apparent motion also at 7T

WUSTL (ahem, WashU) recently set up a 7T scanner (Siemens Terra; 8Tx32Rx_Head_C head coil), and we've started a bit of piloting and exploring what we can do with it. I've been working through multiple aspects: the BOLD images themselves look different because of the stronger magnet, we're trying some new-to-me multi-echo sequences, collecting some noRF and phase images, and even the BIDS and fMRIPrep parts have taken some script updating. I think I'm getting enough of a handle on things now (thanks for help via neurostars, Chris Markiewicz and Taylor Salo!) to share some impressions, but it's all still very much in process.

We've had two pilot sessions so far, each with a different participant and somewhat different sequences. We're most interested in task fMRI, but that's tricky at the moment since the scanner doesn't (yet) have a way to present stimuli, nor to record responses. Eventually I want to do sequence comparisons with the reward-possible DMCC proactive Cued Task-Switching paradigm (and probably other tasks), like in my OHBM 2023 poster. But to get started I asked the second pilot participant to do a self-paced version of the HCP Motor task: blocks of right finger tapping, left finger tapping, right toe wiggling, left toe wiggling, tongue moving, with a bit of rest in between and a few deep breaths before and after each movement block to serve as onset/offset markers. (I'm not going to discuss the movement analysis parts yet, but so far I'm encouraged.)

Both pilot sessions included a pair (PA/AP) of runs with an acquisition similar-ish to what we used in the DMCC at 3T: 2.4 mm iso voxels, MB4, TR 1.2 s. But the 7T allows more acceleration, so they added in-plane acceleration (GRAPPA 2) and collected 4 echoes instead of just 1.

Below are the realignment parameters (from the fmriprep derivative _desc-confounds_timeseries.tsv files) for those runs from the two participants we've had so far. In both cases the grey vertical lines are at one-minute intervals; the first participant (TB7T1)'s runs were about 10.5 minutes and he alternated periods of regular and slow/deep breaths; the second participant (TD7T1)'s runs were about 6.5 minutes each and he did the blocked breathing-motor task.

TB7T1 clearly has much more apparent motion than TD7T1; the periods of regular and deep breathing are obvious, not only in the realignment parameters but also in movies of the BOLD run. TB7T1_run-44_echo-1_bold.avi is the "rawdata" (before preprocessing) version of echo 1; TB7T1_run-44_space-MNI152NLin2009cAsym_desc-preproc_bold.avi is after preprocessing (with fMRIPrep 25.1.0 using Tedana to optimally combine the echoes). The brain sort of looks like it's "jumping" with the breaths in the raw movie; perhaps more like expanding and contracting in the preprocessed version. 

The TB7T1 participant is the same individual as in some of our previous acquisition tests (at 3T); clearly, using 7T doesn't mean we can forget about apparent motion. ... I wouldn't have forgotten about apparent motion regardless since it's a favorite topic of mine, as is probably obvious since I'm starting off this (hopefully) series of posts about our 7T piloting with it.

Each of these sessions included multiple different acquisition sequences; in all cases TB7T1 had more obvious apparent motion than did TD7T1. TB7T1 run 28 had the largest apparent motion; it also had smaller voxels and a longer TR than the other examples.

Finally, the TB7T1 run 44 motion is also striking in the greyplot version created by fmriprep (left); TD7T1 run 24 is below, right:

"crescent" artifacts revisited

Back in 2018 I posted twice about "crescent" artifacts (first, second), and have kept a casual eye out for them since, wondering how much they might affect fMRI analysis results. 

The artifact isn't totally consistent across people, but when present, is very stable over time (i.e., sessions several years apart), and bright in temporal standard deviation QC images. The artifact's location varies with encoding direction: front for PA encoding, rear with AP encoding; the PA encoding versions tend to be much brighter and obvious than the AP. 

Below is an example of the artifact, pointed out by the pink arrows. This is the temporal standard deviation image for sub-f8570ui from the DMCC55B dataset (doi:10.18112/openneuro.ds003465.v1.0.6), the first (left, AP encoding) and second (right, PA encoding) runs of Sternberg baseline (ses-wave1bas_task-Stern), after fmriprep preprocessing:

These two runs were collected sequentially within the same session, but the artifact is only visible in the PA encoding run (right). (Briefly, DMCC used a 3T Siemens Prisma, 32-channel headcoil, CMRR MB4, no in-plane acceleration, 2.4 mm iso, 1.2 s TR, alternating AP and PA encoding runs; details at openneuro, dataset description paper, and OSF sites, plus DMCC-tagged posts on this blog.)

