FastSymmetricForcesDemonsRegistrationFilter.SetUseImageSpacing() not working?

Beginner Questions
1

It seems that FastSymmetricForcesDemonsRegistrationFilter.SetUseImageSpacing() does not work, I get the same result when setting it to True or False, and same thing with FastSymmetricForcesDemonsRegistrationFilter.UseImageSpacingOn().

Here is a script which demonstrates the behavior:

standard_deviations = 1.0
num_iterations = 200

def demons_registration(fixed_path, moving_path, displacement_field_path, use_image_spacing=True, image_spacing_on=True):

    # Open images with SimpleITK
    fixed = sitk.ReadImage(str(fixed_path), sitk.sitkFloat64)
    moving = sitk.ReadImage(str(moving_path), sitk.sitkFloat64)

    print('      Demons registration...')

    # The fast symmetric forces Demons Registration Filter
    # Note there is a whole family of Demons Registration algorithms included in
    # SimpleITK (https://itk.org/Doxygen/html/group__DeformableImageRegistration.html)
    demons = sitk.FastSymmetricForcesDemonsRegistrationFilter()
    demons.SetUseImageSpacing(use_image_spacing)

    if image_spacing_on:
        demons.UseImageSpacingOn()
    else:
        demons.UseImageSpacingOff()

    demons.SetNumberOfIterations(num_iterations)
    # Standard deviation for Gaussian smoothing of displacement field
    demons.SetStandardDeviations(standard_deviations)

    displacement_field = demons.Execute(fixed, moving)
    
    displacement_field_data = sitk.GetArrayFromImage(displacement_field)

    sitk.WriteImage(displacement_field, str(displacement_field_path))

    return displacement_field_path

flair1 = '/data/SEL/emisepTesting0/rennes_20170112_05/flair_time01_normalized.nii.gz'
flair2 = '/data/SEL/emisepTesting0/rennes_20170112_05/flair_time02_normalized.nii.gz'

df = '/data/amasson/test/df/df_spacing11.nii.gz'
demons_registration(flair1, flair2, df, True, True)
df11_image = sitk.ReadImage(df)
df11_data = sitk.GetArrayFromImage(df11_image)

df = '/data/amasson/test/df/df_spacing10.nii.gz'
demons_registration(flair1, flair2, df, False, True)
df10_image = sitk.ReadImage(df)
df10_data = sitk.GetArrayFromImage(df10_image)
print(np.all(df11_data == df10_data))

df = '/data/amasson/test/df/df_spacing01.nii.gz'
demons_registration(flair1, flair2, df, True, False)
df01_image = sitk.ReadImage(df)
df01_data = sitk.GetArrayFromImage(df01_image)
print(np.all(df01_data == df10_data))

df = '/data/amasson/test/df/df_spacing00.nii.gz'
demons_registration(flair1, flair2, df, False, False)
df00_image = sitk.ReadImage(df)
df00_data = sitk.GetArrayFromImage(df00_image)
print(np.all(df01_data == df00_data))

And here is the output showing that all images are equal:

  Demons registration...
  Demons registration...
True
  Demons registration...
True
  Demons registration...
True
2

This is expected to produce different results only if spacing is different from 1.0 along each axis. If spacing is 1, this option will not make a difference for the result, only for speed of computation. What is the spacing of your images?

3

Yes, my image spacing is (0.498046875, 0.498046875, 1.100000023841858)

4

Some filters do not respect spacing properly, but a registration filter should care a lot about spacing. This filter delegates most of the work to other filters, so it is not easy to find out where the spacing is being taken into account. It is possible that this filter ignores the inherited UseImageSpacing setting and always uses image spacing. Why do you want to experiment with this setting?