) to confirm ongoing diagnostic accuracy; systematic logging of edge‑case failures to reveal subgroup gaps; explicit synthetic image validation to ensure fidelity of reconstructed or augmented outputs; and targeted strategies to monitor and mitigate screening overdiagnosis.

The rationale is straightforward: monitoring detects trends early, alerts speed response, PMCF updates benefit–risk assessments, failure logs expose latent weaknesses, synthetic validation prevents artefactual interpretation errors, and overdiagnosis surveillance protects patients from excess intervention. Priorities should focus on performance drift and emergent safety signals as primary triggers for corrective action.

Radiologists carry defined oversight responsibilities, including regular case review of algorithm outputs, structured incident reporting, participation in multidisciplinary governance committees, and negotiation of vendor data‑access and PMS terms. These duties are shared with vendors: providers retain formal PMS responsibility while deployers must ensure robust logging, explicit human oversight, and timely feedback loops. And so departments should consider codifying reporting workflows, assigning accountable personnel, and embedding vendor agreements into device‑lifecycle management to operationalize these duties.

Key Takeaways:

• What’s new — Standardized post‑market surveillance with explicit emphasis on routine drift detection and interoperable PMS platforms, creating uniform expectations for monitoring and reporting.
• Who’s affected — Radiology departments, AI vendors, and procurement teams will share operational and legal responsibilities for monitoring, data access, and corrective actions.
• What changes next — Implement scheduled monitoring cycles, centralized logging and validation protocols, and governance checkpoints tied to vendor contracts and institutional risk management.