qform not saved

[mrjeffs]

hi matthew et al, just converted both some of my parrec data and then the nibabel/test/data and found no qform in the nii files. the sform is there but no qform. it is in the pr_hdr and the nhdr object. i can access it using the get_qform() method but it does not make it into the actual saved file. This is an issue because some packages do not fall back on the sform when the qform is missing and without offsets things get a little unpredictable for example with ANTS.
can you confirm this behavior and offer a solution?
jeff

[mrjeffs]

FYI found a thread on FSL listserve that says for FSL the sform is primary and qform is fall back, so it depends. j

[matthew-brett]

Could you say more about what exactly you are doing? I guess loading into memory as PARREC format, and then converting to nifti, using a script? Could you give a minimal script that demonstrates the problem?

[mrjeffs]

sure. here is nii header dump straight out of parrec2nii:
toddr@scotty:/...f_control/hbm_group_data/qT1/scs317/source_nii$ fslhd -x SCS_317_WIP_T1_MAP_20B_SENSE_9_1.nii
<nifti_image
nifti_type = 'NIFTI-1+'
header_filename = 'SCS_317_WIP_T1_MAP_20B_SENSE_9_1.nii'
image_filename = 'SCS_317_WIP_T1_MAP_20B_SENSE_9_1.nii'
image_offset = '39824'
ndim = '3'
nx = '240'
ny = '240'
nz = '116'
dx = '1'
dy = '1'
dz = '1.1'
datatype = '512'
datatype_name = 'UINT16'
nvox = '6681600'
nbyper = '2'
byteorder = 'LSB_FIRST'
cal_min = '0'
cal_max = '4.09248e+06'
scl_slope = '2532.47'
scl_inter = '0'
xyz_units = '2'
xyz_units_name = 'mm'
sform_code = '2'
sform_code_name = 'Aligned Anat'
sto_xyz_matrix = '-1 0 0 115.319 0 1 0 -101.889 0 0 1.1 -59.98 0 0 0 1'
sto_ijk matrix = '-1 -0 -0 115.319 -0 1 -0 101.889 -0 -0 0.909091 54.5273 0 0 0 1'
sform_i_orientation = 'Right-to-Left'
sform_j_orientation = 'Posterior-to-Anterior'
sform_k_orientation = 'Inferior-to-Superior'
num_ext = '1'
/>

you see only sform data in the header.
here is header with both qform and sform data present:

toddr@scotty:/...ntrol/hbm_group_data/ants_template/template_v1$ fslhd -x SCS353_WIP_MEMP_VBM_SENSE_11_1_rms.nii.gz
<nifti_image
nifti_type = 'NIFTI-1+'
header_filename = 'SCS353_WIP_MEMP_VBM_SENSE_11_1_rms.nii.gz'
image_filename = 'SCS353_WIP_MEMP_VBM_SENSE_11_1_rms.nii.gz'
image_offset = '352'
ndim = '3'
nx = '210'
ny = '352'
nz = '352'
dx = '0.8'
dy = '0.682'
dz = '0.682'
datatype = '16'
datatype_name = 'FLOAT32'
nvox = '26019840'
nbyper = '4'
byteorder = 'LSB_FIRST'
xyz_units = '2'
xyz_units_name = 'mm'
time_units = '8'
time_units_name = 's'
descrip = 'FreeSurfer May 13 2013'
qform_code = '1'
qform_code_name = 'Scanner Anat'
qto_xyz_matrix = '-0.794786 0.0593181 0.0502433 61.8522 0.0693908 0.679415 -0.00438754 -110.569 0.0591607 9.24452e-07 0.680133 -102.282 0 0 0 1'
qto_ijk_matrix = '-1.24185 0.108423 0.0924386 98.2543 0.127532 1.46072 1.98699e-06 153.622 0.108021 -0.00943306 1.46226 141.838 0 0 0 1'
quatern_b = '-0.0434998'
quatern_c = '-0.998368'
quatern_d = '-0.00161131'
qoffset_x = '61.8522'
qoffset_y = '-110.569'
qoffset_z = '-102.282'
qfac = '-1'
qform_i_orientation = 'Right-to-Left'
qform_j_orientation = 'Posterior-to-Anterior'
qform_k_orientation = 'Inferior-to-Superior'
sform_code = '1'
sform_code_name = 'Scanner Anat'
sto_xyz_matrix = '-0.794786 0.0593181 0.0502431 61.8522 0.0693908 0.679415 -0.00438753 -110.569 0.0591604 9.33106e-07 0.680133 -102.282 0 0 0 1'
sto_ijk matrix = '-1.24185 0.108423 0.0924382 98.2542 0.127532 1.46072 2.00531e-06 153.622 0.108021 -0.00943304 1.46226 141.838 0 0 0 1'
sform_i_orientation = 'Right-to-Left'
sform_j_orientation = 'Posterior-to-Anterior'
sform_k_orientation = 'Inferior-to-Superior'
num_ext = '0'
/>

