The SDC of fMRIPrep v21-22 seems not robust!

Hi, there.
I used the fmriprep to do proprocessing recently.
After checking the report, I found many subjects have wrong SDC correct.(The brain becomes not like the brain.)
I found some people have same problem in NeuroStars and the three updates of fmriprep fix the SDC-related issues.
So I update to v22 version. Unfortunately, the fmriprep v22 seems have same problem.
I used 8 subjects to test and half of them have different bad correction. Even some of runs have good correction in the previous preprocessing.
As we can see:
The filed map for two proprocessing are same.


But the second correction with v22 have bad correction result.(Both for same subjects)


Distortion in prefrontal(z=-46) and temporal lobe(x=-42)

Would you have a suggestion for this? I would appreciate it very much!

The more information in this page:

Thank you for reporting this.
Could you please give us additional details on your acquisition settings and preprocessing steps?

  • Which FMRIPREP version do you use before testing v22?
  • Which data do you give to FMRIPREP for SDC: are they phase difference fieldmap images?
  • Did you check that the shim settings were the same between the acquisition of the fieldmap images and the functional images?
  • I noticed that in the functional images, the head is quite tilted to the front, do you align your axial slices on the AC-PC line or on another axis?

Thank you for your reply.
Of course. My pleasure.

  • v21
  • Yes, It is their phase difference field map images.
  • I am sorry. What do your mean “shim setting”?
  • Yes. First, we align the axial slices to AC-PC and then tilt the axial slices 30 degree for maxmaize the vmpfc signal which we learned from some previous study.

Sorry for not being clear. Are the shim values the same between your fieldmap acquisition and your functional images acquisition? You can see the shim values applied during the acquisition of each acquisition in the json file under the field: “ShimSetting” (at least for Siemens DICOM images converted by dcm2niix).

I have checked the shim setting for our subjects. Some of them indeed have subtle difference. But the shim settings are same for the three subjects who I showed.

Thank you again for reporting this. These discrepancies between the correction offered by the two fmriprep versions look very localized to highly distorted area and I feel it won’t affect the general results, which a very important point. For instance the mask displayed as a read line does not look very different between the two versions. Don’t you have the same feeling?

Otherwise I have another suggestion: did you look at the images themselves and not only at the report and did you see this strange behavior on the NIFTI images themselves? I know that sometimes for creating figures for the visual reports fmriprep uses different functions/interpolations for a better display which are not used for the preprocessed images themselves.

OK, I will check it.