In the previous "crescent" posts we speculated that these could be N/2 ghosts or related to incomplete fat suppression; I am now leaning away from the Nyquist ghost idea, because the crescents don't appear to line up with the most visible ghosts. (Some ghosts are a bit visible in the above image; playing with the contrast and looking in other slices makes the usual multiband-related sets of ghosts obvious, but none clearly intersect with the artifact.) It also seems odd that ghosts would be so much brighter and change their location with AP vs. PA encoding; I am no physicist, though!

link to cortex volume?

This week I gave three lab members a set of temporal standard deviation images (similar to the pair above) for 115 participants from the DMCC dataset. The participant images were in random order, and I asked the lab members to rate how confident they were that each showed the "crescent" artifact or not. My raters agreed that 34 participants showed the artifact, and 39 did not. (Ratings were mixed or less confident on the others; I asked them to decide quickly from single pairs of QC images, not investigate closely.)

We didn't measure external head size in the participants, but did run freesurfer during preprocessing, so I used its CortexVol and eTIV statistics as proxies (a different stat better?): and the group my colleagues rated as having the artifact tended to have smaller brains than those without:

If the appearance of this artifact is indeed somewhat related to head size, then it's logical that it would (as I've observed) generally be stable over time. DMCC's population was younger adults; it'd be interesting to see if there's a relationship with a wider range of head sizes.

only with DMCC or its MB4 acquisition?

Is the artifact restricted to this particular acquisition or study? No, not somehow related to DMCC; I've checked a few DMCC participants with the artifact who later participated in other studies and they have it (or not) in all of the datasets.

To see if it's restricted to the MB4 acquisition, I looked at a few images from a different study, which also has adult participants, a 3T Prisma, 32-channel headcoil, 2.4 mm iso voxels, and CMRR sequences, but with MB6 TR 0.8 s, and PA encoding for all runs. Below are standard deviation images for three different people from this MB6 study, one run each of the same task, after fmriprep preprocessing. (I chose these three because of the artifacts; not all are so prominent.)

Since this study has all PA runs I can't directly compare artifacts across the encoding directions, but there are clearly some "crescents", and more sets of rings than typical with MB4 (which makes sense for MB6). The rings are especially obvious in person #3; some of these appear to be full-brain ghosts. I suspect the artifacts would be much clearer in subject space; I haven't looked (I'm not directly involved in the study). But a substantial minority of these participants' standard deviation images resemble #1, whose artifacts strike me as quite similar to the "crescents" in some DMCC MB4 PA images.

but does it matter?

Not all artifacts that look strange in QC images actually change task-related BOLD statistics enough to be a serious concern. (Of course, how much is too much totally depends on the particular case!) I suspect that this artifact does matter for our analyses, though, both because of where it falls in the brain and because it affects BOLD enough to be visible by eye in some cases.

The artifact's most prominent frontal location with PA runs puts it uncomfortably close to regions of interest for most of my colleagues, and is one reason I have advised shifting to all AP encoding for new studies I'm involved with. Preprocessing, motion, transformation to standard space, and spatial smoothing blurs the artifact across a wider area, hopefully diluting its effect. But the artifact's location is somewhat consistent across participants, and present in a sizable enough minority (a third, perhaps, in the datasets I've looked at), that it seems possible it could reduce signal quality in our target ROIs.

For showing that it does indeed actually affect task-related BOLD enough to matter, so far I mostly just have qualitative impressions. For example, below left is the standard deviation of one DMCC person's PA Sternberg run, with the cursors on the artifact. The right side is from averaging together (after voxel-wise normalizing and detrending) frames after pressing a button with the right hand. Squinting, the statistical image is brighter in sensible motor-related grey matter areas, marked with green. But the "crescent" may also be faintly visible, as pointed out in pink.

I can imagine quantitative tests, such as comparing the single-run (so separating AP and PA encoding runs) GLM output images from the group of participants with and without the artifact. Differences in estimates in parcels/searchlights/areas overlapping the artifact would be suggestive, particularly as the estimates vary with encoding direction and participant subgroup (with-artifact or without).

thoughts?

I'm curious, have you seen similar? Do you think this artifact is from N/2 ghosting, incomplete fat suppression, or something else? (What should I call it? "Crescent" is visually descriptive, but not standard. 😅) Seem reasonable it could be related to head size? And that it can significantly affect BOLD? Other reactions? Thanks! (And we can chat about this at my OHBM 2025 poster, which will be on this topic.)

What happened in this fMRI run? ... happened again.

Back in July I posted about a strangely-failed fMRI run, and yesterday I discovered that we had another case not quite two weeks ago. This is the same study, scanner (3T Siemens Prisma), headcoil (32 channel), task, and acquisition protocol (CMRR MB4) as the July case, but a different participant. I've contacted our physicists, but we probably can't investigate properly until after the holidays, and are hampered by no longer having access to some of the intermediate files (evidently some of the more raw k-space/etc. files are overwritten every few days). 

