Handling deformable transforms in Slicer meeting minutes

Operations

• Harden transform

  Harden transform* is an existing capability that collapses linear transforms in a transform hierarchty. Hardening is available as a context menu on a (data) node in the Data module. When hardened, the data object is moved in the Data module to be outside all transformations with the effect of the transform hierarchy applied. *Harden transform* is only available for transform hierarchies involving linear transforms.

• Clone and harden

  Clone and harden* differs from the current *Harden transform* capability in two ways:

  1. Instead of the data being moved outside of all transforms, a "copy" of the data will be moved outside of all transforms.
  2. All transform types, linear and deformable, will be part of the hardening.

Proposed data type support

• Scalar volumes
• Vector volumes
  • RGB
  • Vector
  • Displacement field
• Tensor volumes
• Models
• Fiducials and annotations
• Multivolume

Questions

• Should we maintain a link between the original and hardened data?
• Should only the hardened version be available for visualization?
  • Probably yes, but display a warning on hardening
• Can you clone and harden part of the transform hierarchy?

  Harden just the non-linear portions

• Which tools do we use for each type of data to harden/warp the data?
• When hardening applied, the user would be presented with a list of modules that can harden each type of data?
• What happens to
  • pixel lattice - stays the "same"
    • create an axis-aligned bounding box in the world space, the number of pixels along each dimension will change
  • spacing - affected by the transform
  • orientation - affected by the transform

Design

• Level of performing the hardening
  • Application level: module can register itself as the capable of hardening certain types of data.
    • For now the modules containing the algorithms (BRAINSResample, Resample Scalar/Vector/DWI Volume) will be known by the class performing the hardening
  • Logic level
• Need to know the inverse of each transform
  • to transform the models and fiducials
  • to figure out how to define an appropriate lattice
    • need to back transform all the edge/face voxels to find the bounding box in the world space
• Three options presented to the user
  • Resample in source lattice (still cloned)
  • Resample into lattice that bounds the warping of the edge/face voxels
  • Resample into a target volume
• Where?
  • Tie into the drag and drop of the Data module and prompt the user to harden the transform.
  • Transforms module: display tree, and add a 'Harden' button to the panels
  • Subject Hierarchy right-click -> apply transform to branch
• Look at the logic of handling transforms in DICOM

Subject hierarchy

• Transforms resulting from registration point back to the fixed and moving images. This information is stored in the *attributes* of the hierarchy node.
• Should we use the DICOM frame of reference concepts (frame of reference UID).
• Transform map between two frames of references.
• Currently, Slicer combines two concepts:
  • Frame of reference
    • Everything "under" a transform is in the same frame of reference
  • Transform that maps between two frames of references
    • The two frames of reference are implicitly defined by the transform hierarchy