Add reg_f3d2 support

[ALEXJAZZ008008]

reg_f3d2 works, however currently it flips the output by 180 degrees. If an affine transformation is given as input it does not flip, weirdly.

Insight on the flipping would be appreciated, I'm not certain if the issue is in SIRF or NiftyReg.

[ALEXJAZZ008008]

FYI, addresses #944 for completeness.

[ALEXJAZZ008008]

Furthermore to what I mention above, I should mention that I've tested NiftyReg on the command line; I don't see the flip with -vel (but I do with -sym, as expected from KCL-BMEIS/niftyreg#71). In NiftyReg, reg_f3d2 inherits from reg_f3d_sym so it may be assumed that the issue comes from reg_f3d_sym, however people using reg_f3d2 not through SIRF don't see this issue (and I mentioned I tested it on the command line and couldn't reproduce).

I can produce the issue with:

import sirf.Reg as reg


nifty_f3d_sym = reg.NiftyF3dSym()

nifty_f3d_sym.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
nifty_f3d_sym.set_floating_image(reg.NiftiImageData3D("./flo.nii"))

nifty_f3d_sym.process()

resampler = reg.NiftyResample()

nifty_f3d_sym.get_deformation_field_forward().write("./dvf.nii")

resampler.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
resampler.set_floating_image(reg.NiftiImageData3D("./flo.nii"))
resampler.add_transformation(nifty_f3d_sym.get_deformation_field_forward())

resampler.set_interpolation_type_to_cubic_spline()

resampler.forward(reg.NiftiImageData3D("./flo.nii")).write("./output.nii")

I do not see the issue with:

import sirf.Reg as reg


nifty_f3d_sym = reg.NiftyF3dSym()

nifty_f3d_sym.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
nifty_f3d_sym.set_floating_image(reg.NiftiImageData3D("./flo.nii"))

nifty_aladin_sym = reg.NiftyAladinSym()

nifty_aladin_sym.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
nifty_aladin_sym.set_floating_image(reg.NiftiImageData3D("./flo.nii"))

nifty_aladin_sym.process()

nifty_f3d_sym.set_initial_affine_transformation(nifty_aladin_sym.get_transformation_matrix_forward())

nifty_f3d_sym.process()

resampler = reg.NiftyResample()

nifty_f3d_sym.get_deformation_field_forward().write("./dvf.nii")

resampler.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
resampler.set_floating_image(reg.NiftiImageData3D("./flo.nii"))
resampler.add_transformation(nifty_f3d_sym.get_deformation_field_forward())

resampler.set_interpolation_type_to_cubic_spline()

resampler.forward(reg.NiftiImageData3D("./flo.nii")).write("./output.nii")

or:

import sirf.Reg as reg


nifty_f3d_sym = reg.NiftyF3dSym()

nifty_f3d_sym.set_reference_image(reg.NiftiImageData3D("./ref.nii"))
nifty_f3d_sym.set_floating_image(reg.NiftiImageData3D("./flo.nii"))

nifty_f3d_sym.get_deformation_field_forward().write("./dvf.nii")
nifty_f3d_sym.get_output().write("./output.nii")

Thus is the problem with the resampler? Would you expect the issue to be resolved if you give an input affine transformation if the issue was the resampler?

[ALEXJAZZ008008]

Does #832 have anything to do with it? I have pulled master.

[ALEXJAZZ008008]

Example code and results: resample is the first code snippet from above, affine is the second code snippet from above and no_resample is the third code snippet from above.

resample: https://drive.google.com/file/d/1vubD6Lx6m7ao_zLILxJ-Ae_stBfFVOW4/view?usp=sharing
affine: https://drive.google.com/file/d/1pQgO8ij2qOBttdFZ0IC3SjM0ZE3qsYkz/view?usp=sharing
no_resample: https://drive.google.com/file/d/1ZOAThIfLD7cG_L0D88hhjFzVUDvypwpe/view?usp=sharing

Alternatively: https://github.com/ALEXJAZZ008008/SIRF_niftyreg_f3d2_test

[KrisThielemans]

what's the status of this one? Should we review it again?

[KrisThielemans]

I guess we should check if both are present and then throw an error that we cannot handle it

[KrisThielemans]

is this correct? create_from_cpp always call this?

[ALEXJAZZ008008]

This isn't correct, this will only work with f3d2, I wrote it this way for speed of implementation. Really we should check to see if _use_velocity is set

[ALEXJAZZ008008]

The problem being that _use_velocity is a member variable of NiftyF3dSym

[KrisThielemans]

not sure what you mean, but maybe this just needs 2 different functions create_from_cpp and create_from_velocity_field. Or it needs some header inspection

[KrisThielemans]

Let's convert this to Draft at the moment, till you're done.

By the way, for whatever reason the PR removed requirements.txt