EDRN DICOM Image Headers Subgroup Call Minutes

(This meeting is not online.)

Attendees:

  • Radka Stoyanova, University of Miami
  • Indu Kohaar, NCI
  • Guillermo Marquez, NCI
  • Jackie Dahlgren, DMCC
  • Savannah Partridge, University of Washington SOM
  • Biswas Debosmita, University of Washington
  • Suleeporn Yui Sujichantarara, University of Washington SOM
  • Yency Forero Martinez, Vanderbilt
  • Yoga Balagurunathan, Moffitt
  • Will Hsu, UCLA
  • Dan Crichton, JPL
  • Ashish Mahabal, Caltech
  • Heather Kincaid, JPL

Summary from Slides

EDRN DICOM Header Tag Standards Working Group Update

Our goal is to define a set of minimal required DICOM header fields that make cancer biomarker imaging data reusable and AI-ready, while minimizing burden on participating sites.

This will support future standardization and help guide SOP development for EDRN studies.

Current Plan:

  • Start with a Core set of of minimal DICOM header fields
  • Add modality-specific extensions in next version or as needed

Initial Use Cases:

  • Prostate MRI Reference Set
  • Lung Team Project 2 Reference Set

Working Drafts for Review

Developed using the TCIA process, refined with ChatGPT:

Questions and Issues with Current DICOM Headers

  • Inconsistent header values across sites
  • No unique ID to link images from multiple timepoints
  • Risk of over-anonymization—has important metadata been lost?
  • Do we need additional fields to support AI-readiness?

Prostate MRI Imaging Project

A workflow developed by JPL updates specific DICOM tags using DMCC-supplied values. The transformed dataset is stored as a new version to preserve provenance.Determining what’s needed to ensure pMRI images are reusable and AI-Ready

Current DICOM Header Tags identified to be updated via workflow:

DICOM Tag Tag Name Source JPL Workflow to Update Header
(0010,0020) Patient ID DMCC API Blinded ParticipantID
(0020,0010) Study ID DMCC API EventID
(0012,0050) Clinical Trial Time Point ID DMCC API VisitCode
(0008,0080) Institution Name Overwrite to “Anonymous”

MRI Specific

  • Python Script (Radka’s group) - Included in LabCAS Workflow
  • Run high-B value algorithm

Key Discussion Points

De-identification Tools and Site Support

  • The group discussed the importance of making the de-identification process more intuitive and less reliant on manual work by sites.
  • Many sites currently use PACS systems with varying default configurations, which raises concerns about PHI removal and consistent retention of necessary tag metadata.
  • Several tools were discussed as potential options for de-identification:
    • TRIAD (ACR): A push-button tool that supports multi-modality and multi-vendor formats. Savannah will follow up with ACR to confirm access and licensing details.
    • RSNA CTP: A free, open-source tool run at the site, though possibly too technical for some sites.
    • Lung Hive: A web-based, third-party tool used by Vanderbilt that requires a paid license.
  • If these tools aren’t feasible for EDRN-wide use, the group discussed developing “how-to” guides for common PACS systems.
  • The group suggested creating a list of approved or recommended de-identification tools.

Standardization and Automation

The goal of standardization is to ensure that images are de-identified, FAIR and AI-ready.

  • If possible, sites should not be expected to perform detailed header modifications manually.
  • Over-anonymization by sites could hinder metadata updates, though no such issues have been observed so far.
  • In some cases, PDFs containing identifiers are included in image directories, requiring additional manual QA.
  • The group stressed the importance of avoiding full reliance on human review and supported automated, standardized tools where feasible.

In the future, sites may retrieve the BlindedParticipantID from the DMCC (e.g., via VSIMS) and organize de-identified images in folders named with this ID. JPL can then use this to match and update the DICOM headers using the DMCC API, minimizing site burden while ensuring metadata consistency.

De-Identification Process

  • The updated process should emphasize which metadata must be removed before upload.
  • Instructions should include how sites obtain the Blinded Site ID from the DMCC.
  • Will suggested testing the updated process with the UCLA LTP2 site to ensure clarity and usability.

Next Steps

Short-term (for existing datasets in LabCAS: pMRI and LTP2):

  • Review and finalize the required DICOM header tags and identify which tags can be updated via JPL’s workflow.

Long-term (exsiting and future EDRN studies like TRACER and CVC Hepatocellular Carcinoma):

  • Subgroup members to review and comment on the Draft EDRN DICOM De-identification Process.
  • Continue evaluating options for scalable, site-friendly de-identification workflows:
    • de-identification tools (e.g., TRIAD, RSNA CTP, Lung Hive)
    • “How-to” guides for common PACS configurations
    • A JPL-developed automated workflow using BlindedParticipantID folders, API metadata lookups from the DMCC, and header validation

Action Items:

  • Subgroup members: review and comment on the Draft EDRN DICOM Header Tags for Review. This will also define what should be removed and the minimal required DICOM Tags.
    • Add comments in the “Notes” column.
    • Enter your name in a reviewer column and use the dropdown to select tag action
  • Heather: Replace the June EDRN Data Sharing call with the next subgroup call
  • Subgroup members: review and comment on Draft EDRN DICOM De-identification Process - two current EDRN studies ready to submit images:TRACER and the CVC Hepatocellular Carcinoma
  • Savannah: Contact ACR about TRIAD licensing and feasibility

Next Meeting:

June 16 at 10am PT / 1pm ET

(Replacing the regular June EDRN Data Sharing Call)

When

Los Angeles
June 2, 2025, 10 a.m.
Denver
June 2, 2025, 11 a.m.
Chicago
June 2, 2025, noon
New York
June 2, 2025, 1 p.m.
UTC
June 2, 2025, 5 p.m.