Hi,
See my responses below:
> After more reading (particularly http://www.nitrc.org/forum/forum.php?thread_id=3785&forum_id=1456) I think I have a few questions regarding the optimal pipeline to incorporate FUGUE/epi_reg along with slice timing, physio and volume motion correction for resting state data.
>
> A) Having a siemens interleaved EPI sequence with physio collected I want to create a pipeline to incorporate physio correction, slice time , motion as well as Fugue for resting state data. FUGUE is used to correct for B0 inhomogeneity. The question I have is since the head moves a bit from volume 1 to volume 280 do I still run fugue first or do I need to run some sort of motion correction prior to that fugue applies the distortion correction to the appropriate voxels (ie in case of shift)? On one hand the map is collected prior to the first volume so should I motion correct all EPI volumes to the first timepoint or can I use the last (my ref frame for motion correction) in order to ensure the map matches the brain position. On the other hand the brains were in reality shifted when the EPI was taken so would the field map correction be incorrect since we adjust the position through volume correction, which does not represent the voxels true position in the magnetic field when the scan was acquired (ie run FUGUE PRIOR to motion correction). My guess is that we should motion correct first and then run FUGUE and I can align to the last time point, but please confirm.
I've partly answered this in another reply, but essentially we ignore the second-order effects of varying distortion with motion since these changes are much smaller than the static distortions (since the head motion is really very small typically). We do correct for any differences between the position/orientation of the fieldmap acquisition and the EPI by registering the fieldmap to the structural with epi_reg. I would strongly recommend that you use epi_reg for dealing with the fieldmaps rather than fugue directly.
> B) Aside from just motion correction I am trying to have the optimal pipeline to incorporate fugue with standard slice timing, motion correction etc.
> The following passage is from the link I attached above:
>
> "slice timing correction will make physiologic noise in your data unrecoverable by any means. The slice time correction resamples the data acquired at X msec after the start of the volume and assumes that everything is Nyquist-sampled (ie. no freq higher than 1/(2*TR) Hz). Any frequency lower than the Nyquist is correctly resampled as if it occurred at the start of the volume. However, any frequency in the data that is higher than the Nyquist gets smeared all over the Fourier spectrum randomly depending on where it happened to bin. Most (possibly anything slower than 200-400msec TR, due to higher harmonics of cardiac) acquisitions do not have a volumetric sample rate fast enough to capture physiologic noise."
This is true in that slice-timing-correction performs temporal interpolation that causes such problems.
> For some earlier datasets we do not have physio so I was going to use seed-based regression of CSF/Saggital sinus which captures physio signal and use as a regressor. Based on the above does this mean that I would not be able to use this method due to the fact that slice-timing was performed? Or would this still be a valid approach?
I would not perform slice-timing-correction prior to removing physio signals. You should also try and extract slice-based signals if possible, as the nature of the signal can vary a lot with the slice.
> C) For data that I do have physio collected for, does the order below make sense to incorporate fugue, or any modifications you suggest? :
>
> 1) Run some sort of physio correction using physio data
> 2) Run slice time correction
> 3) Discard initial timepoints since steady state was not reached yet.
> 4) Run motion correction to the last time point (since first few time points will be discarded we align to last timepoint)
> 5) Align the magnitude/fieldmap image to the last time point and apply then apply FUGUE
There is no perfect pipeline here, as these effects all interact but the tools currently only deal with things separately. You will therefore find that opinions differ on the "best" way to do things like this. Partly it will depend on the nature of the subjects (e.g. whether they move a lot or very little). Our general advice is that motion (spatial displacement) induces much larger changes in the signal than slice timing (temporal displacement) and so we would rather correct for as much of the motion-related signal as possible and live with some unrecoverable temporal-displacement-related signals. This involves doing motion correction before slice timing correction. However, if motion was very small and exact timing was a big issue (such as in short event-related fMRI experiments) then an argument can be made the other way around. In your case I would stick with doing motion correction before slice-timing-correction. Also, as slice-timing-correction negatively impacts on the ability to deal with physio noise, then I would do physio noise correction before slice-timing-correction too. The issue of applying distortion correction is largely independent, as we ignore the temporally-varying aspect of it, so this can go last.
Therefore I would recommend the order of: deleting non-steady-state timepoints (which is often done automatically by the scanner anyway); motion correction; physio-noise-correction; slice-timing-correction; distortion correction (with epi_reg). Although we often omit the slice-timing-correction step as it often makes little difference, particularly with shorter TR scans. Note that this order is an approximation, and in an ideal world these corrections would all be rolled into one, but you should be able to get a reasonable result using this order.
All the best,
Mark
> Sorry for the long email!!!!
>
> Thanks,
> Ajay
|