Tuesday, August 16, 2016

that's motion? respiration

In the previous post I showed some motion regressors with very regular oscillations. Several of you pointed towards respiration, and I was able today to extract the signal from a respiration belt for a couple of runs. The respiration signal (black line plots) is not perfectly temporally aligned to the movement regressors, but close enough to convince me that respiration is driving the oscillation. We're still looking into various acquisition details for why it is so prominent in this dataset; I am certainly not an MR physicist, but will summarize when we figure something out.


  1. Any chance you can run your sequence on a (preferably spherical) phantom and show us the motion regressors?

  2. Some more details extracted from emails on investigation so far:

    We can attribute the modulation of B0 via magnetic susceptibility of the chest, rather than changes in arterial CO2 or some other physiological consequence of respiration-coupled BOLD signal changes because the effect is instantaneous. If there was an offset we would be looking for downstream effects, but this is a direct mechanical effect. If you got in the magnet with a phantom and only scanned the phantom you'd still see the effects of your breathing on the phantom data. It would appear to be moving with your breathing!

    The Prisma is running VE11B.

    Okay, good. Then the next crucial question is whether the latest version of MB-EPI (which is R014) shows the same problem as the last version (likely R013a) which was presumably used to collect these scans. Is there a new feature in R014 that we can use to affect the breathing modulation...?

    Version R014 of MB-EPI from CMRR was released a couple of weeks ago and is probably on your Prisma now, because R013a expired on 1st Aug. I am going to ask the CMRR folks directly but from clues in the release notes my guess is that you will now have a per TR on-resonance frequency adjustment. This comes from the newly created ability, in R014, to turn one off! (If it's not there, you can't turn it off!) The correction was put in as far back as R07 in 2012, but only for the VD line of software. That is, for the s/w driving Skyra and some Verio scanners. VE11 (now at rev c) is the first version of software designed to work across all Siemens scanners on the market today. (So it doesn't work on my Trio, because that's off the market. We are stuck at VB forever.) Since the Prisma doesn't have a drift problem (it's not a wide body magnet with slim gradients) there would have been no apparent reason to include the frequency drift correction in Prisma-line software until having to include it by virtue of consolidating s/w into the VE11 platform. Hence also the need to be able to shut it off, for continuity.

    Okay so it turns out Russ used a Skyra for MyConnectome. This will almost certainly have been with a version of MB-EPI that had a per volume frequency adjustment, because the thermal drift on the Skyra was a show-stopper when it first came out. (Siemens had to add the frequency adjust feature to their product EPI in VD software.)

    Now I'm wondering if you might have been using the frequency adjustment version inadvertently (and unnecessarily?) on your Prisma, and also whether this adjustment is producing (exacerbating) the chest motion effects. And finally I'm wondering why there is now a feature to disable the adjustment in R014. Is it because a problem was identified by CMRR?

    1. Even if you can't turn off the frequency adjustments, you can go to your scanner settings (as advanced user) and force an empty ICE Program. Running the sequence with an empty ICE Program eliminates the realtime frequency feedback, and once the acquisition is complete you can Retro Recon your data normally.

      This should be enough to check if things change for better or worse.

    2. Good idea. Thanks! Once we've got a slightly better handle on the extent of the issue (I've not looked on Trio data specifically, for example) and any patterns, I'll probably open a new issue on the CMRR github so they can investigate.

  3. Open questions:

    1. Did problem exist only for VD versions of MB-EPI?

    2. Can problem be replicated on VE version now, on both Skyra and Prisma scanners?

    3. Problem observed by Power (http://www.sciencedirect.com/science/article/pii/S1053811916303871) for MyConnectome data using MB factor 4, as well as here for MB = 8. Any MB dependence?

  4. A question for Skyra owners running VDx software: do you see similar oscillations in EPI data acquired with the Siemens product sequence? My suspicion is that you do/will, due to the per volume frequency adjustment. If you don't then we are looking at a specific interaction of the freq adjustment with the SMS scheme.

    1. Very unlikely that the issue has something to do with the SMS scheme. All that matters for the frequency feedback is the frequency at the isocenter at the end of each volume. All slices are then corrected based on their *anatomical* position relative to the isocenter, which is independent of MB factor.

      What changes with SMS is that the TR gets short enough so we sample multiple phases of the respiration and get to see these oscillations.

