Bold inhomogeneity correction

[JonasVejmelka]

Hi all,

I am currently preprocessing our 4.7T lister hooded rat data, preparing for seed-based analysis and I ran into some issues.

It generally seems to run pretty good, however, I am encountering some problems with the BOLD images.

1. I have 4 EPI runs per subject, and on some runs it seems that while building the robust EPI inhomogeneity template, RABIES generates an intermediate NIfTI file that does not have an orthonormal direction cosine. This leads to ITK not being able to read it, which leads to a crash on that specific run which is then missing from final preprocessing. On some subjects, it crashed the whole process. Is there a way I can fix this or rather prevent this from happening? I am attaching a logfile of a subject with one crashed run.

2. Second issue I run into is inhomogeneity correction for BOLD images failing completely, creating a mask that is small and outside of the brain borders on some runs while working perfectly for others of the same subject. Sometimes it seems to have picked up a signal from outside the brain, other times that doesn't seem to be the case. I already have pretty strict parameters and am not sure what to improve further. I was thinking I might use autobox cropping for the bold images of subjects with this issue. Are there any other ways I could fix this? I am attaching QC images from subjects with this issue.

Thank you in advance!

rabies_preprocess-11.log

[gdevenyi]

Hello.

Could you please include all the information requested in the issue template. You have not provided the call you used.

[JonasVejmelka]

Hi again,
Apologies for not including the call used. The call I used was as follows:

  taskset -c 0-12 singularity run \
    -B "$bids_input":/input_BIDS_full:ro \
    -B "$preprocess_outputs":/preprocess_outputs \
    -B "$sigma_dir":/SIGMA \
    -B "$tmp_dir":/tmp_rabies \
    "$rabies_image" \
      -p MultiProc \
      --local_threads=13 \
      --scale_min_memory=1 \
      preprocess \
      /input_BIDS_full/$subj \
      /preprocess_outputs/$subj \
      --apply_STC \
      --TR=0.801 \
      --anat_inho_cor="method=SyN,otsu_thresh=2,multiotsu=false" \
      --anat_robust_inho_cor="apply=true,masking=true,brain_extraction=true,template_registration=SyN" \
      --commonspace_reg="masking=true,brain_extraction=true,template_registration=SyN,fast_commonspace=false" \
      --commonspace_resampling=0.3x0.3x0.3 \
      --nativespace_resampling=inputs_defined \
      --HMC_option=intraSubjectBOLD \
      --bold_inho_cor="method=Rigid,otsu_thresh=3,multiotsu=true" \
      --bold_robust_inho_cor="apply=true,masking=true,brain_extraction=true,template_registration=SyN" \
      --bold2anat_coreg="masking=true,brain_extraction=true,registration=SyN" \
      --anat_template=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Anatomical_Imaging/SIGMA_Rat_Anatomical_InVivo_Template/SIGMA_InVivo_Brain_Template.nii \
      --brain_mask=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Anatomical_Imaging/SIGMA_Rat_Anatomical_InVivo_Template/SIGMA_InVivo_Brain_Mask.nii \
      --WM_mask=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Anatomical_Imaging/SIGMA_Rat_Anatomical_InVivo_Template/SIGMA_InVivo_WM_mask.nii \
      --CSF_mask=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Anatomical_Imaging/SIGMA_Rat_Anatomical_InVivo_Template/SIGMA_InVivo_CSF_mask.nii \
      --vascular_mask=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Anatomical_Imaging/SIGMA_Rat_Anatomical_InVivo_Template/SIGMA_InVivo_CSF_mask.nii \
      --labels=/SIGMA/SIGMA_Wistar_Rat_Brain_TemplatesAndAtlases_Version1.2.1/SIGMA_Rat_Brain_Atlases/SIGMA_Anatomical_Atlas/InVivo_Atlas/SIGMA_InVivo_Anatomical_Brain_Atlas.nii \

[CMonnin]

hi! it looks like you applied the SIGMA_InVivo_CSF_mask.nii as a vascular_mask

[JonasVejmelka]

Hey, that's true - it was more of a placeholder but I will remove it. However, it shouldn't be the issue now because I am not using it in any way at this stage.

[Gab-D-G]

Hello @JonasVejmelka , sorry for the late response.

issue 1) This is odd, I'm not sure why this arise so late into the pipeline. You are using a relatively old version, 0.4.8, and the more recent one has an option to correct orthonormal directions, so how about you try applying --oblique2card with the new version?

issue 2) So this looks like a classic masking failure (3rd column in the report) on your EPI. Make sure you have checked that structural preprocessing steps have not failed for that particular subject first, because if the structural scan failed, then later EPI steps fail as well. Otherwise, I think you have the right intuition about autobox - I see the registratino is failing because of extra tissue outside the brain, so I advise to try that before anything else.

let me know if I can help further!