Hi all, I am using fmriprep with multi-echo data and was wondering how to prepare the fmap folder for the multi echo data? We have acquired the multiecho sequence in AP and PA directions, so we’d have 4 AP echo trains and 4 PA echo trains. Outside fmriprep we use topup on the means of the echo trains and then apply the unwarp to each echo train separately. How does fmriprep handle this and how do we prepare the bids folders for it to recognize the AP / PA maps?
Bw,
Peter
Hi @peterzhukovsky and welcome to Neurostars!
Does this help?: Multi-echo, fmriprep 21.0.2 distortion correction, the -t task selection parameter, and tedana - #12 by jrdalenberg
Best,
Steven
1 Like
The short answer is that we’ll use the first echo to calculate the fieldmap. The same fieldmap is then applied to all echos.
As to how to prepare the folders, if you follow the BIDS standard, we should be able to find them. (Note that for IntendedFor, we do not yet support BIDS-URIs, so subject-relative paths are required.) If you run into an issue, please update with what you tried.
2 Likes
Awesome, thanks Chris, so it’s only the first echo that we include in the fmap folder, right? Will try it out next
If multiple echos are in the fmap folder, I believe we will select the shortest echo. If you would like to use an alternative echo, you may place that in the fmap folder without the shortest. And yes, if you want to just be unambiguous (or if I’m wrong and fMRIPrep gets confused by multi-echo fieldmaps), you can place the first echo in the fmap folder.
It would be good to get some clarification more with regard to if the “fieldmap” images are gradient-echo (GE) or spin-echo (SE) based.
The default for susceptibility distortion corretion for fmriprep is to use topup. When you then follow the thread with regard to topup (which was developed for correction DWI data) at FSL, topup was developed for use with spin-echo-based epi (and not ge-based EPI input images) for the estimation of the distortion correction that can then be applied to DWI EPI data (and by extension to GE-EPI data as the community is doing with time-series BOLD data).
It seems that some folks have been erroneously using reversed phase encode GE-EPI to calculate the correction when then correction should be calculated using a AP/PA SE-EPI pair.
See this thread, which includes Jesper Andersson along with his paper:
https://www.jiscmail.ac.uk/cgi-bin/wa-jisc.exe?A2=ind2212&L=FSL&D=0&P=18631
1 Like
I’d like to make one addition. While I know this thread is specific to top-up and its usage in fmriprep I think it is worth mentioning that this has been explicitly tested at 7T (where field homogeneity is an even greater challenge) and it seems that GE based approaches work quite well: https://onlinelibrary.wiley.com/doi/10.1002/hbm.25540. This is the case even when using topup rather than 3dQwarp (as is done in afni for “blip-up/down” correction). In fact, GE data seemed to outperform SE and fieldmaps in this work (with the exception of the vmpfc).
Of courts, this is not to say a single study is the final word on the subject, but just to add a bit more details for folks even though GE is not the intended use case.
2 Likes
There is indeed a controversy here, when you can find papers supporting one conclusion and others the opposite: Are GE-EPI images better than SE-EPI for correcting BOLD (GE-EPI images), or the opposite?
As always in the MRI field, it is a multi-components problems, since many bricks are being put together between the acquisition and the preprocessing steps and the analysis.Depending of your choice for each brick, the result and the conclusions drawn may be different.
In the discussion cited above in the FSL forum, involving two well known specialists, Jesper Andersson being the developper of FSL topup, it is mentioned that in the paper you cited ( Schallmo et al. Assessing Methods for Geometric Distortion Compensation in 7 T Gradient Echo Functional MRI Data. Hum Brain Mapp 2021, 42 (13), 4205–4223. https://doi.org/10.1002/hbm.25540), the authors were mistaken when they drew the conclusions of their paper. Indeed the issue is that GRE-EPI has both distortion and dropout, which cannot be distinguished by topup when estimated the fieldmap, resulting in a biased fieldmap estimation.
This point was further developed in this presentation (even if here topup was not the tool use for fieldmap estimation):
https://cds.ismrm.org/protected/18MProceedings/PDFfiles/2334.html
To add to the controversy, below is another paper at 7T that claims the opposite: SE-EPI giving better results than GE-EPI for susceptibility distorsion correction 
Clarke, W. T.; Mougin, O.; Driver, I. D.; Rua, C.; Morgan, A. T.; Asghar, M.; Clare, S.; Francis, S.; Wise, R. G.; Rodgers, C. T.; Carpenter, A.; Muir, K.; Bowtell, R. Multi-Site Harmonization of 7 Tesla MRI Neuroimaging Protocols. NeuroImage 2019, 116335. https://doi.org/10.1016/j.neuroimage.2019.116335.
In this work we have carried out EPI distortion correction using FSL’s topup routine, with SE-EPI reversed phase-encode blip images providing the estimate of the unwarping to apply. This is done on the basis of previous work (Driver et al., 2018). A simple analysis of the data collected as part of this study replicated the results of that previous work, finding that correction based on field-maps or on topup using SE-EPI measurements provide superior distortion correction to topup correction based on GRE-EPI measurements.
Thanks for sharing that, good to have more details. One complication of course is the use of different metrics (another “brick” thrown on the pile…), but it does seem clear from this that, if you are focused on regions with large dropout, then taking the time to collect 2 spin-echo scans is probably valuable.