      Now, is it really a bad thing that the frequency is being corrected? Aren't these curves actually showing us that things are going well?

      It'd be interesting to look at the phase of freq. feedback corrected images and see if the usual respiratory fluctuations diminish.

      I wouldn't be surprised to find out that there is a trade-off between "fake" motion artefacts and physiological noise.

      Interested to hear more about these effects.

    2. Hi Daniel, I agree that it's unlikely a feature of SMS, but I'll keep an open mind pending further evidence. If it's in product EPI on the Skyra with typical TR of 2 sec then it should be visible. That's plenty fast to sample respiration above Nyquist. The oscillations may be harder to see, which is perhaps why nobody (?) has noticed until now.

      But we have to bear in mind that using blipped CAIPI means that frequency shifts between the reference (single band) images and the current undersampled volume could have interesting, novel effects. There is much to learn yet!

      The frequency correction in VD was intended to compensate from slow drift due to heating, primarily on the Skyra. Trios don't experience very large drifts because of the gradient and passive shim construction. I'm expecting/hoping Prismas to be the same. So only the wide bore gradient/magnet scanners likely need it, to avoid requiring a massive linear term in the realignment.

      Is it a bad thing that respiration is being corrected? That's the thing! I'm not sure it is! If it were being corrected fully the images should remain stationary. I'm not familiar with how the correction is done (yet) but I would expect that a good correction would yield negligible apparent shift in the PE direction. The way the adjustment is either computed or calculated may need slight modification. It probably depends on what is being used now for the reference frequency. Is it, for example, a single volume of data at the start of the time series?

    3. "computed or calculated" should have read "computed or implemented." Sorry!

    4. Am looking into the product transmit (Tx) frequency feedback, which was enabled for scanners running VD11 from R007 on, and for VD13A from R011 on. It's also now in VE11C for R014. Apparently there is a facility to disable Tx frequency feedback on all VD, VE versions that incorporate it. Not sure yet whether disabling is removing the oscillations.

  5. If it's of any help, I have seen the same oscillatory pattern in the motion regressors of data acquired with an SMS sequence on a 3T GE MR750 system, and it does seem to be predominantly related to respiratory rate. It is particularly obvious when comparing data from infants/toddlers (with a faster respiratory rate) to that of adults.

    I also have data acquired with different multiband acceleration factors (no parallel imaging acceleration), and the oscillatory pattern is visible even in the mb1 (TR=2456ms) motion regressors.

    I have uploaded some figures here, if you are interested: https://www.dropbox.com/s/ylfci85on9zao17/multiband_oscillations.pdf?dl=0

    1. This is very, very interesting! Are your analyses (resting state? event-related?) going well event with the oscillatory patterns? Are you going forward with the study with the oscillations?

      I think many people will be interested in your findings, especially across the multi-band levels; is it ok if I put this comment (and you pictures) into a separate post? You're welcome to write more, too; comment here or email me directly (jetzel@wustl.edu). Thanks!

    2. I believe that GE scanners use a per volume on-resonance frequency feedback adjustment, as Siemens Skyra scanners running VD software were required to do, in order to combat drift from heating. So this is quite likely consistent behavior. I have had one report of similar behavior with a product EPI sequence on a Skyra which leads me to believe that we are looking at a frequency feedback issue rather than an SMS-specific issue. I don't have a scanner capable of running these tests but I'm going to ask CMRR to test with and without the frequency feedback because by best information suggests that their custom sequence (for VE) is the only one for which the freq feedback can be turned on/off.

    3. Jo, definitely feel free to use any of these images! I have been wondering about this pattern for a while, and pointed it out to a few people but so far nobody has had an answer. I was really exited when I saw your post, and all the excellent comments!

      Following up, on my previous comment, I have looked at some additional data acquired with a product GE sequence (same scanner but 8 channel head coil) with a 2s TR, and while by far not as obvious and visible in far fewer subjects, I think there is a similar pattern in those motion regressors as well: https://www.dropbox.com/s/iho3q4yyielbc0u/multiband_oscillations_GE_EPI.pdf?dl=0. I have also noticed the same thing you did when comparing your data to the four random HCP subjects: the directions most impacted change between the GE scanner and the data I have from the Prisma. Feel free to email me too, if you have any questions or need anything else: annikalinke@gmail.com