Education, tips and tricks to help you conduct better fMRI experiments.
Sure, you can try to fix it during data processing, but you're usually better off fixing the acquisition!

Saturday, July 6, 2013

12-channel versus 32-channel head coils for fMRI

At last month's Human Brain Mapping conference in Seattle, a poster by Harvard scientists Stephanie McMains and Ross Mair (poster 3412) showed yet more evidence that the benefits of a 32-channel coil for fMRI at 3 T aren't immediately obvious. Previous work by Kaza, Klose and Lotze in 2011 (doi: 10.1002/jmri.22614) had suggested that the benefits were regional, with cortical areas benefiting from the additional signal-to-noise ratio (SNR) whereas the standard 12-channel coil was superior for fMRI of deeper structures such as thalamus and cerebellum. The latest work by McMains and Mair confirms an earlier report from Li, Wang and Wang (ISMRM 17th Annual Meeting, 2009. Abstract #1614) that showed spatial resolution also affects the benefit, if any. In a nutshell, if a typical voxel resolution of 3 mm is used then the 32-channel coil provides no benefit over a 12-channel coil. The 32-channel coil was best only when resolution was pushed to 2 mm, thereby pushing the SNR down towards the thermal noise limit, or when using high acceleration, e.g. GRAPPA with acceleration, R > 2.

What's going on? In the first instance we need to think about the regimes that limit fMRI at different spatial resolutions. In the absence of subject motion and physiologic noise, the SNR of an EPI voxel will tend towards a thermal noise-limiting regime as it gets smaller. Let's assume a fairly typical SNR of 60 for a voxel that has dimensions 3.5x3.5x3.5 mm^3, as detected by a 12-channel head coil at 3 T. If we shrink the voxel to 3x3x3 mm^3 the SNR will decrease by ~27/43, to about 38, while if we shrink to 2x2x2 mm^3 the SNR will decrease to about 11. (Here I am assuming that all factors affecting N are invariant to resolution while S scales with voxel volume, which is sufficient for this discussion.) If we decrease the voxels to 1.5x1.5x1.5 mm^3 the SNR decreases to below five. The SNR is barely above one if we push all the way to 1x1x1 mm^3 resolution, which is why you don't often see fMRI resolution better than 2 mm at 3 T. Thus, if high spatial resolution is the goal then one needs to boost the SNR well beyond what we started of with to achieve a reasonable image. Hence the move to larger phased-array receive coils.

So that's the situation when the thermal noise is limiting. This is generally the case for anatomical MRI, but does it apply to fMRI? If something else is limiting - either physiologic noise or subject motion - then increasing the raw SNR may not help as expected. In fMRI we are generally less concerned with true (white) thermal noise than we are with erroneous modulation of our signal. It's not noise so much as it is signal changes of no interest. For this reason, Gonzalez-Castillo et al. (doi: 10.1016/j.neuroimage.2010.11.020) recently proposed using a very low flip angle in order to minimize physiologic noise while leaving functional signal changes unchanged.

From ISMRM e-poster 3352, available as a PDF via this link.

What if we can't even attain the physiologic noise-limiting regime? It's quite possible to be in a subject motion-limiting regime, as anyone who has run an fMRI experiment can attest. In that case, the use of a high dimensional array coil (of 32 channels, say) could actually impose a higher motion sensitivity on the time series than it would have had were it detected by a smaller array coil (of 12 channels, say), due to the greater receive field heterogeneity of the 32-channel coil. This was something a colleague and I considered last year, in an arXiv paper ( and accompanying blog post. In an e-poster at this year's ISMRM Annual Meeting (abstract #3352; a PDF of the slides is available via this Dropbox link) we simulated the effects of motion on temporal SNR (tSNR), as well as the potential for spurious correlations in resting-state fMRI, when using a 32-channel coil. In doing these simulations we assumed perfect motion correction yet there were still drastic effects, as the above figure illustrates.

Whether the equivocal benefits of a 32-channel coil for routine fMRI (that is, using 3-ish mm voxels) are due to enhanced motion sensitivity, higher physiologic noise or some other factor I'm not in a position to say with any certainty. My colleagues and I, and others, are investigating ways that we might reduce the effects of receive field contrast on motion correction. The use of a prescan normalization is one idea that might help, at least a bit. The process has many assumptions and potential flaws, but it may offer the prospect of getting back some of what might be lost courtesy of the enhanced motion sensitivity. We simply don't know yet. The bigger problem, however, seems to be that a heterogeneous receive field contrast will impart motion sensitivity on a time series even if motion correction were perfect. Strong receive field heterogeneity, of the sort exhibited by a 32-channel head coil, is a killer if the subject moves.