I've asked our experimenters to be on the lookout, and while hopefully it won't happen again, if it does, I hope they can catch it during the session so all the files can be saved. If anyone has ideas for spotting this in real time, please let me know.

A possibly-relevant data point: the participant asked to have the earbuds adjusted after the first task run. The technician pulled the person out of the bore to fix the earbuds, but did not change the head position, and did not do a new set of localizers and spin echo fieldmaps before starting the second task run (the one with the problem). I've recommended that the localizers and spin echo fieldmaps be repeated whenever the person is moved out of the bore, whether they get up from the table or not, but the technician for this scan did not think it necessary. What are your protocols? Do you suggest repeating localizers? No one entered the scanning room before the problematic July run, so this (pulling the person out) might be a coincidence.

Here's what the this most recent case looks like. First, the three functional runs' DICOMs (frame 250 of 562) open in mango, first with scaling allowed to vary with run:

Then with scaling of 0 to 10000 in all three runs, showing how much darker run 2 is:

And finally the SBRef from run 2:

In July the thinking was that this is an RF frequency issue, possibly due to the FatSat RF getting set improperly, so that both fat and water were excited. But this seems hard to confirm from the DICOM header; this time, the Imaging Frequency DICOM field (0018,0084) is nearly identical in all three runs: 123.258928, 123.258924, 123.258924 (runs 1, 2, and 3 respectively), which is very similar to what it was in July (123.258803).

What happened in this fMRI run?

This is one of those occurrences (artifacts?) that is difficult to google, but perhaps someone will recognize it or have a guess.

This run is from a session in which a person completed four fMRI runs of a task sequentially. They did not get out of the scanner between these runs, nothing was changed in the protocol, no one entered the scanner room. Later participants (with the same protocol, scanner, etc.), have been fine. This study uses CMRR MB4 acquisitions, so we have an SBRef image for each run; the artifact is the same in the SBRef and functional run.

Runs 1 (not shown), 2, and 4 are normal, but run 3 is much darker than the others and has an obvious ghost-ish artifact, here are the DICOMs from each run's SBRef, allowing mango to adjust the contrast in each:

And here they are again, with contrast set to 1-15000 in all three images:

The functional run's DICOMs are also dark and have the prominent artifact; here's a frame:

When the run is viewed as a movie in mango the blood flow, small head movements, etc. are plainly and typically visible. The artifact does not appreciably shift or change over the course of the run, other than appearing to follow the (small) overt head motions (when the head nodded a bit, the artifact shifted in approximately the same way). The two surrounding runs (2 & 4) are typical in all frames (no sign of the artifact).

Given that this artifact is in the DICOMs, it's not introduced by preprocessing, and I am assuming this run is unusable. I'd like an explanation, though, if nothing else, so we can take any steps to reduce the chance of a recurrence. Our best guess at this time is some sort of transient machine fault, but that's not an especially satisfactory explanation. 

Any ideas? Thanks!

update 13 July 2022:

In response to Ben and Renzo's suggestions, I skimmed through the DICOM headers for fields with large differences between the three runs; if there are particular fields to look for, please let me know (this is a Siemens Prisma); I am not fluent in DICOM header! The most obvious are these, which I believe are related to color intensity, but I'm not sure if it's reporting a setting or something determined from the image after it was acquired.

run 2 (typical)

(0028,0107) Largest Image Pixel Value 32238

(0028,1050) Window Center 7579

(0028,1051) Window Width 16269

(0028,1055) Window Center & Width Explanation Algo1

run 3 (dark/artifact)

(0028,0107) Largest Image Pixel Value 3229

(0028,1050) Window Center 1218

(0028,1051) Window Width 3298

(0028,1055) Window Center & Width Explanation Algo1

run 4 (typical)

(0028,0107) Largest Image Pixel Value 31787

(0028,1050) Window Center 7423

(0028,1051) Window Width 15912

(0028,1055) Window Center & Width Explanation Algo1 

 

And here's yet another view from the three functional runs, in which I played with the contrast a bit. There's definitely a difference in which structures are brightest between the three.

 

more crescents: with sequence variations

Back in January I posted about the "crescent" artifacts that show up in the PA runs of some people. (In a few people reversed crescents appear in the AP runs as well.) The working hypothesis is still that these are N/2 ghosts and perhaps related to insufficient fat suppression. I am still extremely interested in any advice people have for avoiding these, especially as I now have evidence that task signal is noticeably reduced in the "crescent" areas (details soon, hopefully).

