Poor functional and anatomical registration

1

Hi,

In our experiment, participants are scanned twice so there are two sessions. The participants have some type of lesion in their brain. For one participant, we are experiencing poor anatomical and functional coregistration for the second session only for all functional runs except one. I tried manually setting the origin for each scan, so there is good initial overlap between functional and anatomical images. Here is one example of a processed functional overlaid on the processed T1w:

image
image891×1099 197 KB

We are using fmriprep 20.2.1 and here is the command used:

singularity run --cleanenv \
    -B /cm/shared/apps/freesurfer/6.0.0:/work_dir \
    -B /scratch/global/jheffernan/experiments/tdcs/templateflow:/opt/templateflow \
    /scratch/global/jheffernan/experiments/tdcs/singularity_images/fmriprep-v20.2.1.simg \
    /scratch/global/jheffernan/experiments/tdcs/bids /scratch/global/jheffernan/experiments/tdcs/derivatives/fmriprep-v20.2.1/sessions-2_manualorigin participant \
    --participant_label mcwa002 \
    --nprocs 16 \
    --omp-nthreads 4 \
    --ignore slicetiming \
    --output-spaces T1w \
    --random-seed 1234567890 \
    --fs-license-file /work_dir/license.txt \
    --fs-no-reconall \
    --work-dir /scratch/global/jheffernan/experiments/tdcs/tmp/fmriprep_mcwa002_ALL_sessions-2_IDvxTZ \
    --stop-on-first-crash

We did not include a lesion mask, because we are only using T1w output-space, so I don’t think it would help. We use --fs-no-reconall, because the brain mask refinement with freesurfer tends to exclude the lesioned areas of the brain. I tried --force-no-bbr option and the results did not look good either. The functional scanning dimensions also changed between sessions, but not much so I do not think it should be a big problem. Here are the parameters for both sessions:

sub-mcwa002_ses-1_task-bh_run-1_bold.nii.gz   int16 [128, 128,  72, 330] 1.62x1.62x2.00x0.80
sub-mcwa002_ses-1_task-rest_run-1_bold.nii.gz int16 [128, 128,  72, 378] 1.62x1.62x2.00x0.80
sub-mcwa002_ses-1_task-rest_run-2_bold.nii.gz int16 [128, 128,  72, 378] 1.62x1.62x2.00x0.80
sub-mcwa002_ses-1_task-rest_run-3_bold.nii.gz int16 [128, 128,  72, 378] 1.62x1.62x2.00x0.80
sub-mcwa002_ses-1_task-SDTD_run-1_bold.nii.gz int16 [128, 128,  72, 378] 1.62x1.62x2.00x0.80
sub-mcwa002_ses-1_task-SDTD_run-2_bold.nii.gz int16 [128, 128,  72, 378] 1.62x1.62x2.00x0.80

sub-mcwa002_ses-2_task-bh_run-1_bold.nii.gz   int16 [128, 128,  68, 176] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-rest_run-1_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-rest_run-2_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-rest_run-3_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-rest_run-4_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-SDTD_run-1_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
sub-mcwa002_ses-2_task-SDTD_run-2_bold.nii.gz int16 [128, 128,  68, 201] 1.62x1.62x2.00x1.50
2

I tried several different options (6, 9, and 12 bold2t1w-dof, manual setting the origin in all images for better initial T1w-func alignment, longitudinal) to fix the T1w-func alignment. Nothing worked for all funcs. I uploaded the fmriprep report after manually setting the origins here: flirt_bbr. All funcs in session 2 fail to align with the T1w except for the single run in the bh task. I have not examined if there is a difference between it and the others, but I do not think there should be.

I did manage to edit the fmriprep code to use align_epi_anat.py from afni for T1w-func alignment rather than flirt from fsl. I built the docker image and ran it on the same data. All the T1w-func alignments succeeded when using align_epi_anat.py. I uploaded the fmriprep report for that pipeline here: align_epi_anat.py. My solution is hacky and I am on a time crunch so throughout the report it says FSL flirt was used to generate transformations from EPI-space to T1w-space, which is not true. The ending methods correctly says the BOLD reference was then co-registered to the T1w reference using align_epi_anat.py.