Single band reference scan settings

I am adding a single band reference scan to be processed in fmriprep along with MB fMRI data. What are the suggested setting of this scan? From what I’ve read, this is used to have a scan with better contrast than the functional scans, so:
1.I assume this is mainly an issue of TR, so should the single band reference scan use the minimum single-package TR (> 2 sec, currently the MB TR is 1 sec)?
2. Or is the main point here to have an MB-artifact-free reference scan, in which case TR isn’t a concern?
3. Should this include multiple dynamics (or averages) to improve the SNR for this scan?

The rationale for using the SBRef is described in Glasser et al. used for full tissue contrast. The very short TRs associated with multi-band will reduce the contrast of the tissue boundaries. Presumably, the very first volume of the multi-band time series has a long effective TR (with the corresponding T1 signal), but this is typically not reconstructed. As you note, the single-band reference is also free from the aliasing inherent in multi-band.

So the goal is for the SBref to be identical to the multi-band time series except without multi-band on. This will inherently require a long TR. Make sure this image is acquired with the same angulation as the multi-band data, to ensure the same shim. The question of whether one wants to average multiple single band images is a good one. Presumably, the first volume has the most T1 and best tissue contrast, but signal averaging might be a benefit for longer sequences. Regardless, I would try to make sure the scanner does not discard the first volume, as here the goal is to maximize T1-effects, not ignore the initial volumes where the signal is not saturated.



A follow-up question on this one. Our functional scans are run with both PA and AP PE directions, but not our SB-ref (which we just added in without dummy scans and 6 dynamics - our plan was to average these across the 6 dynamics, with motion correction if needed). From what I can tell, the SB-ref is being used to improve on the field distortion correction. If that’s the case, should we be running the SB-ref with both PE directions as well? Not sure how the software handles this, but don’t have the time to dig into it. Any help would be much appreciated.


Mark W

Diffusion scans use Spin-Echo sequences to remove susceptibility induced signal loss. EPI artifacts are seen as spatial distortion but not signal loss.

fMRI scans use Gradient-Echo sequences that retain susceptibility induced signal loss (since the BOLD effect is a susceptibility contrast). EPI artifacts are seen as BOTH spatial distortion and signal loss.

For DWI SE sequences, you can acquire AP and PA sequences. Both show spatial distortion of equal magnitude, but opposite polarity, This allows tools like FSL’s Topup to undistort the image.

For fMRI GE sequences, you can not use Topup. Regions of inhomogeneity are not merely moved, they become invisible. Therefore, it is hard to calculate the undistorted image,

For fMRI GE sequences you have two options:

  1. Acquire a series with a GE FieldMap and use a tool like FSL Fugue.
  2. Acquire two series of SE images, where one matches the phase encoding direction of the GE. fMRI and the other reverses the phase encoding direction. Use a tool like Topup to estimate the distortion of the SE sequences and applytopup to apply the estimate to the GE sequence.

In both cases, it is critical that the resolution, angulation and other parameters match the GE fMRI series, so the same shim is used.

What is the vendor of your MRI (GE, Philips, Siemens)? Since you mention multi-band, I assume you have a modern 3T instrument with coils designed for multi-band. A good place to start might be using the ABCD sequences on your instrument. Also, assuming you are at a research center, the vendor has a research collaboration manager who is responsible to helping you. These individuals tend to work like lawyers on retainer, if they do not hear from you they assume you are happy with their services. These individuals have exquisite and proprietary knowledge of your instrument. I urge you to work with them to optimize your protocols for your instrument.

Thanks, Chris. We have field maps, and can use these for field correction. Although, I’m still not sure I completely agree with your comment. Our protocol is modelled after the HCP protocol, which also collects PA and AP scans for all of their functional data.
You can’t reconstitute regions of signal loss, even with a field map. The only issue which you mentioned is the ability to calculate the undistorted image, which will fail in regions of no signal but I don’t care about those regions - they are “lost” in any event. The issue is whether or not the undistorted image in regions with signal will be accurately reconstructed, in close proximity of these signal loss regions. We have a few data sets, so can see how it performs.

It would appear that fmriprep is doing “something” with respect to susceptibility distortion correction, this is pretty clear in the output html (SDC before/after scans), although it does note “susceptibility distortion correction - none” in the list of corrections done, which was a bit puzzling. Any insight into what’s going on there would be most appreciated.

Finally, we do have a Philips clinical scientist “on campus” (whatever you would call on-campus these days!). I can ask him to see if he has any thoughts on the subject.

Thanks, Mark

Note that the HCP uses spin-echo sequences (SpinEchoFieldMap_AP, SpinEchoFieldMap_PA) to correct the spatial distortion of their gradient-echo T2* images (rfMRI_REST_AP, rfMRI_REST_PA).
The spin-echo sequence preserves signal pile-up/dilution caused by local stretching/compression. This allows you to map the compression/dilation. You can apply these maps to the gradient-echo images to recover the shape, but you can not recover the signal lost with the gradient echo.
On the other hand, the gradient echo does not preserve the pile-up/dilution, so you can not map the spatial distortion.
The FSL Topup web page suggests this article for comparison of the undistortion methods.

Thanks! All makes sense.