2012 Summer Project Week:AIM for QIN

• 

  Projects List

• 

  Diagram of various data types and connections that need to be organized

• 

  Diagram from Daniel Rubin

Key Investigators

• BWH: Andrey Fedorov, Nicole Aucoin, Paul Mercea, Fiona Fennessy
• Northwestern U.: Pat Mongkolwat
• Stanford: Daniel Rubin
• U. Iowa: Reinhard Beichel, Markus Van Tol
• MGH: Jayashree Kalpathy-Cramer
• NCI: Justin Kirby

Objective

Participants of the Quantitative Imaging Network (QIN) are in need of a standard-based or consensus-based format for consistent storage and organization of various types of data related to quantitative imaging of clinical patients undergoing cancer treatment. Typical dataset includes a subset of the following:

• several time-points with the multi-modal imaging data at each time-point
• segmentations of the lesions and/or reference and/or relevant anatomical structures
• measurements/calculations derived from the images and segmentations
• clinical and demographical information about the patient
• information related to treatment
• pathology correlation data

Ideally, we would like to be able to consistently organize these various types of information into a self-sufficient document/data structure to facilitate exchange of the information for subsequent analysis, presentation and archival. The possibilities currently considered include DICOM-based formats (DICOM Structured Reports), Annotation Image Markup (AIM) format, and Slicer MRML scene.

The goal of this project is to focus on the specific project of U. Iowa in longitudinal PET/CT imaging, and attempt to utilize the capabilities of AIM to organize the data provided by U. of Iowa collaborators. Specifically, the dataset we will use for experimentation includes:

• n PET/CT datasets acquired over the course of treatment
• binary segmentations of lesions, represented as 3d segmentations (lesions may disappear over the course of treatment)
• binary segmentations of the reference structures

BWH and MGH teams are also participants of the QIN, and will provide input related to how the use-cases at their sites are different from U. Iowa use case. BWH will also provide expertise in 3D Slicer in discussing its capabilities in visualizing and organizing the data for the Iowa QIN use case.

Big questions we would like to answer in the end:

• what are the capabilities of AIM with respect to addressing the needs of specific QIN use cases?
• how these capabilities are supported by the tools from the user perspective?

Approach, Plan

• We will have discussions and hands-on demonstrations of AIM capabilities to understand its applicability to QIN use cases.
• Tuesday 6/19 11am to 1pm Eastern time we will have a conference call that will include Daniel Rubin and Reinhard Beichel (joining remotely); project week participants interested to join will meet at room 32-262 at MIT (on the 2nd floor of the Stata Center)
• We will summarize our findings at the end of the project week

Progress

• We had a TCON that included participants from BWH, MGH, Northwestern, Stanford and NCI.
• discussed the specific mechanisms available in AIM for storing multi-timepoint data with multiple lesions at each time-point:
  • separate AIM files should be used for each individual segmentation
  • each annotation has to be associated with a single finding on an image or a set of images in the same series
  • associations across the timepoints that correspond to a given patient can be recovered by means of querying the "AIM database" for a certain patient
  • direct associations can be encoded using AnnotationOfAnnotations type; however, it is not supported by ClearCanvas (the only AIM workstation that is available to the QIN users). There is no plan to add this functionality in the foreseeable future
• discussed the plans for supporting cross-platform exchange of 3D segmentations
  • support for DICOM segmentation objects is very limited and not available on the workstations readily available
  • there is no plan for supporting DICOM seg objects in ClearCanvas
  • BWH is in the process of implementing DICOM seg support in 3D Slicer, but testing is limited to cross-check with syngo.via
  • an alternative has been suggested to use DICOM secondary capture to store segmentation label in NRRD format, which is widely supported by libraries and viewers
  • BWH will work with Daniel Rubin to exchange NRRD date wrapped in secondary capture and point to the code that does this packaging

Tuesday TCON

Attended: Pat Mongkolwat, Daniel Rubin, Vlad Kleper, Nicole Aucoin, Andrey Fedorov, Reinhard Beichel, Marcus Van Tol, Jayashree Kalpathy-Cramer, Justin Kirby, Pallavi Tiwari

• Pat Mongkolwat:
  • AIM stores informaton related to the image, so patient information will not be represented
  • Separate nodes require separate AIM objects
  • Elements stored:
    • Annotations
    • Measurements derived from annotations
    • Markup types (point, multipoint, polygon, circle) + segmentation object
  • Annotations of annotations can be used to organize multiple timepoints and multiple modalities
• Support of Annotations of annotations
  • in workstations is not currently available, but if this is justified by QIN use cases, Daniel and Pat can work on this
  • Pat: this is not in going to be available in the foreseeable future
• Sharing of 3d segmentations
  • support of DICOM SEG in ClearCanvas is not going to be available anytime soon
  • since DICOM SEG is supported by very limited set of tools, we may consider using NRRD as a format referenced from AIM

Action items

• Daniel:
  • will talk with Reinhard to learn more about the use case and learn if AIM is suitable
  • Andrey will send SEG object example produced and a pointer to the code

• Reinhard:
  • would like to know what infrastructure is available in NIH for sharing AIM, and also for storing patient information (age, tests, clinical info)

• Andrey:
  • will send Slicer data bundle and a pointer to the code that creates/unpackages the scene and referenced data
  • will look into wrapping just the NRRD file into a DICOM SC for the purposes of sharing segmentations

Delivery Mechanism

The summary of our findings will be available to the NA-MIC and QIN community on this wiki page.

References

• Quantitative Imaging Network (QIN) overview: https://wiki.nci.nih.gov/display/CIP/QIN
• BWH AIM/Slicer integration project: Projects:QIN:3D_Slicer_Annotation_Image_Markup
• QIN Motivating use cases: Projects:QIN:3D_Slicer_Annotation_Image_Markup:Use_cases