'''PaLM Conf Minutes 2021-June-15

From IHE Wiki
Jump to navigation Jump to search


Name Name
Venkat Murugan Norman Zerbe
Riki Merrick Luca Desogus
Alessandro Sulis Mikael Wintell
Dan Rutz Takuya Haga
Raj Dash Joachim Schmid
Megumi Kondo Steven Borg
Ralf Herzog Arvydas Laurinavicius
JD Nolen Kevin Schap
Nick Haarselhorst Bruce Beckwith
Mary Kennedy Bjorn Lindequist
Gunter Haroske Craig Sayers
David de Mena Gabor Garanyi
Brian Bialecki Wei-Chung Cheng
David Beckman Geregly Szollosi
Markus Herrmann Uwe Horchner
Graham King David De Mena
Simon Doran David Clunie
Mickey Mandel Veronica Klepeis
Jan Schutrups Mike Riben
  1. Transferring an annotation back and forth
  2. Highlighting the annotation / region of interest to be reviewed
  3. DPIA LAB-80 and Container ID discussion
  4. Selection of schemes/systems and value sets for coding anatomical concepts and pathological findings


  • DPIA profile
    • Change request for inclusion of SAC segment
      • Slides = container
      • Specimen & Container number are two distinct things
      • Specimen could be a higher level identifier (or container) which would be separate from container ID.
      • Specimen is different from the slide
        • Can have more than 2 smears on a slide – example cervix and vaginal canal?
        • Also, can you split a WSI into multiple sections – this is different from region of interest
        • Example: https://lumea.net/solutions/labs
    • As you move into the digital world, might need to come up with a separate ID to relate them back to the original slide they were derived from.
      • Right now the profile cannot differentiate that
      • In practice the identifier of the specimen and the container are the same – at least for blood etc.
      • DICOM metadata supports identifying when slides have samples from the same specimen
      • DICOM translated what scanners are currently doing into the DPIA profile, but there is a need to be able to put this in to support the future
      • Can we add this segment as 0..1?
        • We could, but in LAW it is 1..1, would be good for it to match
        • Should not make comments on the content of the container ID vs specimen ID about matching
      • Raj will draft the CP and we will discuss it on the next call
      • The separate spreadsheet for mapping to DICOM should also be updated
  • DICOM annotation work update
    • Has 2 mechanisms
      • Structured report can reference coordinates on slides and regions of interest – this does not scale well for AI algorithms
      • Supplements 222 is going through approval now; = Binary arrays of coordinates (how to encode the graphical representation of the region of interest) – but still need work on how to describe what these regions represent
    • This is addressing real world scenarios – will need to see how this translates into real world uses – will try at a Hackathon (need LIS and scanner and AI vendors)
      • Also need to figure out how to link descriptions of gross work to the micro results (this is on the backburner for DICOM – we are aware of that gap – this would be great for IHE PaLM to take on for LIS / scanner vendors and vendors supporting gross (not many involved) – DICOM probably has multiple mechanisms that could be used, so getting the specific use cases and workflows described so that we can then decide which it use would be useful
      • Hard to find vendors in the grossing space, as that they don’t have interest in standards development – if we can push this as a patient safety issue, maybe we can get more interest
      • Would a whitepaper be of value to resolve this to show off the ROI of implementing this and then give to vendor.
        • Get help from the academic world?
        • Writing up specific use cases for this and show where the issues are now vs where they could be
        • Could we take/find a proprietary grossing system and LIS and PACs to show if it is possible (academic project?)
    • Next Step:
      • Make this topic of a next IHE PaLM call
      • Consider writing the white paper to show the gaps to get vendors more involved
        • Tumor board review
        • Review tumor margins
        • AI Annotation objects into report (explicit workflow and show where we need the interoperability between the AI and the LIS – a simple use case and how the data travels after that)
  • DICOM Hackathon on Friday 6/18
    • We create the standard, but we don’t often test it first (so test it before we write it)
    • For the connectathons there is so much work before folks can try out
    • Hackathons provide a less formal venue to test these mechanisms
    • Could we expand that into the IHE PaLM profiles?
    • Or at least build a reference implementation to help test the standards?
    • https://github.com/DICOMWG26AnnotationAHG/hackathon
    • Would be good, if we can get larger community involvement:
      • Academia
      • AI vendors (open-source community that is working on QPath – openslide is working on DICOM support, they currently have a proprietary annotation mechanism)
