Issue about segmentation result with data including lesion-mask

[shz8012]

Hi,

I had a dataset of stroke patients(T1w, fmri, lesion-mask) and tried to run fmriprep to get structural and functional results. Lesion-mask image was placed and named correctly.

When I checked the images of segmentation (e.g./.../fmriprep/sub-FCS029/anat/sub-FCS029_label-GM_probseg.nii.gz), I found that there were values in the voxels belonging to lesion-mask, as shown below (red is lesion-mask)

However, I would like to get segmentation results with voxels value of 0 at the location of lesion, like this: (I used Atropos with mask to get this segmentation result, forgive me for forgetting to skull stripping)

Here were my question:

1. In the workflow of fmriprep, was lesion-mask used in segmentation?
2. Was the output of node fmriprep_wf.single_subject_FCS029_wf.anat_preproc_wf.brain_extraction_wf.atropos_wf.01_atropos the source of segmentation results?
3. What nodes were lesion-mask used for ? (what I know was fmriprep_wf.single_subject_FCS029_wf.anat_preproc_wf.anat_norm_wf.registration)

Thanks! shz8012

[effigies]

To answer the questions, we'll need to know the version of fMRIPrep you were using. Your full command and a listing of our input dataset would also be helpful.

The segmentation results will be either FAST or FreeSurfer derived, depending on your command and version. The lesion mask is only currently used in normalization. It seems sensible to also use it in skull-stripping and segmentation; if you have example commands of how that would be done normally, that would be a great help.

[shz8012]

I am sorry about that.

Here is additional information:

• fmriprep version: 20.2.1(docker)

• command:

   fmriprep-docker \
  /data/disk1/test/ \
  /data/disk1/test_output \
  --participant-label sub-FCS029 \
  --cifti-output 91k --fs-license-file /opt/software/freesurfer/license.txt \
  --use-aroma --user $(id -u):$(id -g) 

• input data:

• segmentation command using lesion-mask: To exclude changes in parameters, I used the exact same command as in node fmriprep_wf.single_subject_FCS029_wf.anat_preproc_wf.brain_extraction_wf.atropos_wf.01_atropos

  Atropos --image-dimensionality 3 --initialization KMeans[3] \
  --intensity-image /scratch/fmriprep_wf/single_subject_FCS029_wf/anat_preproc_wf/brain_extraction_wf/inu_n4_final/mapflow/_inu_n4_final0/sub-FCS029_T1w_corrected.nii.gz \
  --likelihood-model Gaussian \
  --mask-image /scratch/fmriprep_wf/single_subject_FCS029_wf/anat_preproc_wf/brain_extraction_wf/atropos_wf/get_brainmask/tpl-OASIS30ANTs_res-01_label-brain_probseg_trans_resampled_maths_maths.nii.gz \
  --mrf [0.1,1x1x1] --convergence [3,0] \
  --output [sub-FCS029_T1w_corrected_labeled.nii.gz,POSTERIOR_%02d.nii.gz] --use-random-seed 1 

The only difference was that mask image that followed flag --mask-image is a mask that did not contain lesion region (the mask image like this):

I was looking forward to fmriprep taking more account of lesion data in the future!

Thanks shz8012