A supplemental PACS could help validate AI imaging workflows, according to a presentation given June 12 at the Society for Imaging Informatics in Medicine (SIIM) annual meeting.
In his talk, Stephen Clark from Cincinnati Children’s Hospital in Ohio reported success from his team’s experience with developing its supplemental PACS to make way for AI testing and validation.
“There are some [electronic medical record] vendors that have supplemental environments, so we thought, why not do this with PACS?” Clark said.
Validating imaging workflows that are aided by AI can be a difficult process, since clinical operations may be disrupted. Clark noted the need for a “reliable and realistic” testing platform for these AI-enabled workflows without interfering with patient care. He added that typical test systems contain studies that don’t have metadata to properly and fully test workflows.
Clark and colleagues developed their supplemental PACS environment that replicates key parts of production and data streams while staying separate from clinical systems. The system can receive copies of production HL7 interfaces.
“This is not a replacement for PACS production or task,” Clark said. “Think of it as an extension designed for testing, validation, and experimentation.”
The team used controlled DICOM forwarding rules built within the DICOM router to replicate and route data streams and production imaging into the supplemental PACS. This way, the supplemental environment could receive real-time imaging volumes, metadata, and variations specific to imaging modalities.
Clark reported that the system has been successful in enabling safe and realistic validation. This includes the following benefits: isolating risk from production systems and data, mirroring real-world imaging and HL7 workflows, allowing for safe testing of edge cases and failures, providing repeatable one-to-one order-to-result validation, and supporting regulatory and governance requirements.
Clark also confirmed that outputs from the system are correctly produced, stored, and interpreted, and ensure integrations for application programming interfaces (APIs) properly trigger downstream workflows. The supplemental PACS also supports validation of changes through PACS workflows, including protocol and user interface behavior testing, Clark added.
“We also can task integration with third-party systems and evaluate the actual user experience before anything reaches your radiologists,” he said.
However, Clark noted some challenges the team faced in implementing the system. These included storage likely to be depleted quickly and network congestion’s impact on overall system performance.
The team configured watermarking for properly managed storage and configured the DICOM router to send data at a lower priority or during non-peak hours.
Clark said the supplemental PACS helped the hospital team push its in-house AI system through the hospital’s internal AI regulatory committee.
“Ultimately, it bridges the gap between development and clinical employments,” he added.
Check out AuntMinnie’s full coverage of SIIM 2026 here.
