Skip to content

Send studies to an AI vendor, de-identified

Goal: chest CTs go to an outside AI service, de-identified. Its results come back to the router, which gives them the patient’s identity and original UIDs back and sends them to the PACS. Every study is tracked until its results arrive.

Configuration › De-identification › Add profile: name AI vendor. The defaults suit most AI services: identities removed, pseudonyms for patient ID and name, new UIDs, dates shifted, patient characteristics (age, sex, size, weight) kept. If the images can carry burned-in text (ultrasound, secondary captures), add mask rules or keep the default of refusing them. See De-identification.

Add a destination for the service: DICOM (host, port, AE title), or DICOMweb (the service URL and its sign-in). Under De-identification, choose the profile AI vendor.

A source for the chest CTs: the CT scanners’ AE titles and network, with a header condition such as StudyDescription Contains CHEST. A route from it to the PACS and the AI service, with Hold until the study is complete ticked, so the service gets whole studies. Put it above the general CT route.

The service sends results back to the router. Add a source that matches it: its AE title and address, or, if it uploads over DICOMweb, a DICOMweb sender (tick Accept DICOMweb uploads from these senders and give the service its token). Add a route from this source to the PACS.

Configuration › AI workflows › Add AI workflow: the AI service (the destination), Its results arrive as (the results source), Expected within (minutes) (how long results may take; then an alert), and Complete after (minutes).

  • Send a chest CT. History › AI jobs shows it Sending, then Waiting for results.
  • When the result arrives: Results coming in, then Complete. In the PACS the result sits in the original study, with the real patient name and ID.
  • Nothing with the real identity ever reached the AI service: open a delivery to it in History to see what it was sent.

See AI workflows.