Qsiprep preprocessing anat native files wrong space

snp2003 · December 2, 2021, 2:50pm

Dear team,

I have ran a subject with qsiprep preprocessing pipeline with the following command:
docker run -ti --rm --runtime=gpus -v /home/ubuntu/tracktbi/niftis:/data:ro -v /home/ubuntu/tracktbi/derivatives:/out -v ${fslicense}:/opt/freesurfer/license.txt -v /home/ubuntu/tracktbi/work:/work -v /home/ubuntu/tracktbi/config_files:/config_files pennbbl/qsiprep:0.14.3 /data /out participant --participant-label 011007s20140402 --skip-bids-validation --denoise-method dwidenoise --unringing-method mrdegibbs -v --use-syn-sdc --output-space template --template MNI152NLin2009cAsym --output-resolution 2 --eddy-config /config_files/eddy_params.json --low-mem -w /work

After completion, I noticed that native files in the anat folder (files without space label in the filenames) are in the MNI space as well, please see below. Is this an expected behavior or an anomaly?

raw:

/data/sub-011007s20140402/anat/sub-011007s20140402_T1w.nii.gz            int16 [176, 240, 256]      1.20x1.00x1.00        sform
/data/sub-011007s20140402/dwi/sub-011007s20140402_dwi.nii.gz             int16 [128,  84,  59,  65] 2.73x2.73x2.70x9.00
/data/sub-011007s20140402/func/sub-011007s20140402_task-rest_bold.nii.gz int16 [ 64,  64,  48, 200] 3.31x3.31x3.96x3.00

anat:

anat/sub-011007s20140402_desc-brain_mask.nii.gz                             float64 [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_desc-preproc_T1w.nii.gz                            float64 [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_dseg.nii.gz                                         int32  [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_label-CSF_probseg.nii.gz                           float32 [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_label-GM_probseg.nii.gz                            float32 [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_label-WM_probseg.nii.gz                            float32 [193, 229, 193] 1.00x1.00x1.00   sform
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_desc-brain_mask.nii.gz   float32 [193, 229, 193] 1.00x1.00x1.00
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_desc-preproc_T1w.nii.gz  float32 [193, 229, 193] 1.00x1.00x1.00
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_dseg.nii.gz              float32 [193, 229, 193] 1.00x1.00x1.00
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_label-CSF_probseg.nii.gz float32 [193, 229, 193] 1.00x1.00x1.00
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_label-GM_probseg.nii.gz  float32 [193, 229, 193] 1.00x1.00x1.00
anat/sub-011007s20140402_space-MNI152NLin2009cAsym_label-WM_probseg.nii.gz  float32 [193, 229, 193] 1.00x1.00x1.00

dwi:

dwi/sub-011007s20140402_space-T1w_desc-brain_mask.nii.gz   uint8  [ 97, 116,  98]      2.00x2.00x2.00
dwi/sub-011007s20140402_space-T1w_desc-eddy_cnr.nii.gz    float32 [ 97, 116,  98]      2.00x2.00x2.00        sform
dwi/sub-011007s20140402_space-T1w_desc-preproc_dwi.nii.gz float32 [ 97, 116,  98,  65] 2.00x2.00x2.00x0.00   sform
dwi/sub-011007s20140402_space-T1w_dwiref.nii.gz           float64 [ 97, 116,  98]      2.00x2.00x2.00

Thank you,
Sneha

Steven · December 2, 2021, 6:50pm

Hi Sneha,

The files without the space label should be in ACPC space (native space, but realigned to ACPC). Then the DWI derivatives should be aligned to those ACPC T1 images.

Can you post some screenshots of the raw and preprocessed T1s?

Best,
Steven

mattcieslak · December 2, 2021, 8:13pm

Hi Sneha,

QSIPrep won’t write preprocessed images out to any space other than native AC-PC aligned. The reasoning is that the T1w-driven registration to template space, while great for getting cortical atlas regions into alignment with the DWIs, is not very accurate for white matter structures. It is better to spatially normalize reconstructed data, which you can do on the results of the the mrtrix or DSI Studio reconstruction workflows. We plan on adding spatial normalization for reconstructed ODF images in a future release.

snp2003 · December 3, 2021, 6:40pm

Dear @Steven and @mattcieslak thank you for clarification. I think filenames in general and the output size and resolution of anat preprocessed files from fmriprep makes it confusing when we refer to non-MNI space labels in their “native space”. As stressed by both of you I think it makes sense when we say native space but ACPC aligned. However, is it possible to not perform resampling once realigned to ACPC and maintain the same size and resolution as that of raw T1 image?

@Steven please see below my raw and preprocessed T1s

Raw (*_T1w.nii.gz):

Preprocessed native ACPC aligned (*desc-preproc_T1w.nii.gz):

Preprocessed MNI aligned (*space-MNI152NLin2009cAsym_desc-preproc_T1w.nii.gz):