EPI to T1 registration

Hello Everyone,

I runned 1 patient trought fmriprep and spm. The result of fmriprep is on the left and spm on the right (screenshot). The quiston is why fmriprep doesn’t make angle correction for Nifti slices?

Which FMRIPREP output file are you plotting? What software are you using to plot the images?

@ChrisGorgolewski, I am plotting MNI<…>_preproc.nii through MRIcron.

Thanks. Could you share the HTML report (with figures which should be in a separate folder) as well as the NIFTI file?

1 Like

@ChrisGorgolewski, here are links for confounds and preproc, as i can see in confounds fmriprep write a head-motion estimation workflow result to confounds.

https://drive.google.com/open?id=1ELERM3pzZ63JlW17sfP8BmoS3YVRIWEj - preproc nii
https://drive.google.com/open?id=1hfjaIXytZv4U0aUIxy-6aU6JdoG44Yb2 - confounds

Figures - https://drive.google.com/drive/folders/1qyLL8xLlC_wB_mhMcIWhCurxgABvZwpj?usp=sharing
html - https://drive.google.com/file/d/1r0XSMsXc95OY-ZCnPhNcmBkAynq6q55o/view?usp=sharing

Thanks for sharing.

The issue you are experiencing has nothing to do with motion correction. The problem is with EPI to T1 registration (as shown in the animation in the HTML report). Due to the lesion in this particular participant the coregistration did not do well.

Could you try rerunning FMRIPREP without the --debug flag (it turns on suboptimal, but fast coregistration options which we use for testing/debugging) and with the --force-bbr flag? You can check the HTML report if the coregistration improved.

Hello @ChrisGorgolewski, Thanks for your reply and attention to the issue. I just launch this patient with the flag you proposed. I have another patient wthout any lesions - from control group. But the aignment if fmri doesn’t work as well. Here is the link https://drive.google.com/file/d/1eBT36eBTsOepdb1CJsHQgEePXwQwczYR/view?usp=sharing. Any help would be appreciated.

Indeed looks problematic. I wonder if there’s something wrong with the NIFTI headers (we recommend using https://github.com/rordenlab/dcm2niix to create NIFTI from dicom). Would you mind sharing input data for one participant?

Another thing to try is to not to disable recon all - even if you don’t need the surface, it leads to better coregistration.

Another idea would be using the fieldmaps or enable --use-syn-sdc. Although I agree that bbregister has more chances of succeeding than flirt (hence, removing the --fs-no-reconall).

My preprocessing pipeline include heudiconv conversion from RAW dicoms to BIDS structured dataset.