I have been very much impressed with the fMRIprep ease-of-use and the general community vibe so thank you to all the developers for creating such environment.
My question regards missing DICOM files in one of the clinical datasets I hope to pre-process with fMRIprep.
I have pre-intervention fMRI and T1 DICOMs scans for all patients but only post-intervention fMRI and T1 NIFTI scans for all patients. Given the scanning parameters are kept constant, would it be possible to re-use the JSON files from the pre-interventions scans for the post-intervention scans? If so, is there anything I have to specifically take care of (such as disabling slice-time correction)?
I am aware of the fMRIprep philosophy and the fact this is not the standard way of preparing datasets for fMRIprep, however, my experience with clinical dataset is that they are often incomplete or missing and some solution to such issues might be useful in general to the community. I hope this is a reasonable question for the developers.
Hi Jakub,
Thanks a lot for sharing this very relevant question to the list. I have the very same problem and I would love to know how people fix these type of issues.
Best wishes,
Diego
I would consider this a data management issue, rather than one specific to fMRIPrep. As long as the metadata doesn’t contradict the data (e.g., with SliceTiming mismatching the number of voxels in the slice-encoding axis), fMRIPrep will do its best to work with what you give it.
The nice thing about coming from DICOMs is that you can format in BIDS automatically and keep a record of the process, so others with access to the DICOMs can rerun to verify or update. However, BIDS is just the dataset format and does not prescribe how it must be generated.
So yes, it’s reasonable to reuse JSON metadata when you know it’s the same. I would just make a note of it in your README so that somebody who comes to your dataset later on can understand how the dataset was constructed and decide how confident they are in your choices.
Thank you very much for your valuable insights. It is reassuring.
As someone on the analysis side of fMRI, I will certainly expect DICOMs for any clinical experiment that is to be carried out in the future . Currently, in clinics this is not always the case… I presume dlozanosoldevilla has encountered similar barriers on that front.
My first thought was just to disable slicetime correction with --ignore slicetiming. But, if I understand you correctly, you suggest it should be possible to reuse the information about slice timing as well - after all the scans have the same TR and number of voxels in the slice-encoding direction. Does that seem reasonable?
I also like the README suggestion to allow anyone after me to see the choices made. And will certainly implemented.
If you know that the two images were generated by the same MR sequence, copying SliceTiming is reasonable. If this is just a best guess, then I would be hesitant to do so. It’s generally going to be better to skip STC than to correct with bad timing information, so I would be very conservative here.
That said, you can try to check whether your best guess SliceTiming is consistent with your NIfTI images. There is a slice_code field in the NIfTI header (https://nifti.nimh.nih.gov/pub/dist/src/niftilib/nifti1.h) that you can compare to the SliceTiming. I might also search for some tool that can give a confidence score that a data image actually was acquired according to a given slice sequence, though I don’t know if anybody’s written such a thing.
Upon inspection of the header information, the slice_code field is either “0” or “UNKNOWN”. So even though informative in general, in my case it is unfortunately not.
However, the two images are for a pre and post-treatment experimental scans. So, I don’t see how the MR sequence could be different. Would you agree?
It does seem likely that the sequences are the same, but I can’t advise you to reuse that metadata based only on that likelihood. I would suggest asking your MR techs or whoever actually ran the scanner; they should be able to look up the scan cards. Others involved in designing the study would also be worth consulting.