[mrjeffs]

here is the command with print statements inserted in parrec2nii to expose affine and qform values:
toddr@scotty:~/Software/nibabel/nibabel/tests/data$ parrec2nii -o parrec2nii_test --overwrite --scaling='fp' phantom_varscale.REC
orig affine = [[ -3.64994708 0. 1.83564171 123.66276611]
[ 0. -3.75 0. 115.617 ]
[ 0.86045705 0. 7.78655376 -27.91161211]
[ 0. 0. 0. 1. ]]

ornt after LAS dot affine = [[ -3.64994708 0. 1.83564171 123.66276611]
[ 0. -3.75 0. 115.617 ]
[ 0.86045705 0. 7.78655376 -27.91161211]
[ 0. 0. 0. 1. ]]
affine after LAS+ = [[ -3.64994708 0. 1.83564171 123.66276611]
[ 0. 3.75 0. -120.633 ]
[ 0.86045705 0. 7.78655376 -27.91161211]
[ 0. 0. 0. 1. ]]
affine at save = [[ -3.64994708 0. 1.83564171 123.66276611]
[ 0. 3.75 0. -120.633 ]
[ 0.86045705 0. 7.78655376 -27.91161211]
[ 0. 0. 0. 1. ]]
nhdr getqform = [[ -3.64994711 0. 1.83564145 123.6627655 ]
[ 0. 3.75 0. -120.63300323]
[ 0.86045693 0. 7.78655383 -27.91161156]
[ 0. 0. 0. 1. ]]

toddr@scotty:/Software/nibabel/nibabel/tests/data$ fslorient -getqform parrec2nii_test/phantom_varscale.nii
3.75 0 0 0 0 3.75 0 0 0 0 8 0 0 0 0 1
toddr@scotty:/Software/nibabel/nibabel/tests/data$ fslhd -x parrec2nii_test/phantom_varscale.nii
<nifti_image
nifti_type = 'NIFTI-1+'
header_filename = 'parrec2nii_test/phantom_varscale.nii'
image_filename = 'parrec2nii_test/phantom_varscale.nii'
image_offset = '352'
ndim = '4'
nx = '64'
ny = '64'
nz = '9'
nt = '3'
dx = '3.75'
dy = '3.75'
dz = '8'
dt = '2'
datatype = '64'
datatype_name = 'FLOAT64'
nvox = '110592'
nbyper = '8'
byteorder = 'LSB_FIRST'
cal_min = '-10516.8'
cal_max = '645360'
scl_slope = '1'
scl_inter = '0'
xyz_units = '2'
xyz_units_name = 'mm'
time_units = '16'
time_units_name = 'ms'
sform_code = '2'
sform_code_name = 'Aligned Anat'
sto_xyz_matrix = '-3.64995 0 1.83564 123.663 0 3.75 0 -120.633 0.860457 0 7.78655 -27.9116 0 0 0 1'
sto_ijk matrix = '-0.259552 -0 0.0611881 33.8047 -0 0.266667 -0 32.1688 0.0286819 -0 0.121665 -0.15102 0 0 0 1'
sform_i_orientation = 'Right-to-Left'
sform_j_orientation = 'Posterior-to-Anterior'
sform_k_orientation = 'Inferior-to-Superior'
num_ext = '0'
/>

[mrjeffs]

i have a fix which uses fslorient -copysform2qform in a subshell. its inelegant but needs must.
the qform is in the nimg object when saved as you can see from nhdr.get_qform() from last print stmnt middle of last post. somewhere in nifti1.save it is discarded and only sform is written to header.

[mrjeffs]

Having now looked in detail at nifti1.py header_dtd definitions there is no qform matrix definition (or sform matrix for that matter), just the qform code, quaternions, and qoffsets but they don't seem to survive writing to file. the sform matrix gets generated presumably from that, tho i havn't figured out how yet. seems possible the qform matrix could be easily generated as well, then written.
I personally vote to include the qform matrix in the nii header when its appropriate in your development cycle.

[matthew-brett]

The qform matrix derives from the translations, composed on the quaternion rotations composed with the zooms with an extra flip - description here : http://nifti.nimh.nih.gov/nifti-1/documentation/nifti1fields/nifti1fields_pages/qsform.html - implemented at https://github.com/nipy/nibabel/blob/master/nibabel/nifti1.py#L889

I'm guessing that the problem here is that, when you create a new blank NIfTI header, and ask to set an affine matrix into the header, the default qform has code 0 ('unknown') - see https://github.com/nipy/nibabel/blob/master/nibabel/nifti1.py#L1738 . I guess you want the affine code to be something else, such as 2 'aligned'. This will then show up with (for example) the FSL output you are looking at.

[mrjeffs]

yes indeed. how could than implemented without new code? do
nhdr.set_qform(affine, code=2) after setting up nifti1 object and before saving?
i will check if qform code = 0 eventho qform matrix was present. i had assumes it was 2 since the affine was there.

[matthew-brett]

Right - set_qform(affine, code=2) should do it.

[mrjeffs]

thanks matthew, that worked. problem solved. i would recommend the one line added to the parrec2nii binary script at line 225 in a pull request. is it faster for you or shall i?

# Make corresponding NIfTI image
nimg = nifti1.Nifti1Image(in_data, affine, pr_hdr)
nhdr = nimg.header
nhdr.set_data_dtype(out_dtype)
nhdr.set_slope_inter(slope, intercept)
nhdr.set_qform(affine, code=2)                       #line 225 added to pull request

[matthew-brett]

Would you mind doing it? If you have time, it would be good to have a test too.

[mrjeffs]

no worries. i'll give it a go.

[matthew-brett]

@mrjeffs - any progress? Can I help?

[matthew-brett]

We're getting to close to a release - any chance of PR for this in the next couple of days?

[mrjeffs]

congrats to jasper for doing the heavy lifting on building the tests.