AIR_EC_Capabilities

Overview

The AI Results (AIR) Profile specifies how AI Results encoded as DICOM Structured Reports (SRs).   Depending on the AI algorithms implemented on the AIR Evidence Creator (EC) actor, the EC will create/encode one or more of the different result primitives in its SRs, e.g. qualitative findings, measurements, locations, regions, parametric maps, tracking identifiers, image references.

For the Connectathon, there is a set of no-peer tests to evaluate how the Evidence Creator encodes its AI results; the tests follow the naming pattern AIR_Content_*. Each of these tests align with a different result primitive included in an AI results SR.  We have created separate tests for the different result primitives to make it test execution and evaluation more manageable.  The Evidence Creator will perform Connectathon tests that are applicable to the SRs and primitives it has implemented.

The purpose of this Preparatory test is to have the Evidence Creator describe in narrative form the nature of its AI results implementation.     Reading this description will help the Connectathon monitor have the proper context to evaluate your Evidence Creator application, the AI results you produce, and the result primitives included in your AI SR instances.

Instructions for Evidence Creators

For this test you (the Evidence Creator) will produce a short document describing your implementation in the context of the AI Results Profile specification.  The format of the document is not important.  It may be a PDF, a Word or google doc, or some other narrative format.

Your document shall include the following content:

  1. Your system name in Gazelle Test Management (eg. OTHER_XYZ-Medical)
  2. AI Algorithm Description - this should be a sentence or two describing what your algorithm does (e.g. detect lung nodules)
  3. DICOM IODs implemented (one or more of:  Comprehensive 3D SR Storage IOD, Segmentation Storage IOD, Parametfic Map Storage IOD, Key Object Selection (KOS) Document Storage IOD)
  4. Result primitives encoded in the AI Result SR. (one more of: qualitative findings, measurements, locations, regions, parametric maps, tracking identifiers, image references)
  5. If you encode measurements, indicate whether your measurements reflect a planar region of an image (i.e. use TID 1411, a volume (TID 1410), or are measurements that are not tied to a planar region or volume (TID 1501). (Refer to RAD TF-3: 6.5.3.3 in the AIR TI Supplement for details.)
  6. If you encode regions, indicate whether they are contour-based regions (i.e. use TID 1410 or 1411) or pixel/voxel-based regions (i.e. use the DICOM Segmentation Storage SOP Class) (Refer to RAD TF-3: 6.5.3.5 for details).
  7. Please add any additional information (e.g. screen shots) that would help the reader understand your algorithm, and output.
  8. REPEAT 2 - 7 for each AI Algorithm that produces result(s) on your Evidence Creator
  9. Finally, in order to share it with your test partners, upload your document as a Sample in Gazelle Test Management.  On the 'List of Samples' page, use the dropdowns to find your test system, and on the 'Samples to share' tab, add a new "AIR_EC_Capabilities" sample and upload your document there.   When you save your sample, it will be visible to your test partners.

 

Instructions for AIR 'consumer' systems

    1. Each Evidence Creator should have uploaded a description of their AI algorithms and inputs and outputs into Gazelle Test Management (Gazelle TM under Testing-> Sample exchange.    On the Samples available for rendering tab under the AIR_EC_Capabilities entry,  This page will evolve as your partners add samples, so be patient. 
    2. Retrieve the document uploaded. The purpose is to give you an understanding of the types of AI content that your Image Display, Image Manager or Imaging Doc Cosumer will store/display/process  Refer to these help pages for details on this task.

Evaluation

There is no "pass/fail" for this test.  However, you must complete it because it is a prerequisite for several Connectathon tests.  The Connectathon monitor will be looking for the document you produce here and use it when s/he is evaluating your AI result content.