search

FCP-INDI / C-PAC_templates

Templates and atlases that aren't included in Neuroparc and are packaged in C-PAC
BSD 3-Clause "New" or "Revised" License
2 stars 1 forks source link

🐛 `mni_icbm152_t1_tal_nlin_asym_09c` vs. `tpl-MNI152NLin2009cAsym` #18

Open shnizzedy opened 1 year ago

shnizzedy commented 1 year ago

Describe the bug

Both of these templates (mni_icbm152_t1_tal_nlin_asym_09c.nii packaged in C-PAC & tpl-MNI152NLin2009cAsym from TemplateFlow) seem to be templates for space-MNI152NLin2009cAsym but they aren't identical.

Expected behavior

Templates are identical or have different names

Acceptance criteria

C-PAC version

v1.8.5-dev

Additional context

They both have b'mnc2nii mni_icbm152_nlin_asym_09c/mni_icbm152_t1_tal_nlin_asym_09c.mnc mni_icbm' as the description in the header, but mni_icbm152_t1_tal_nlin_asym_09c.nii has

cal_max         : 100.0
qoffset_x       : 0.0
qoffset_y       : 0.0
qoffset_z       : 0.0

where tpl-MNI152NLin2009cAsym_res-01_T1w.nii.gz from TemplateFlow has

cal_max         : 10000.0
qoffset_x       : -96.0
qoffset_y       : -132.0
qoffset_z       : -78.0

They look the same visually,

mni_icbm152_t1_tal_nlin_asym_09c tpl-MNI152NLin2009cAsym_res-01_T1w
mni_icbm152_t1_tal_nlin_asym_09c.png tpl-MNI152NLin2009cAsym_res-01_T1w.png

but get_fdata() shows roughly a 100× scaling between the data points (like the cal_max in the headers, although less precisely):

count   8530021.00
mean    101.04
std 7.21
min -1844.27
25% 100.85
50% 100.92
75% 101.10
max 1029.66

histogram

Gist with files, headers, and images: https://gist.github.com/shnizzedy/03e8ba5ead9987b5a299e2501ec77ae0

gkiar commented 1 year ago

Thanks for using the quickviz script to compare them! Super helpful to see that the correlation of the two is 1.0!

Can we track down a list of all nodes that use the mni_icbm152_t1_tal_nlin_asym_09c file by default? We should do a run that compares a complete pipeline with all of those nodes enabled (if possible), and then repeat it with the Templateflow-named version. I suspect we will see no difference, and then can just remove the old-named version.