Status: Currently the code base in MITK often assumes and checks that a segmentation has to have completly the same geometry/grid than the respective image.
- E.g. a lot of CL apps (like CLGlobalImageFeature, Fitting, PixeldDump...)
- The segmentation views and utils
- The statistic classes /modules / views
- The registration views
to name a view, but there are certainly many more.
In practice it would be ofte enough to just ensure that the segmetation has the very same grid or a subregion out of it. Thus you could also use cropped segmentation that do only cover a smaller part of the grid.
Currently we allready offer such possibilities (crop of ML segmentations in context menu) or saving of DCM Seg (see T27416) with skipping empty slices and they lead to unsupported segmentations.
Introducing such an feature would be a major rework overhaul of the code base. It should be done in conjunction with other discussed planned and reworks, like
- T24303 (as we now would blow up the segmentation any way if you want to alter it again regarding a given reference image)
The core things needed for this task are quite easy:
- A helper/predicate that can be sued to easily check if a geometry is equal to a ref geometry or a valid sub part of the ref geometry grid.
- A helper that allows to extend a segmentation (that is a valid subpart of a grid) to the grid/geometry of a reference image. E.g. to use in code parts that cannot properly handle subgrid segmentations.
The tricky part will be to rework all the code/business logic to properly handle such subpart segmentations.