Unless you are attempting to use highly accelerated parallel imaging (in particular the multiband sequences) and/or pushing your voxel size towards 2 mm, then, you're almost certainly better off sticking with the 12-channel coil as far as fMRI performance is concerned. Other scans, in particular anatomical scans and perhaps some diffusion-weighted scans, may benefit from larger array coils (because these scans may be in the thermal noise-limiting regime), but each application will need to be verified independently.


  1. Very interesting - thanks. So, from what you've said here, I'm guessing that the new 64-channel head-coil systems available with the Siemens Prisma range ( won't provide much (or, any) added benefit either?

  2. Hi Matt,

    If the 64 channels are all used around the head then, yes, it's possible that performance of that coil will be worse - more motion-sensitive - than the current 32ch coil. But there are situations where the coil may well be better. For fMRI, the multiband (or simultaneous multislice) EPI sequences may well benefit, permitting improved high-resolution (< 2.5 mm voxels) whole head fMRI at 3 T. The MB-EPI sequences themselves need validation, but there is hope that larger arrays will only help there.

    The real benefits of a 64ch coil on a Prisma will be for diffusion imaging. The Prisma has x2 gradient strength compared to a Trio. That's worth x4 for b value. And since diffusion is often in the thermal noise-limiting regime, having higher SNR from the 64ch coil will help, too.

    What we really need for fMRI is better motion correction. Prospective motion correction could fix T1 effects. What we need after that is a better way to update the shims and receive field contrast in real time, so that spurious signal changes aren't imprinted onto the time series. Until we have these, however, the only solution to the motion problem is to avoid it. Big array coils have potential benefits when motion is low, but they sharpen the failure profile when motion is a possibility.

    My suggestion is to minimize risks in the acquisition unless those risks are warranted for the experiment. If you can answer your neuroscience question with 3 mm voxels then stick with the 12ch coil, etc. Only if you need 2 mm voxels is it worth amplifying the risk with a 32+ channel coil. And in those cases you have to bear in mind you're a test pilot!



  3. 64 ch coils are exciting, I'm just not sure how all the raw data will be transferred/stored. If it really does generate twice the raw data as the 32 ch, we're talking about upwards of 10 GB per min!

    Anyways, definitely anyone wanting to run something similar to the Human Connectome protocol will have to use the 32 ch coil (i.e. 2mm iso, < 800 ms TR, and whole brain coverage).

    Both the high spatial resolution and short TR (small Earnst flip angle) will certainly bring you into the thermal noise dominated regime. With the fast TR you also reduce the frame to frame sensitivity to motion as well as the need for slice time correction (although it would not help with the supposed receive field heterogeneity effect on tSNR).

    The nice examples of prospective motion correction by kineticor using a bore mounted camera are quite nice.

    However, even just having a camera to provide subjects visual feedback of their motion and training them to keep REALLY still - is very powerful in terms of minimizing motion artifacts (at least for normal subjects). The camera is also very handy for scan operators to visually detect subject motion during ipat/MB reference scans (even before images are reconstructed) to abort motion contaminated runs.

  4. Hi Joseph,

    Thanks for the feedback, and for the link. Very useful information! To be sure, the issues surrounding MB-EPI are far more complex and have yet to be fully worked out. I understand that the most recent MB-EPI version from CMRR (R009) will work with a 12ch coil, and I've tested it myself very briefly, but I agree that to get the MB-EPI running well would necessitate a 32ch coil.



  5. Please see a recent article comparing the two coils under discussion.

  6. Hi Sheeba,

    Thanks for the link. Unfortunately I can't review the methods because the paper is behind a pay wall and my institution doesn't subscribe. In any event, my guess is that you've used high resolution - probably around 2 mm - and possibly acceleration. And in one or both of these situations it's likely that the larger coil arrays will perform better.

    The 2011 paper by Triantafyllou, Polimeni and Wald (doi: 10.1016/j.neuroimage.2010.11.084) sets out the case for high array coils nicely, I think. The benefits for high resolution and/or acceleration are evident. Where we have to be careful, however, is in extrapolating from one study to another; in this case, extrapolating from high resolution (2 mm) and assuming that the benefits will hold at low resolution (3+ mm). Each application probably needs to be evaluated separately.

    In my opinion we need lots more studies of the sort that you've just published, studies that set out the benefits, limits, etc. of each innovation that appears on our scanners. We can't (rather, we shouldn't) just assume that things will work the same - or better - every time we change a part of the acquisition. A 32ch coil may - or may not - offer a benefit over a 12ch coil for application X. Most fMRIers would prefer that the only novelty in their experiment be in their neuroscience question, and would probably prefer that their tools weren't additional variables in the mix!



