Apply transformation to the points given in Markups fiducial file

[echonax07]

Hi
I am new to medical imaging field.
I have a landmark that specifies different anatomy of the brain in fcsv file(which the 3Dslicer generates).
These landmarks are ground truth for my model.
So In the data augmentation part, I want to apply the transformation to both training images as well as the landmarks,
Any idea how I could do that?

[fepegar]

Hi, @echonax07. We would need some major features added to the library, but I think it's a good idea. It's related to #537, so we can maybe discuss this there. What transforms are you interested in?

[romainVala]

I did not realize at first read, but yes that is true if you consider surface as a point cloud, then it is the same datatype as for landmarks.
Since landmarks are just a list of 3D coordinate, (rigth ?)

there are also a surface format, with triangle (meshes) that is then a little bit different. but may be not for the transform point of vue, because, even with mesh, you end up with a list of 3D coordinate ...

We could then consider the 2 format, mesh and point cloud ... or just only one new torchio datatype is needed

[fepegar]

[@romainVala I tried to comment on your issue earlier today but my comment got somehow lost]

I suppose we could add a Surface and Points classes. It would be nice to make the internal data compatible with ITK transforms, so maybe VTK could be used for this (I think it's not very hard to convert from ITK to VTK nowadays). ITK has classes for mesh and point clouds, but not SimpleITK, unfortunately.

The points class should be able to read ROIs and fiducial files from old (acsv, fcsv) and recent (acsv, .mrk.json) versions of Slicer, and simpler things like .txt or .npy, I suppose.

[fepegar]

Here's some handling of point clouds using Python ITK: https://examples.itk.org/src/registration/metricsv4/registertwopointsets/documentation