T1w image and its manual segmentation does not match in 3 sub-brnoUhb scans

[renelabounek]

I see 3 scans where manual T1w segmentation mismatch T1w image on the latest database release. It seems like qform and sform matrices needs tobe matched.

See yourself:
cd data-multi-subject
fsleyes sub-brnoUhb02/anat/sub-*_T1w.nii.gz derivatives/labels/sub-brnoUhb02/anat/sub-*_T1w_seg-manual.nii.gz &
fsleyes sub-brnoUhb07/anat/sub-*_T1w.nii.gz derivatives/labels/sub-brnoUhb07/anat/sub-*_T1w_seg-manual.nii.gz &
fsleyes sub-brnoUhb08/anat/sub-*_T1w.nii.gz derivatives/labels/sub-brnoUhb08/anat/sub-*_T1w_seg-manual.nii.gz &

I have not checked T2w and T2star segmentation for these subjects. There may be the same issue.

This issue impacts quality of process_data.sh outcomes as noted here

[renelabounek]

It is not just about transformation matrices. The images have different resolutions:

fslinfo sub-brnoUhb02/anat/sub-*_T1w.nii.gz
data_type INT16
dim1 188
dim2 512
dim3 512
dim4 1
datatype 4
pixdim1 0.999941
pixdim2 0.625000
pixdim3 0.625000
pixdim4 0.004784
cal_max 0.000000
cal_min 0.000000
file_type NIFTI-1+

fslinfo derivatives/labels/sub-brnoUhb02/anat/sub-*_T1w_seg-manual.nii.gz
filename derivatives/labels/sub-brnoUhb02/anat/sub-brnoUhb02_T1w_seg-manual.nii.gz
data_type UINT16
dim1 188
dim2 320
dim3 320
dim4 1
datatype 512
pixdim1 0.999941
pixdim2 1.000000
pixdim3 1.000000
pixdim4 0.000000
cal_max 0.000000
cal_min 0.000000
file_type NIFTI-1+

[renelabounek]

T2w and T2star segmentations match. So the issue is only with T1w scans.

[jcohenadad]

I suspect these issues came up when we had to re-upload the properly defaced images 082dd0d

[renelabounek]

I suspect these issues came up when we had to re-upload the properly defaced images 082dd0d

That seems to be the cause. Luckily, only 3 scans seem affected. Do you need help to re-segment these 3 scans? Or is it easier to fix it directly in Montreal? Either ways are fine with me.

[jcohenadad]

Do you need help to re-segment these 3 scans?

The segmentation should absolutely not be re-done. It is only a matter of resampling the T1w data so that it matches that of the segmentation. Given the large matrix of the T1w (512x512, due to post-acquisition oversampling by the proprietary MR software) I suspect during the early curation process we resampled all T1w scans to ~1mm iso. So in principle by applying this transformation we should be able to get the T1w in the space of the T1w_seg.

[jcohenadad]

I also noticed the same issue (ie: mismatch of q/sform between image and segmentation) for T2w images. Eg: sub-cmrrb03_T2w_seg-manual.nii.gz. We should ideally resample/reorient the segmentations.

In the meantime, to properly use the segmentations, the native images should be preprocessed (reorientation, resampling) as instructed in the analysis pipeline:

https:/spine-generic/spine-generic/blob/master/process_data.sh#L150-L151