SDC-pepolar in fmriprep: rationale for limiting displacement estimation direction

[markushs]

Hi!

As the title suggests, I’m curious about the reason for limiting the estimation of susceptibility distortions to the phase encoding direction (as described here https://fmriprep.readthedocs.io/en/stable/api/index.html#phase-encoding-polarity-pepolar-techniques).

In my experience (from inspecting fieldmaps estimated through FSL topup) the directions of susceptibility distortions in a region seem to be a product of interactions between phase encoding direction and subject anatomy (e.g. volume of air-filled cavities). For example, with PE-direction A–>P the frontal lobes will mostly be compressed, but some parts will also show slight stretching in the anterior direction. (see https://github.com/poldracklab/fmriprep/issues/1547 for an extreme example).

Are there any obvious negative effects of allowing the displacement to be estimated in both directions of the PE axis?

And a related question: does anyone know if topup estimated fieldmaps (in Hz) can be used as input to the “Direct B0 mapping”-workflow for susceptibility distortion correction? (i.e.: https://fmriprep.readthedocs.io/en/stable/api/index.html#direct-b0-mapping-sequences)
If so, maybe this could be a workaround (unless, of course, displacement estimation is limited also in this workflow).

Thanks!
Markus

[oesteban]

Are there any obvious negative effects of allowing the displacement to be estimated in both directions of the PE axis?

By restricting the deformation to happen along the PE axis we mean that no deformations can happen along the other axes. It does not mean that all displacement vectors point to the same direction (which I what I think you are suggesting here). Indeed, displacements happen in both directions of the PE axis.

Conversely, if you are suggesting to estimate the distortion using the two possible polarities of the encoding blips, then I would argue against the idea because it opens the door to hide implementation and/or metadata problems away. It could be pretty dangerous.

And a related question: does anyone know if topup estimated fieldmaps (in Hz) can be used as input to the “Direct B0 mapping”-workflow for susceptibility distortion correction? (i.e.: https://fmriprep.readthedocs.io/en/stable/api/index.html#direct-b0-mapping-sequences )
If so, maybe this could be a workaround (unless, of course, displacement estimation is limited also in this workflow).

The current implementation of fMRIPrep actually does the same with the outputs of both alternatives (i.e. it is equivalent). Displacement estimation is limited in the same sense as for the PEPOLAR option.