      • We all can help get the word out
      • Create a demonstration implementation
        • Example: get info from David Clunie!
    • This was short notice, so plan another event to give folks more of a heads up
      • Create a wish list of use cases to be tested
      • Goals for the hackathon
    • Phillips is working on adding annotations into their products
    • Use case:
      • Breast biomarker (nuclear stains) use case and lung cancer and PDL1 (algorithm development is currently ongoing under FDA review) maybe just select region of interest, with additional annotations (is there tumor in the lymph nodes and how much to get very specific) = Ask Raj to WRITE THIS UP
      • From the use cases we can then describe the requirements for the vendor systems to close the gaps
      • Is this to communicate that the annotation exist, or to communicate the
      • On the slide can see the region of interest, then copy and paste the annotations from the AI vendors viewers into the LIS
      • Telepathology consult would need
      • This is a subset of observations that need to be grouped – need to define the entity and collect the reporting about it = similar to Genomic discussions at HL7 OO.
      • Need to find a good way to integrate the image reports and the evidence created from the images (which are observations like regular lab results)
      • Need to have an identifier for the region to tie these together
      • Need to be able to manage the workflow (track the events, the status of the process, tag images for teaching/consulting, reference in interpretations as supporting evidence (for example when a tumor has different expressions of tumor markers, making it hard to predict the treatment response of the patient)
      • Need to define the boundaries of the data creation, annotation, exchange need, error correction
      • Need to categorize the value sets for the region of interest, disease identified etc
      • DICOM needs to define the basic use case for the evidence creator actor and the display of the annotation
    • Next Steps:
      • Will need to develop IHE sponsored testing tools for the DPIA profile (need to work on the Gazelle profile and we may need to get some testing tools (proxy and actor creator)
      • Riki to reach out to the IHE testing and tooling committee – will ask Anne Gaelle for how to move this forward
      • Write test scripts for the Gazelle testing
      • Make this agenda item for each of the groups and how to progress this work jointly
      • Raj to write the use case and then derive the workflow
      • Markus will write out the requirements for the hackathon
  • IHC Specimen description correction
    • Preparation, fixation, staining – how to codify these elements; some of these stains have more than one component,
    • What elements do we need to capture, how do we communicate that detail to the systems that do the staining and what needs to be sent to the scanner (if anything)
    • For PDL1 - what is the antibody, what is the clone, in what step is the Antibody used – we don’t have a good identifier (catalog numbers are the only thing we have now) how do we track the procedure (encoding the SOP might be overkill)
    • The HL7 specimen DAM uses a reference to the protocol / SOP and allow for identification, if there was a variance (what and why)
    • Reference to the device that was used for processing
    • LIS sends message with single local code, that is pre-programmed (SOP) to the stainer – no baggage on the message, that changes the slide from unstained to stain X; that is then sent to the scanner
    • Problem with referencing the SOP that is local (and also time-bound)
    • For research use case may want to be explicit what has been done
    • IHC Specimen Description Correction Proposal
    • When we describe a biomarker or annotation of a particular biomarker finding, we will need to specify which antibody is being used and measured.
    • Current Specimen Preparation Model - Fixation, Embedding, Staining
    • If you try to extend it beyond traditional stains things get complicated
    • Try to determine the information you might want to capture when doing IHC and codeset that should be used to capture it.
    • What does the LIS know inside itself about the use of Ki-67 and how does it communicate to the devices applying the staining device, and how does it send that data to the scanner creating the image?
    • What is the right level of abstraction to record into the DICOM file?
    • Rather then put the entire SOP into the Specimen DAM, it's referenced
    • Good as long as you remain in the institution
    • Trying to expand beyond a protocol code, which is likely local
  • Update on IHE Digital Pathology Profiles – next planned:
    • Ordering (Francois)
    • Evidence creator (Raj and Jim Harrison – looking for AI vendors)
  • It was agreed that DICOM WG 26 and IHE PaLM should continue to meet jointly – possibly every three months.