7T resting-state fMRI data in CIFTI format showed bad FC maps

We used fMRIprep to process our 7T resting-state fMRI data(CMR 7T), but the performance of vertex-based functional connectivity in cerebral cortex was not very satisfactory. I have checked the alignment of the cortical extraction, and there is no big problem here. But the vertex-based functional connection shown in the figure below did not meet our expectations compared with the HCP 7T data. We think it may be the step of cortical projection got something wrong, but it‘s still not clear exactly what the problem is. Our 7T fMRI protocols including TR=2440ms, TE=20ms, 1x1x1mm isotropic, with 250TRs. Thanks for your help.

Best regards,

Hi @Guoyuan what command did you use for fmriprep and what version?

Hi Steven,

The shell script I have run is shown below, which should be the latest version of fmriprep.

mkdir -p .../output/tmp09
docker run -ti --rm \
    -v .../bids_7T:/data:ro \
    -v .../output_7T:/out \
    -v $HOME/license.txt:/opt/freesurfer/license.txt \
    -v .../output/tmp09:/tmp \
    -v .../out/sourcedata/freesurfer:/freesurfer_path \
    -w /tmp \
    nipreps/fmriprep:latest \
    /data /out/out \
    participant \
    --participant_label 001 \
    --ignore fieldmaps \
    --fs-license /opt/freesurfer/license.txt \
    --cifti-output 91k \
    --n_cpus 12 \
    --bold2t1w-dof 12 \
    --use-syn-sdc \
    --fs-subjects-dir /freesurfer_path \
    --fs-license-file /opt/freesurfer/license.txt


Can you try the --project-goodvoxels flag?

I’ll test the effect of this right away.

Hi Steven,

We have tested two runs with –project-goodvoxels flag and the vertex-based FC map as shown below, which did not improve the results very well. Are there other parameters that can be further optimized?


How are you denoising / post processing your data?

Hi Steven

We tried further post-processing, using the GLM model to regress head movement and physiological signals, and performed 4mm smoothing on cortical surface mesh. The results became a bit better as shown below. However, we found resting-state fMRI with 1mm isotropic resolution is not as good as that of traditional 2mm isotropic resolution when performed fMRIprep to generate 91k cortical surface. But we are not sure why. Do you think this is a problem of SNR or fMRIprep is not suitable for high spatial resolution fMRI data?


By the way, we found that high-resolution fMRI had higher DVARS compared to conventional-resolution data. The DVARS of our results has reached an average value above 90, but the head movement curve of the SPM analysis shows that our head movement is not very large.

Try --cifti-output 170k instead. 91k is about the same as 2mm isotropic, so the 170k resolution may match your data better.

What confounds specifically did you use? Some people like to use derivatives and squared expansions of the head motion parameters.