Lesion mask not reoriented to LPS

[rbakker]

Summary

When I use qsiprep with a lesion mask in my bids folder, it does apply the mask in the nonlinear transformation to MNI-space part, but not correctly so. The mask is used in a wrong orientation.

Additional details

• QSIPrep version: 1.0.2.dev0+gfc89945.d20250405
• Docker version:
• Singularity version: apptainer

What were you trying to do?

Run the anatomy-only pipeline with a lesion mask

What did you expect to happen?

lesion mask applied correctly to the lesion

What actually happened?

if I compare with and without lesion mask, I see subtle differences elsewhere in the brain, but not around the lesion.
I have the suspicion that qsiprep starts to resample all files to LPS space, but not the lesion mask, hence the error.
All files I have have RAS orientation, with identity matrix affine transforms.

Reproducing the bug

Run qsiprep anat-only with T1w.nii.gz in RAS orientation and identity affine transform.
Put lesion mask in sub-{subject}_ses-{session}_T1w_label-lesion_roi.nii.gz (and disable bids validation).
The lesion mask will be applied but in a wrong orientation.

-->

[rbakker]

This error is actually caused by a bug in niworkflows/interfaces/norm.py

Line 516

def create_cfm(in_file, lesion_mask=None, global_mask=True, out_path=None):

where it subtracts the lesion_mask from the global mask.
Before it does so, it calls in Line 576

lm_img = nb.as_closest_canonical(nb.load(lesion_mask))

But this as_closest_canonical destroys the LPS alignment of the lesion_mask.

And then in the next line the now RAS-oriented lesion mask is subtracted from the LPS-oriented in_file.
If any reorientation is applied, then it should be a specific one from the lesion_mask orientation to the in_file orientation.
Nibabel has the machinery to do this properly. But one can also assume that in_file and lesion_mask have the same orientation.

[kikiluvbrains]

Hi,

I suspect that the T1w and lesion mask might be in different orientations, which could explain the behavior you’re seeing. Could you try printing the following outputs to check their affines?

py

import nibabel as nb
t1 = nb.load("T1w.nii.gz")
lesion = nb.load("lesion_mask.nii.gz")

print(t1.affine)
print(lesion.affine)

If you have any sample data or a similar dataset available, I’d be happy to help investigate further.