We have begun a set of pilot scans to see if some parameter combinations produce better subcortical and frontal signal in a reward task. So far all scans have been on a Siemens Prisma, 64 32 channel head coil, CMRR MB4 sequences; we'll be doing the tests on a Siemens Vida as well in a few weeks. So far, scans are acquired with 2.4 or 3.0 mm isotropic voxels, either "flat" (AC-PC aligned as usual) or "tilted" (30 degrees off AC-PC); more acquisition details below the jump.

Of the two pilot people (so far), one has the crescent artifact and the other does not, and the appearance of the crescents in the different acquisitions is interesting. All of these images are voxelwise standard deviation, calculated over the entire run (no censoring, but extremely low motion), and on raw images (preprocessing is in progress).

First, here are sagittal views of a flat (left, scan 15) and tilted (right, scan 37) 2.4 mm isotropic run. The crescent artifact is visible in both; I marked the approximate ends with green arrows (click to enlarge). The multiband slice boundary is visible in both a fourth of the way up the image (red arrows).

Here are axial slices of the set of runs we have so far for this person. All are with the same color scaling (0 to 200); brighter is higher standard deviation. These are raw images, so the slice appearance varies quite a bit between the "flat" and "tilted" runs.

The "crescents" are visible in all PA runs, though perhaps easiest to spot with the 2p4 (2.4 mm isotropic) voxels. The slices in which the crescents appear varies between the tilted and flat acquisitions (e.g., k=31-46 for run 15_2p4flat_PA; 22-31 for run 37_2p4tilt_PA). It will be easier to compare the crescent locations after preprocessing.

The 3p0 (3.0 mm isotropic voxels) images are generally more uniform and dark than the 2p4 runs, likely reflecting improved signal-to-noise, particularly in the middle of the brain. While the large vessels are brightest in all runs (as they should be), the runs with 2.4 mm voxels (2p4) generally have a "starburst" type effect (brighter in the center, darker towards the edges), which is worrying, particularly since we want good signal in reward areas.

I will share other observations on this blog as the piloting and analyses progress. Please contact me if you'd like to run your own analyses; we'd be happy to share and are very interested in others' thoughts.

UPDATE 18 April 2018: I've wondered before if head size was a factor in which people have the crescent artifact, using the total intracranial volume measurement produced by freesurfer as a proxy. I don't have those measurements yet, but they kindly allowed me to measure their heads as if fitting them for hats, and they were nearly identical: about 58 cm for the person without the crescent artifact, and about 57 for the person shown in this person (with the artifact). Both people have a normal healthy body size; the person without the artifact was a bit shorter (around 5'2") than the other (around 5'7"). So, at least for these two pilots, external head size doesn't seem to matter for the artifact.

more acquisition parameters below the jump

Holy crescents, Batman!

Quite a few of the posts over the last year or so have arisen from things that catch my eye as I review the SMS/MB4 images we're collecting in our ongoing project, and this is another. For quick comparison, I make (with knitr; we may give mriqc a try) files showing slices from mean, standard deviation, and tSNR images for participants, runs, and sessions.

Some participants have obvious bright crescent-shaped artifacts in their standard deviation images (the examples above are from two people; both calculated from non-censored frames, after completing the HCP Minimal Preprocessing pipeline). Looking over people and runs (some participants have completed 6 imaging sessions, over months), people have the crescents or not - their presence doesn't vary much with session (scanning day), task, or movement level (apparent or real).

They do, however, vary with encoding direction: appearing in PA phase encoding runs only. Plus, they seem to vary with subject head size, more likely in small-headed people (large-headed people seem more likely to have "ripples", but that's an artifact for another day).

All that (and thanks to conversations with practiCal fMRI and @DataLoreNeuro) gave a hint: these crescents appear to be N/2 ghost artifacts.

Playing with the contrast and looking outside the brain has convinced me that the crescents do align with the edges of ghost artifacts, which I tried to show above. These are from a raw image (the HCP Minimal Preprocessing pipelines mask the brain), so it's hard to see; I can share example NIfTIs if anyone is interested.

So, why do we have the bright ghosts, what should we do about it, and what does that mean for analysis of images we've already collected? Suggestions are welcome! For analysis of existing images, I suspect that these will hurt our signal quality a little: we want the task runs to be comparable, but they're not in people with the crescent: voxels within the crescent areas have quite different tSNR in the PA and AP runs.

Holy crescents, Batman! (We've been watching the 1966 Batman TV series.)

not all MB4 fMRI scans are free of gradients and banding

I was encouraged by the images in the previous post: it looked like the gradient and banding ("washing out") artifacts were not visible in the MB4 people. I need to revise that a bit: I do have bands and gradients in at least some MB4 people, though to a much lesser degree than in our MB8 datasets. How much this affects our task analyses is still to be determined.

