Mni coords to array indices missing in nilearn?

1

Hi,

I needed the inverse function to nilearn.image.coord_transform
I didn’t find in nilearn, so I had to implement it.
I believe it should be there, already.
(maybe I should have opened this as an enhancement request in github)

What I’ve done is simply inverting the affine and multiply it by the needed coordinate, and then I get the indices…
Any comments about this pseudo code or a suggested pseudo code for implementing this inverse function?

@ChrisGorgolewski
@GaelVaroquaux

(Edit: I’ve tested my implementation and it seems fine (extracted some known region and compared its mean with masker.fit_transform and asserted that coord_transform(myfunction((x,y,z))=(x,y,z) ))

Thanks!

2

Hi @brai,

Inverse transforms of affines sounds like something nibabel would have solved,
you can check out their affine example (but the entire page is a great read to understand coordinates/coordinate systems)

Best,
James

1 Like
3

Hi, indeed going from coordinates to voxel indices is just mulitplying by the inverse affine and taking the integer part of the result. you can pass the inverse affine to coord_transform or just use np.dot, not sure a new function is needed for that

1 Like