I am working on setting up an acquisition protocol for fMRI and structural MRI data and have a couple of questions:
What is the best way to acquire field maps for susceptibility distortion correction of fMRI data? We’re using a Siemens scanner and collecting 2 magnitude + 1 phase difference field maps. Should the FOV and voxel resolution match those of the main fMRI data?
In general, are slice gaps to be avoided? My impression is that if the acquisition is interleaved, slice gaps should be eliminated as much as possible, but I’m a newbie so I might be missing something. And what if the acquisition is ascending/descending (as for the structural scans)? Would slice gaps be advisable (to avoid cross-talk)?
The method for acquiring a Siemens product field map (%SiemensSeq%\\gre_field_mapping) and setting saving the phase and magnitude images is well established - here is the dcm2niix validation dataset. I would certainly have the image match the FOV, number of slices, resolution match the fMRI images. On your console, set up the fMRI and field map to copy references to use the same slice positioning and angulation. This should also ensure the scans use the same shim setting.
Adding slice gaps reduces interference between neighboring slices for sequential acquisitions, and reduces spin history motion artifacts for interleaved acquisitions. The correct choice depends on your sequence, equipment and population. For stationary objects like water bottles, an interleaved sequence with no gap makes sense, but most real world situations are more complicated. A study of children who exhibit more head movements might consider a larger gap and sequential slices. Older scanners with analog RF transmission show more ripple at the edges of slice selection, so benefit from a gap. All else being equal (number of slices, field of view in slice direction), a gap means less hydrogen being sampled, so less SNR. Modern multi-band sequences have unintuitive slice sampling patterns to balance the scans, so conventional ideas of sequential and interleaved do not apply. I am sure more informed MRI physicists can provide more details, but my notes are here and I would encourage you to acquire tSNR maps and look at them across the slice direction. You may want to share the sequence PDF or a BIDS JSON file for your sequence here for more comments. In general, if you have a multi-band capable Siemens scanner with a 32-channel head coil, the ABCD sequences look pretty well thought out (with the exception that the DWI scans are inherently imbalanced as they have 60 slices and mb=6),
Many thanks for the information @Chris_Rorden - it really helped. Below I am including links to the PDF sequences we’re using. For fMRI, the slices were either acquired in interleaved fashion (protocol 2) or sequentially, in descending order (protocol 3). We used the copy references setting you suggested but for the resolutions of the EPI scans and gradient field maps to match we had to allow asymmetric echo. We set the distance factor to 0% for both the EPI scans and the gradient field maps when using interleaved acquisition, and 25% (again, for both EPI and gradient field maps) when collecting the slices in descending mode. In the latter case though, it appears that the gradient field maps were still acquired in interleaved fashion. I am not sure why this is so.
The sequences also include the parameter specifications for MPRAGE. The main difference between protocols 2 and 3 is the values of the following parameters: Orientation, Phase encoding direction, FoV phase, TR, TI, flip angle, and table position.
The slice order for the field maps does not need to match the fMRI scan. The controversy regarding interleaved and sequential slicing really focuses on 4D datasets. My sense is that interleaved acquisition of field maps makes sense unless you are working with a population prone to head movements.
In general, the protocols look fine to my eyes, but I think there are others with far more expertise than me. My one comment would be that the 0% gap EPI sequence with 45 slices and 2.5mm between slice centers means that the whole slab is just 112.5mm thick. This is a bit tight for most adults, and can also limit registration quality. Following the FSL FEAT defaults I usually try to have at least 120mm in the slice direction for the EPI data.
Repetition was intentional. Multiband slice timing patterns described by the paragraph'Note that the number of slices acquired varies with multiband.... while the whole page are the notes I compiled talking to several MR experts (Ben Inglis, Paul Morgan, Jens Jensen). I have similar pages of notes for other modalities largely to describe to users at our center the trade offs we made to select specific sequences.
Thanks again for your reply @Chris_Rorden. We can set the number of slices to 51 so that the number of slice stacks is still odd (multiband acceleration = 3).
Do you think a 0% gap is okay for the EPI sequence and gradient field maps when using interleaved acquisition?
@Chris_Rorden I have one more question, please: what slice gap is it advisable to use for the EPI sequence and gradient field maps when using interleaved acquisition?
@cfarr adding a slice gap reduces interference for sequential acquisitions, and reduces motion-based spin history issues for interleaved acquisitions. On the other hand, it reduces the overall amount of hydrogen sampled which hurts SNR. More modern MRI systems have better RF so the slice selection is more accurate. Common in our field is 0-25%, with studies of children (where head motion is common) going toward the high end of this range. I think this is a known tradeoff and I do not have a clear value to give you. If I were you, I would contact your center’s Siemens Research Collaboration Manager - they have a lot of inside information regarding the hardware capabilities of your system. Your RCM is an invaluable resource in getting the most out of your instrument, and they have a lot of knowledge that is not publicly disclosed.
Hi @Chris_Rorden ! First of all, many thanks for your insight and helpful comments here and apologies for reopening an old thread. If preferred I can create a new thread or contact you in another way.
My question - duo to an error in the settings, a handful of our siemens phasediff/magnitude fmaps were acquired with the FOV and orientation of the T1 anatomical image rather those of the EPI. Is there a way to save this (e.g. by applying SDC after alignment of EPI to T1 or reverse aligning the fmap to the EPI)?
This question is outside my expertise. In general, for undistorting gradient echo fMRI/resting state data where signal is lost and moved I acquire two phase and two magnitude images suitable for use with FSL FUGUE - see example data here. For spin echo diffusion data where the primary impact is moving signal, I acquire pepolar images suitable for FSL TOPUP/Eddy. In both cases, I try to use the same slice geometry, centers and angulation to ensure that the scanner is not re-shimmed between series. I really do not have insight into how robust these methods are when these properties are not matched.