On the left is the same MB4 person and run from the previous post (27 September 2017), voxelwise standard deviation over time of a task fMRI run, after going through the HCP minimal preprocessing pipelines. I was glad to see that the vessels were brightest (as they should be), though concerned about the frontal dropout. The person on the right is another person scanned at MB4 (doing the same task; same encoding, scanner, etc.); same preprocessing and color scaling for both.

The vessels clearly don't stand out as much in the person on the right. It's hard to tell in the above image, but there's a gradient in the standard deviation and tSNR images, with better signal on the edges than in the center of the brain. Below is another run from the person on the right, tSNR calculated on the images ready to go into afni for GLM calculation (so through the HCP minimal preprocessing pipelines, plus smoothing and voxelwise normalizing). This tSNR image is shown at six different color scalings; it's easier to see interactively (download the NIfTI here), but hopefully it's clear that the darker colors (lower tSNR) spreads from the center of the brain to the edges, rather than uniformly.

Here is the same person and run again, with the standard deviation (first two) and tSNR (right pane) of the raw (no preprocessing, first and lower) and minimally preprocessed (right two) images. I marked a banding distortion with green lines, as well as the frontal dropout. The banding is perfectly horizontal in the raw image, at 1/4 of the way up the image (see larger image), which makes sense, since this is an MB4 acquisition. I included the larger view of this raw image since all three banding artifacts are somewhat visible; in our dataset the inferior band is generally the most prominent.

The banding and gradient artifacts are certainly less visually prominent in our MB4 than our MB8 images, but they are present in some MB4 people. I haven't systematically evaluated (and probably won't be able to soon) all of our participants, so don't have a sense of how often this occurs, or how much it impacts detection of task BOLD (which is of course the key question).

Below the jump: movement regressors for the two runs in the top image; the person with the banding and gradients had very low motion; less than the person from the September post.

voxelwise standard deviation at different MB/SMS acquisitions

I've previously posted about using standard deviation and mean images to evaluate fMRI data quality, and practiCAL fMRI has a very useful series of posts explaining how various artifacts look in these images. In this post I use HCP (MB/SMS 8, RL/LR, customized scanner), and some images from one of our studies (MB4 and MB8, AP/PA, Siemens Prisma scanner); see links for acquisition details, and afni commands to generate these images is after the jump.

These tests were inspired by the work of Benjamin Risk, particularly his illustrations of banding and poor middle-of-the-brain sensitivity in the residuals (e.g., slides 28 and 29 of his OHBM talk). He mentioned that you can see some of the same patterns in simple standard deviation (stdev) images, which are in this post.

Here is a typical example of what I've been seeing with raw (not preprocessed) images. The HCP and MB8 scans are of the same person. The HCP scan is of the WM_RL task; the other two are of one of our tasks and the runs are about 12 minutes in duration (much longer than the HCP task). These MB8 and MB4 runs have low overall motion.

 Looking closely, horizontal bands are visible in the HCP and MB8 images (marked with green arrows). None of the MB4 images I've checked have such banding (though all the HCP ones I've looked at do); sensible anatomy (big vessels, brain edges, etc.) is brightest at MB4, as it should be.

Here are the same runs again, after preprocessing (HCP pipelines minimal preprocessing pipelines), first with all three at a low color threshold (800), second, with all three at a high color threshold (2000).

The HCP run is "washed out" at the 800 threshold with much higher standard deviation in the middle of the brain. Increasing the color scaling makes some anatomy visible in the HCP stdev map, but not as much as the others, and with a hint of the banding (marked with green arrows; easier to see in 3d). The MB4 and MB8 runs don't have as dramatic a "wash out" at any color scaling, with more anatomic structure visible at MB8 and especially MB4.The horizontal banding is still somewhat visible in the MB8 run (marked with an arrow), and the MB4 run has much lower stdev at the tip of the frontal lobe (marked with a green arc). tSNR versions of these images are below the jump.

These patterns are in line with Todd et. al (2017), as well as Benjamin Risk's residual maps. I'm encouraged that it looks like we're getting better signal-to-noise with our MB4 scans (though will be investigating the frontal dropout). Other metrics (innovation variance? GLM residuals?) may be even more useful for exploring these patterns. I suspect that some of the difference between the HCP and MB8 runs above may be due to the longer MB8 runs, but haven't confirmed.

yet more with respiration and motion regressors

Last year I did a series of posts on our experiences with multiband (SMS) task fMRI, particularly related to the respiration-related oscillations that are very apparent in the motion regressors of some people. See also practiCal fMRI's posts, especially this one. The tests and analyses here extend those, and convince me that the oscillations are not due to actual head motion, but rather B0 modulation or something else related to chest motion and/or blood oxygenation ("apparent head motion").

