BBR registration algorithm performed poorly on 7T fMRI data

Hi fmriprep experts,

I have a question about using fmriprep to preprocess 7T fMRI data. We acquired high spatio-temporal resolution multi-band fMRI data on Siemens Terra 7T, and tested the performance of fmriprep on this data. The results show that when performing cortical reconstruction of fMRI images, the output CIFTI data is incorrect because BBR cannot accurately align BOLD to T1W images on both 6dof and 12dof. But fmriprep only provides BBR registration algorithm for fmri to native T1W space. In this case, how should we use fmriprep to process 7T fMRI data with large distortion? Thanks for your help.


These are susceptibility artifacts. I don’t know how well fMRIPrep will perform on 7T data, but without susceptibility distortion correction (which will be more significant in 7T acquisitions) there will be significant anatomical misalignments. You can try --use-syn-sdc to see how well that works.

You need to correct the distortion first with either:

  1. Topup using a B0 map or reverse phase encode epi pairs. (note technically this works better if the AP PA pairs are matched spin echo epi images with matched bandwidth, echo time, resolution, and parallel imaging parameters.
  2. You can co-register with ANTs to the T1map (from mp2rage) using antsRegistrationSyn (this does a pretty good job but would be experimental). I have modified my version of antsRegistraionSyn to use mutual information as cost function. Theoretically I one should be able to then do fmri to uniden but it never works as well as fmri to t1map

Thank you very much for your reply, we are testing the performance of topup including B0 field and SpinEcho_EPI with AP and PA phase encoding direction. But I believe that larger distortions cannot be completely corrected by the above method. Can you share your modified antsRegistrationSyn code, we want to test how this works before fmriprep preprocessing. Thank you so much.


Hi @Guoyuan , actually the PEPOLAR method with SpinEcho EPI images with AP and PA should work best for susceptibility correction. But it is true that if the distorsion are too big, especially in the compressed area, SDC will be difficult. You can mitigate the distortion on the acquisition side by minimizing the echo spacing, reducing the FOV in the phase encoding direction and using parallel imaging for instance.

I don’t know how this would work with 7T data, but we recently tested this tool which is very good for SDC if you don’t have any fieldmap images to correct your bold images:

We discussed its implementation with fmriprep processing here:

Apologies, I missed the request here for the modified antsRegSyn. If you PM me I can share the code. But basically I just modified the antsRegistrationSynQuick to use a MI cost function. However, I agree the distortions are never perfectly corrected. IMO the phase encode PA always seems to work better than AP. So I believe in getting the fMRI time series in PA, but work with your radiographers to make sure the PA distorted frontal regions don’t wrap!