Rad Tech Minutes 2019-11-11 to 2019-11-14

Participants

Participants: - Andrei - Chris Lindop - Salt - Steve N. - Kevin - Jonathan - Kinson - Kevin Schap (CAP, IHE PaLM) - Antje - Wim - Elliot Silver - Sridhar B (Nuance) - Neil - Brad - David Kwan - Hamid - Charles Parisot

TF Maintenance decisions

Maintenance:

• CP approved for ballot
  • CP-RAD-258
  • CP-RAD-356
  • CP-RAD-378
  • CP-RAD-379
  • CP-RAD-382

• CP cancelled
  • CP-RAD-239
  • CP-RAD-259

Notes on break up of RAD Tech Framework

See https://wiki.ihe.net/index.php/RAD_TF_Maintenance_2019-2020#Proposed_breakdown_-_initial_proposal_from_Lynn

TI Supplements final text / deprecation decisions

• DBT extension

• Technical Committee consensus has been achieved to promote the DBT Extension too Final Text, see DBT Extension Evaluation

• IID
• It is proposed that the profile is not advanced to the Final Text until CPs 349 and 364 against it are finalized.
• MAWF
• Steve to draft Final Text evaluation for SWF.b FT Evaluation

Kickoff Evaluations

AI Workflow

• • Describe gaps in Use Case coverage
    • Currently reconstituting the Simple Case (scan through result distribution)
    • Thinking through variants on the simple case
    • Some open discussion around which places proxy, e.g. Orchestrator proxying for the Model to the local data systems
  • Review the "uncertainty points" in the evaluation. Is there a resolution plan for each?
    • Mostly they got copied into Open Issues and that OI list has been reviewed and resolutions planned.
    • (Didn't actually review the UP list from the proposal - think we got them all)
  • Do the effort points in the evaluation still seem right?
    • Mostly seem right. The UPS-RS issue was scoped for no discussion, but we have spent some time.
    • (Learning point: make sure resolutions during evaluation are carefully captured and clearly expressed)
    • We've been trying hard to hold our scope management which has helped.
  • Describe unresolved technical issues/tasks
    • Need to fully put the UPS-RS issue to bed
      • Chose to use existing transactions on an existing standard (that is not yet widely implemented). The proposal could also have proposed to develop new transactions on a soon to be existing standard (FHIR Task), or develop new transactions on a standard we would have to standardize (). The latter would have refocussed this cycle work onto that transaction work instead of the workflow level we've been looking at.
    • Kevin will help document the UPS-RS equivalents for the fields highlighted by Neil. That information should be covered in the Workitem Concept section or somewhere more normative if needed.
    • Procedure update has some questions around how AI-driven worklist reprioritization should work - who decides what, what need to be communicated, how is that encoded.
    • A number of sections that need technical content, but it seems like we have an understanding of what goes there.
  • Describe potential practical issues
    • Some of todays AI Models are used to being handed PNG files. It will be a step up to do what we are describing.
    • Haven't really described how our framework applies to the current pattern of AI platform products. (May be covered in Actor Descriptions)
    • AI Models being in the cloud vs locally deployed might involve some practicalities we dont have experience with yet, e.g. security and need for configuration/proxying
    • Questions were raised about regulatory requirements/monitoring, but we are putting those a bit out of scope for now.
  • Review the open issue list. Does it feel complete
    • Pretty much? Maybe think about questions we want to ask/highlight for the Public Commenter community
  • Which open issues feel most risky
    • Workflow is different from the data-push model
  • How is the work fitting in the allocated bandwidth? (Time to spare? Just right? Things were left undone?)
    • Mostly OK.
  • How does the scope feel? (Room to expand? Just right? Pretty ambitious?)
    • Keeps nudging to expand; we keep reining it back in...
  • If you had to reduce scope, what would you drop?
    • Can't drop anything. If we run out of time, simply don't publish this cycle and request again next cycle.
  • Have the promised resources manifested?
    • Mostly - Brad, Neil, Sridhar, Dave Kwan
  • What tasks would benefit from additional expertise? (e.g. each actor, user)
    • AI Model developer, especially standalone. Users (radiologists) who will use multiple models. Users who will configure the logic on the Requesters (IIT? PACS admins? Integrator? Jon Shoemaker?)
  • What vendors are engaged for each actor? Record how many.
    • Requester -
    • Orchestrator - ???
    • AI Model - ???
    • Reporting Worklist - Visage, Nuance, GE, Change
  • How many tcons would you like between now and the PC Prep Meeting?
    • 1 mid-December, 1 mid-January
    • (Joint with WG23 in Dec)

It will be the responsibility of the Profile Editor to lead resolution of these issues before the Public Comment preparation meeting.

AI Results

• • Describe gaps in Use Case coverage
    • None that I can think of. Narrow scope.
  • Review the "uncertainty points" in the evaluation. Is there a resolution plan for each?
    • Yes. Resolved with relevant Concept sections etc.
  • Do the effort points in the evaluation still seem right?
    • Basically yes. Actually used a bit less this meeting. Probably will use full slot next meeting.
  • Describe unresolved technical issues/tasks
    • None currently visible. Mostly exploring practical product design issues to see if anything needs additional tech.
    • Creating the example encodings for each primitive will be a bit of work. (consider both inline explanation, and actual sample objects in the Implementation Materials folder.
  • Describe potential practical issues
    • Navigation and presentation will be challenging for Displays.
  • Review the open issue list. Does it feel complete
    • Seems complete
  • Which open issues feel most risky
    • Not an open issue, but it will be a step for the AI Models to implement the specified DICOM SOPs. Might want to encourage platform/proxies to help them with this. (and make the case you get to read these too to get ground truth)
    • Another adoption risk is "conflict" between multiple standards activities (e.g. FHIR, ACR, local groups, etc)
  • How is the work fitting in the allocated bandwidth? (Time to spare? Just right? Things were left undone?)
    • Well. A bit of time to spare this meeting.
  • How does the scope feel? (Room to expand? Just right? Pretty ambitious?)
    • Good. Constraining scope creep.
  • If you had to reduce scope, what would you drop?
    • STOW AI Sketch, Consumer (not much of a savings)
  • Have the promised resources manifested?
    • Julian, Eliot, Dave Kwan, Sridhar, Jonathan, Andrei, Kinson, etc. (CAD, Displays, etc)
  • What tasks would benefit from additional expertise?
    • Radiologists to vet the navigation and display requirements.
  • What vendors are engaged for each actor? Record how many.
    • Evidence Creator - Nuance
    • Image Manager - Vital, Visage, Change, GE, Siemens
    • Image Display - Vital, Visage, Change, GE,
    • Imaging Doc Consumer - Nuance,
  • How many tcons would you like between now and the PC Prep Meeting?
    • Yes, 1 mid-Dec, 1 mid-Jan. Will likely cancel one or both but hold the spot.