These scans are all using this previously-described MB4 acquisition sequence, and an experienced test person with rather prominent respiratory oscillations (the same person is in these first two plots) but little overt motion. We've been evaluating using Caseforge headcases to reduce movement and ensure consistent head positioning in the headcoil. We have two Caseforge headcases for this person: one fits quite tightly, and the other a little looser. This gives a set of comparison runs: with foam packing only (allowing more movement); with the regular Caseforge (some, but not much, movement possible); and with the tight Caseforge (motion essentially impossible).

The plots below show the 6 motion regressors (calculated from SPM12's realignment) and raw respiration signal (with a Siemens belt). I'm pretty confident that the motion regressors and respiration are precisely aligned here (it was approximate in some of last year's plots). The vertical grey lines mark one-minute intervals; the TR was 1.2 seconds. The tick marks show task onset; each run had three blocks of a verbal-response Stroop task.

First, here is a run without a headcase (normal packing). The difference in raw respiration amplitude between rest and task blocks is clear, as is some drifting over the course of the run. The y and z regressors closely match the respiration trace; hopefully you can zoom in to see that the y, z, and respiration curves are in phase - if the respiration curve goes up, both the y and z lines also go up. This was the case for all AP (anterior-posterior) encoded runs.

Next, here is a run of the same task and encoding (AP), with the very tight headcase. The oscillations in the y and z are still tightly matched to the respiration and about the same amplitude as before, but the drifting and rotation is quite reduced.

Finally, here is the same task, but the looser Caseforge headcase, and the reverse encoding (PA). The rotation is perhaps a bit more than with the tight headcase, but overall drift is quite low. The magnitude of the y and z oscillations is again about the same as the previous plots. If you look closely, the y line is out of phase from the respiration and z lines: the z line still goes up when the respiration goes up, but the y goes down.

We ran other tests, including some breath-holding. This is about two minutes of an AP and PA run breath-holding segment. The y-axis flipping is hopefully easier to see here: the blue peaks match the respiration peaks with AP, but fall in between on PA.

This y-axis flipping with encoding direction, plus the consistency of oscillation size across head stabilizers, has me convinced that we're seeing something other than overt head motion: I just don't believe he could have adjusted his y-axis movement with encoding direction that precisely, even had he known which encoding we were using each run.

If any of you have thoughts, or would like to look at these datasets to run additional tests, please let me know.

UPDATE 11 October 2018: This dataset is now available for download on openneuro.org, DOI:10.18112/openneuro.ds002737.v1.0.1 (was 10.18112/openneuro.ds001544.v1.1.0), called "multibandCFtests". (Note: I updated the links August 2020; the dataset was not properly deidentified the first time, requiring correction by the openneuro team - thanks!)

adjusting my mental model: movement correlation after preprocessing

It's time to adjust my mental model of the fMRI signal: there's a lot more correlation with movement in the timecourses after preprocessing than I'd expected. That movement really affects fMRI is not at all new, of course, and is why including the motion regressors as covariates in GLMs is standard. But I'd pictured that after preprocessing (assuming it went well and included realignment and spatial normalization) the correlation with movement left in the voxel timecourses should be pretty low (something like normally distributed, centered on 0, ranging between -0.2 to 0.2), without much spatial structure (i.e., maybe some ringing, but fairly uniformly present over the brain). Asking around, I think this is a fairly common mental model, but it looks to be quite wrong.

For exploring, I simply used afni's 3dTcorr1D program to correlate the timecourse of every voxel in several preprocessed task fMRI datasets with each of the six motion regressors (generated during preprocessing). 3dTcorr1D makes an image with 6 entries in the 4th dimension (sub-brick, in afni-speak), one for each of the 6 motion columns; the value in each voxel the Pearson correlation between that voxel's timecourse and the movement column. I plotted these correlations on brains, and made histograms to summarize the distribution.

The correlations are much higher than I expected, even in people with very little movement. Here's an example; more follow. Below are the motion regressors from single task run (about 12.5 minutes long; HCP-style preprocessing; MB8 protocol), the correlation with each motion regressor, and a (density-style) histogram of the voxel-wise correlations. Color scaling for this and all brain images is from 1 (hottest) to -1 (coolest), not showing correlations between -0.25 and 0.25.