  7. Please note that we did not use any acceleration. We have done comparisons at different fMRI settings (high and low spatial resolutions):

    It is important to note that the coil geometry is also very important. This becomes even more evident in simultaneous multi-slice applications, especially at high multi-band factors (6 or more, TR<1s) where 12ch coil simply becomes useless for providing a combination of whole-brain coverage and high spatial and temporal resolutions.

  8. Hi Sheeba,

    Great, thanks for the new link! Your results in the 2010 abstract support the view that the 32ch coil is better than the 12ch coil when resolution is significantly better than 3 mm in-plane. So I think there's a consistent story emerging across all the available literature. Essentially,

    if (voxels > 2.5 mm) and (acceleration = OFF) then (RF coil = 12ch)

    if (voxels < 2.5 mm) then (RF coil = 32ch)

    if (acceleration = ON) then (RF coil = 32ch)

    Here I've proposed that the 32ch coil should be superior to the 12ch coil whatever sort of acceleration is used, e.g. in-plane GRAPPA with R=2+, or simultaneous multislice (aka multiband) of any acceleration factor. I agree that there is little point trying to use a 12ch coil for high MB factors, much as there is severe degradation of in-plane GRAPPA for R=3+ when using the 12ch coil.

    But the costs/benefits of MB are a large, separate subject! It isn't yet clear what combination of parameters is best for MB-EPI with a 32ch coil. And I would only attempt MB-EPI on a 12ch coil if I didn't have a 32ch coil to use. In which case I would try to restrict the acceleration to MB=2 or 3. But that's a story for another day...



  9. Please note that I've been using WIP 770A for my SMS evaluations with the two coils. Preliminary results indicate that the total acceleration factor (i.e., slice times in-plane) that a given coil can handle (at 2iso resolution) cannot exceed one-third the coil count. I've also tried 1.5 iso and 1 iso resolutions with the 32ch, SMS combinations at 3T, but may be higher field strengths and or 64Ch coil would be the way to go for that.. :)

  10. Thanks for the rule of thumb!

    I've been using the CMRR version, R009 the most recent revision. Am still waiting to get a hold of the MGH WIP via Siemens. So far I've been trying to go the other way. That is, starting with "Connectome" type of parameters with MB=6 or 8, decreasing the resolution a smidge (from 2 to 2.5 mm) in order to allow shorter echo spacing (from 0.8 ms to below 0.6 ms on a Trio) and full Fourier sampling instead of 7/8ths partial Fourier. These changes also permit the TE to come down from around 40 ms to around 32 ms. To keep the TR reasonably short - a second or less for whole brain - I've dropped to MB=4, but that's just because I figure I should accelerate as fast as necessary, and no farther. I haven't directly compared the different MB factors.

    I've noticed that the distortion and dropout of the Connectome style of parameters are rather drastic; the distortion even more so than the dropout. And I am reluctant to use in-plane GRAPPA for fear of enhancing the motion sensitivity. I am, however, investigating different phase encoding directions. So far, for axial slices, P-A shows reduced dropout for deep brain than does A-P. Frontal lobes also get stretched rather than compressed. But occipital lobes then get compressed. At some point I'll try to submit some examples of my compromise protocols to the CMRR's gallery,, and perhaps do a quick summary post here.

  11. I like the nice compromise between TR and MB (1s and 4 respectively) for whole brain coverage. However, when I drop the TR lower, say 0.6s with a MB of 8 (TE of 30ms, 6/8 pF) I get even better statistics (stronger correlations) but this could purely be due to the merit of having more samples.. Not sure if that serves as a compelling enough argument to go towards lower TRs, unless one is invested in phMRI.. :)

  12. Yeah, that makes sense. The way I'm approaching it I'm trying to restrict echo spacing < 0.6 ms, TE < 35 ms and allow full Fourier. These set the cap on in-plane resolution at about 2.4 to 2.5 mm for whole brain, depending on slight FOV changes. (I generally use 210-224 mm so that large heads will always fit.) But these restrictions don't generally alter what can be done in the slice dimension, so it would be feasible to run the same 2.5x2.5 mm in-plane resolution with 2-2.5 mm slices and choose MB and TR as appropriate for the application. In my examples I'll try to get MB from 3-6, commensurate with the minimum TR in each case, so that people can evaluate the time series for themselves. (I'll probably acquire a fixed time period of around 4 mins.)

    One other benefit of the reduced in-plane res and lower MB: faster recon! With the Step 4 MRIR, the recon will just about keep up with the acquisition if the voxels are 2.5 mm iso and MB=4! How is the recon speed on WIP770?

  13. This comment has been removed by the author.

    1. Yes, recon speed is certainly baseline/imager dependent.