I’ve begun reorientating my data to RAS+ using nibabel’s as_closest_canonical. This has led me to two questions I was hoping to have clarified.
1). Am I correct in assuming that simply reorientating the data doesn’t cause it to become a derivative of the original (by BIDS standards)?
2). Would the PED metadata need to be adjusted? This is in relation to spin-echo fmap acquisitions. Typically the PED of my fieldmaps are in the Anterior-Posterior plane (j- or j). My functional data is normally in RPI, so if I reorient my data to RAS+, the PED should be switched from j- to j? I could see this becoming more difficult if the 2nd axis of the data isn’t A or P
What is your motivation for re-orienting your data to RAS+. Note that when converting DICOM images tools like dcm2niix and dicm2nii will automatically losslessly reorient your data to match canonical space if it was a 3D non-EPI acquisition. However, there is a reason these tools do not apply this spatial transform to the other data. Specifically, we want to maintain the slice order (for slice timing) phase encoding parameters, etc. While in theory the NIfTI header does allow one to describe the slice dimension as something other than K, none of the slice timing tools I know of can handle such data. I also think preserving slice order is helpful for understanding artifacts (multi-band aliasing, phase wrap, etc).
My sense is if you do this, you will be using many tools in ways they were not intended and where few people have validated the results. The opportunity for unintended consequences and edge cases is high. Since the scanner SForm and QForm store the spatial orientation, I am not sure what benefit is derived from this approach.
1 Like
I had hoped to align the orientation to fMRIPrep; however, given the issue of preserving slice-order, I’ll avoid this with acquisitions that aren’t 3D non-EPI.