Left-Right Flip a Warp field?

1

Hi All,

I have a peculiar situation:

I need to estimate a warp field (we are using ANTS). Call this warp1

Later I need to L-R flip my image and apply further warping on the flipped image. Unfortunately, this means I need to estimate the warp field all over again for the flipped image. Call this warp2.

I’m wondering if it would make sense to L-R flip the warp1 field and use that for transformations instead of estimating warp2 for the flipped image. I’m thinking this shortcut would save a lot of time, but I’m concerned it might be wrong-headed.

Thoughts? Thank you.

2

Hi @Dianne_Patterson ,

To my understanding, flipping the warp field warp1 would only work if the target image when you calculated this warp field was symetric. If not, you would need to calculate warp2 that will be different from the L-R flipped version of warp1.

3

Thank you, I was hopeful it might work ; (
-D

4

You’re flipping both the moving and fixed images in the same way?

The flip is perfectly symmetric (e.g., horizontal flip about the midline)?

The warping process is symmetric and deterministic (as with ANTs SyN, assuming consistent parameters)?

import nibabel as nib
import numpy as np

warp_img = nib.load('warp1.nii.gz')
warp_data = warp_img.get_fdata()
affine = warp_img.affine

# Flip L-R axis (axis=0), and negate x-displacement component
warp_flipped = np.flip(warp_data, axis=0)
warp_flipped[..., 0] *= -1

# Save new flipped warp
flipped_img = nib.Nifti1Image(warp_flipped, affine)
nib.save(flipped_img, 'warp1_flipped.nii.gz')

Do you want help writing a script to automate the flip + check process?