If my expectation (correlations normally distributed, centered on 0, ranging between -0.2 to 0.2) was right, there shouldn't be any color on these images at all, but there's clearly quite a bit: many voxels correlate around 0.5 with roll and pitch (4th and 5th brain rows are mostly "hot" colors), and around -0.5 with x, y, and z (first three rows mostly "cool" colors). There's some structure to the peak correlations (e.g., a hotter strip along the left side in slice 46), which may correspond with sulci or large vessels, but it's rather speckly. Note that this is a pretty low motion subject overall: less than 2 mm drift over the 12 minute run, and only 4 volumes marked for censoring (I didn't censor before the correlation).

Looking at other people and datasets, including from non-SMS acquisitions with larger voxels and longer TRs, it appears like correlations of 0.5 are pretty common: this isn't just some sort of weird effect that only shows up with high-resolution acquisitions. For another example, these are the histograms and motion regressors for four runs from one person included in this study (acquired with 4 mm isotropic voxels; run duration 7.6 min, TR 2.5 sec, SPM preprocessing). The corresponding brain images are below the jump.

So, really visible motion (which at least sometimes is linked to respiration) in the voxel activity timecourses (such as here) is to be expected. Unfortunately, the correlation is not always (or even usually) uniform across the brain or grey matter, such as below (just the correlation with x and y translation). It also looks like very little (under a mm) motion is needed to induce large correlations.

What to do? Well, adjust our mental models of how much correlation with movement is left in the activation timeseries after preprocessing: there's quite a bit. I'll be exploring further, particularly isolating the task windows (since I work with task, not resting state, datsets): how are the correlations during tasks? I'm not at all sure that applying a global signal regression-type step would be beneficial, given the lack of homogeneity across the brain (though I know there are at least a few reports on using it with task data). Censoring high-movement trials (i.e., not including them) is likely sensible. Interestingly, I've found similar MVPA performance in multiple cases with temporal compression by averaging and fitting a model (PEIs), which would not have been my guess looking at these correlation levels. Perhaps averaging across enough timepoints and trials balances out some of the residual motion effects? I am concerned, however, about respiration (and motion) effects remaining in the timecourses: it's clear that some people adjust their breathing to task timing, and we don't want to be interpreting a breath-holding effect as due to motivation.

Any other thoughts or experiences? Are you surprised by these correlation levels, or is it what you've already known?

an update and some musings on motion regressors

In case you don't follow practiCal fMRI (you should!), his last two posts describe a series of tests exploring whether or not the multiband (AKA simultaneous multi-slice) sequence is especially liable to respiration artifacts: start here, then this one. Read his posts for the details; I think a takeaway for us non-physicists is that the startlingly-strong respiration signal I (and others) have been seeing in multiband sequence timecourses and motion regressors is not from the multiband itself, but rather that respiration and other motion-type signals are a much bigger deal when voxels are small (e.g., 2 mm isotropic).

This week I dove into the literature on motion regression, artifact correction, etc. Hopefully I'll do some research blogging about a few papers, but here I'll muse about one specific question: how many motion regressors should we use (as regressors of no interest) for our task GLMs? 6? 12? 24? This is one of those questions I hadn't realized was a question until running into people using more than 6 motion regressors (the 6 (x,y,z,roll,pitch,yaw) come from the realignment during preprocessing; transformations of these values are used to make the additional regressors).

Using more than 6 motion regressors seems more common in the resting state and functional connectivity literature than for task fMRI (Power et al. 2015, and Bright & Murphy 2015 , for example). I found a few (only a few) task papers mentioning more than 6 motion regressors, such as Johnstone et al. 2006, who mention testing "several alternative covariates of no interest derived from the estimated motion parameters", but they "lent no additional insight or sensitivity", and Lund et al. 2005, who concluded that including 24 regressors was better than none.

Out of curiosity, we ran a person through an afni TENT GLM (FIR model) using 6 (left) and 24 (right) motion regressors. This is a simple control analysis: all trials from two runs (one in blue, the other orange), averaging coefficients within my favorite left motor Gordon parcel 45 (there were button pushes in the trials). It's hard to tell the difference between the model with 6 and 24 regressors: both are similar and reasonable; at least in this test, the extra regressors didn't have much of an effect.

My thinking is that sticking with the usual practice of 6 regressors of no interest is sensible for task fMRI: adding regressors of no interest uses more degrees of freedom in the model, risks compounding the influence of task-linked motion, and hasn't been shown superior. But any other opinions or experiences?

multiband acquisition sequence testing: timecourses 2

In a previous post I showed timecourses from the same person, doing the HCP MOTOR task, collected with a multiband 4 and a multiband 8 sequence (see previous posts for details). We ran another test person ("999010") through the same task, with the same two sequences (but not the MB0 control).

