Registration of images which do not cover the entire brain

[LuRoe7]

Describe the bug
I want to register a DCE-MRI image into the MNI space and do not achieve a good result. The field of view (FOV) of the DCE-MRI image does not cover the entire brain. I attached a screenshot at the bottom of this post.

To reproduce

# set path to brain extracted DCE-MRI image registered to the native T1w space
patlak_img = ants.image_read('path/to/sub-IMPCSF008_ses-BL_acq-LinReg_PatlakToAnatBrain.nii.gz')

# set path to corresponding mask image
patlak_mask = ants.image_read('/path/to/sub-IMPCSF008_ses-BL_acq-LinReg_PatlakToAnatBrainMask.nii.gz')

# set path to MNI template (only brain)
mni_template = ants.image_read('/path/to/MNI152NLin6Asym_2mm_T1w_brain.nii.gz')

# set path to MNI mask
mni_mask = ants.image_read('/path/to/MNI152NLin6Asym_2mm_T1w_brain_mask.nii.gz')

# run ants registration
reg_result = ants.registration(fixed=mni_template,
moving=patlak_img,
outprefix='/path/to/test_',
type_of_transform='SyN',
mask = mni_mask,
moving_mask=patlak_mask)

# extract the registered moving image
registered_image = reg_result['warpedmovout']

# save it as a NIfTI file
ants.image_write(registered_image, '/path/to/test_out.nii.gz')

# apply transform
patlak_mask_mni = ants.apply_transforms(fixed=mni_template,
moving=patlak_img,
transformlist=['/path/to/test_out.nii.gz', '/path/to/test_0GenericAffine.mat'],
interpolator='linear',
output_dtype='float')

# write image
ants.image_write(patlak_mask_mni, '/path/to/test_Registered.nii.gz')

Screenshots
This is the resulting registration (test_Registered.nii.gz):

And this is how the DCE-MRI image looks like in native T1w space:

Any advice what I could optimize?

[ntustison]

First, try removing the specified masks in the command call.

[cookpa]

Why not register T1w to the MNI T1w, then apply the transforms to the DCE image?

[LuRoe7]

Removing the masks did not change anything:

But registering the T1w to the MNI first and then applying the transforms to the DCE image produces much better results. The code:

# perform registration
registration = ants.registration(fixed=mni_template, moving=t1_img, type_of_transform='SyN')

# get transformation matrices
transforms = registration['fwdtransforms']

# apply the transformation to the DCE image
registered_dce = ants.apply_transforms(fixed=mni_template, moving=patlak_img, transformlist=transforms)

# save the transformed DCE image
ants.image_write(registered_dce,'/path/to/test_Registered.nii.gz')

The output:

Do you see any need for optimization? The fit is not perfect in frontal areas I guess and some voxels appear to be outside the template (possibly due to interpolation?):

[cookpa]

You can try changing syn_metric='mattes', syn_sampling=32 to syn_metric='CC', syn_sampling=2. It will take longer but might be better.