How can I transform two CTs with a known affine transformation matrix?

[angolin22]

hello, thanks for your excellent work! I have a problem:
I have two 3D CT images, and I know the affine transformation matrix between them. e. g.

[[ 9.92477011e-01 -8.45505280e-02  8.85471061e-02  5.05205949e+01] 
[ 8.73986445e-02  9.95757283e-01 -2.87907932e-02 -5.85686522e+00] 
[-8.57371490e-02  3.63130975e-02  9.95655814e-01 -4.96898369e+01]
[-0.00000000e+00 -0.00000000e+00 -0.00000000e+00  1.00000000e+00]]

How can I transform CT with it?
Thanks

[adalca]

Sounds like a job for @mu40

[angolin22]

May I ask if @mu40 has completed this part of the content?

[adalca]

im not sure what you mean, but @mu40 has been spearheading our affine work and code.

[angolin22]

May I ask where is the affine code and how to use it?

[mu40]

Assuming your images moving and fixed are defined on the same voxel grid, and your matrix transforms the centered voxel coordinates of fixed to those of moving, moving moving can be as simple as

trans = matrix[..., :-1, :]
moved = vxm.layers.SpatialTransformer()((moving, trans))

where moving and fixed would have a leading batch as well as a trailing feature dimension.

For the more general case, you can pass shift_center=False to interpret matrix as a transform from zero-based voxel indices of the fixed to zero-based voxel indices of the moving image. You can also set the output shape, should your images live in different voxel spaces:

shape = fixed.shape[1:-1]
moved = vxm.layers.SpatialTransformer(shift_center=False, shape=shape)((moving, trans))

[mu40]

If this is still relevant, @angolin22, we now have a tutorial on affine registration at w3id.org/synthmorph.