Mrdegibbs in partial Fourier

Hi experts,

I was wondering how is your experience with mrdegibbs in partial Fourier data.
In original paper https://www.ncbi.nlm.nih.gov/pubmed/26745823 they use partial Fourier 6/8 to demonstrate that degibbs in this type of acquisition produces spurious ringing after correction.
I have also 6/8 partial Fourier data (AP phase encoding direction) and actually mrdegibbs seems to me to work very well. The ringing is removed very well and I do not see any new spurious ringing in AP direction after correction. See mean diffusivity data after denoising, before mrdegibbs:


After mrdegibbs the big ringing at cursor position is effectively removed:

Actually, I do not know what they mean by new ringing after correction with partial Fourier data showed in fig. 6 in the original paper. I see in the figure some hypointensities around ventricles in the partial Fourier data, but they seem to be present also in their original data before correction. I see similar hypointensities around ventricles in my data (see cursor), but they are present also in the data before mrdegibbs, so it is no new ringing.
Before mrdegibbs:

after mrdegibbs:

Could you please comment on?

Secondly, do you think that it is possible to implement mrdegibbs to perform unringing only in the direction with full k-space coverage (in my case LR)?

Thank you in advance,

Antonin

We had the exact same experience – and I also don’t think there was any.evidence in the original paper of the method introducing artefacts that weren’t already there. So I asked Elias Kellner about this a while back, and he also thinks it’s perfectly fine to use. Here are selected parts of his thoughts on the issue (I hope he doesn’t mind me sharing them here…):

The unringer does not introduce or amplify additional artifacts. If there is “true” ringing, it removes it; to the rest, only a slight smoothing is introduced (but it does never “oversmooth”)

So in general, we fully support the statement that it is safe to apply it to any MR image as a black box.

But again, I think that additional smoothing in PF direction is not too strong (as you also observed), and for now, I would suggest to write in the help something like:

Note that this method is designed to work on images acquired with full k-space coverage. Running this method on partial Fourier (‘half-scan’) data will not fully remove the artifact, and you will observe residuals of the original artifact in the partial-fourier direction, but it is safe and worthwhile to apply it also in this case.

I guess it’s about time we updated the help text as suggested… :blush:

1 Like

Secondly, do you think that it is possible to implement mrdegibbs to perform unringing only in the direction with full k-space coverage (in my case LR)?

If there were sufficient justification to do so, it would certainly be possible: core code for such functionality is already there within the command, it would only require altering the main executing function to permit 1D operation only when the input to the -axes option has one value only. But given it’s generally advocated to run in 2D mode regardless, it might be hard to justify putting the time into it.