Error in bias correction step: header information seems to differ between brainmask and rescaled T2?

1

I have a problem with segmenting several T2’s.
During the bias correction step, it seems that the header information of the brainmask and the rescaled T2 differ and therefore the pipeline stops running after creation of the brainmask.
This is the log:

Description: itk::ERROR: N4BiasFieldCorrectionImageFilter(0x55b1631eb6b0): Inputs do not occupy the same physical space!
InputImage Origin: [-8.6477035e+01, 6.7219341e+01, -6.4898184e+01], InputImage_1 Origin: [-8.6476832e+01, 6.7219339e+01, -6.4898203e+01]
Tolerance: 7.0312500e-07
InputImage Spacing: [7.0312500e-01, 7.0312500e-01, 4.0000005e+00], InputImage_1 Spacing: [7.0312399e-01, 7.0312399e-01, 4.0000000e+00]
Tolerance: 7.0312500e-07

I don’t have this problem with the majority of my scans, and they are all processed the same. The quality of the T2 scans that give this error is fairly well so that does not seem to be the problem.
Manually changing the header information did not solve this issue.

Does anyone have a solution to this problem?
Should I try running the bias correction step without the brain mask or does that give too much variance in my results?

Thanks in advance!

2

Can you share a sample dataset directly with me (e.g. send a personal message to me with a link to sample data on Google drive)? Alternatively, maybe you can report the output of these two images as seen using fslhd.

Be aware that ITK-based tools like N4 give precedence to the NIfTI Q-Form, while most tools give precedence to the S-Form. Therefore, it is possible that these two images appear aligned with each other based on the S-Form, but not the Q-Form. The issue is that ITK demands rectangular volumes, where the Matrix of the S-Form can in theory store shears. The quaternion-based Q-Form is more constrained (9 degrees of freedom) than the S-Form (12 degrees of freedom). fslhd reports both, which can be helpful in troubleshooting.

3

Thank you for your reply.
I sent you an email with the link to the files.