This plot shows the average activation (post-preprocessing) in the same SMmouthL Gordon parcel as before, but with the respiration recording as background (light blue), instead of the motion regressors. As before, since this is a mouth parcel, the activation should be strongest to the green "tongue" movement blocks. The respiration recording, events, and average activation are temporally aligned to the actual TR (x-axis), not shifted to account for the hemodynamic delay.

It is clear in both the MB8 and MB4 recordings that task activation is present, but it is perhaps a bit clearer with MB4. The little "jiggles" in the timecourse are present in all four runs, and look to be related to respiration, though not perfectly aligned with respiration. We're switching our ongoing MB8 imaging study to MB4 in the hopes of improving signal-to-noise, but respiration-related effects still look to be prominent in the MB4 BOLD, so dealing with them is an ongoing project.

respiration and movement: more experiences

Annika Linke, a postdoc at SDSU, contacted me, reporting that she's also seen oscillations in multiband acquisitions, with a Siemens Trio and Prisma, plus a GE Discovery. She kindly sent examples of her findings, and allowed me to share and describe a few of them here. More examples can be downloaded here and here.

Annika ran a multiband 8 sequence on a GE 3T Discovery MR750 scanner (UCSD CFMRI; TR = 800 msec, mb_factor 8, 32 channel head coil, 2 x 2 x 2 mm voxels; AP - PA encoding directions), and also saw prominent oscillations in the motion regressors, linked to respiration:

The subject was sleeping in this scan, and periodically stopped breathing. The movement in the motion regressors stopped and starts with the abnormal breathing periods, very similar to the traces from the purposeful breath-holding experiment we ran. I was also struck by the size of the oscillations in the movement regressors: somewhere between 0.5 and 1 mm, which neatly matches the size of the larger oscillations we've seen. Annika has results for an awake adult and toddlers, all of whom show oscillations (particularly in the A-P (y) and I-S (z) axes); see this pdf. These comparisons suggest the oscillation magnitude is not directly related to participant weight: the toddler and adult magnitudes are similar, though toddlers of course are smaller and have a faster respiration rate.

Here are some motion regressors Annika collected on a different scanner (Siemens 3T Prisma at Robarts Research Institute, Western University; CMRR release 10b, VD13D, 64 channel head coil, 3 x 3 x 3 mm voxels), at different MB factors. The oscillation is in all runs, though lowest amplitude with MB1.  

Finally, here's a set of motion regressors collected on a Trio at MB2 (ipat2 acceleration, 2 mm voxels, TR=1300 msec, AP/PA phase encoding directions, Robarts Research Institute, Western University). Each subplot is a run from a different person. All again have oscillations mostly in the y and z directions, though the magnitude is less than the MB8 plots above.

Annika's results make it clear that the magnitude of the oscillations in the motion regressors is not due to some weird fluke with our scanner, but rather some aspect of the multiband sequences (or some related factor, such as reconstruction, etc.); hopefully these additional examples will help us understand what's going on.

multiband acquisition sequence testing: respiration

This post is about the relationship between respiration and the movement regressors that we've found in our multiband acquisition sequence testing; see this post for an introduction; this post for examples of activation timecourses, and this post for examples from different people and tasks.

During our acquisition testing session, our participant did breath-holding for one set of MB8 runs, instead of the hand/foot/tongue movements, using the task cues to pace his breath-holding periods.

The light blue line is the respiration recording from a chest belt, with the six columns of the movement regressors overlaid, and the colored blocks at the bottom indicating task cues (as in the plots in the previous posts). Note that the time alignment is not totally perfect for the respiration trace: our system records respiration longer than the fMRI run, and I can't (yet) extract the run onset and offset signals. I aligned the traces using the breath-holding periods, but won't guarantee the timing is to-the-second perfect.

Regardless, the breath-holding periods are clear, and it's also clear that the oscillations in the movement regressors are related to the breathing. The start and stop of motion is also visible in a movie of the raw (before preprocessing) image, which can be seen here: the brain "jitters", then stops, then jitters, then stops (the movie is of the PA MOTORbreath run).

Here are two traces from the MOTOR task runs; plots for the rest of the runs can be downloaded here. The oscillations in the movement regressors is clearly closely linked to the respiration.

Interestingly, the biggest oscillation in the movement regressors here is split between several columns (dark blue, medium blue, light red), where before it was confined to one column (same participant as here); perhaps the head position or packing varied a bit?

Again, it's not new to note that fMRI is affected by breathing. What does seem different is the magnitude of the effect: these scans seem more affected, both in the motion regressors and (more importantly for us, the BOLD). For example, this last set of images shows the movement regressors for the MB0 (no multiband) runs from or test session, and a person from a different MB0 dataset ("BSF117"; a different scanner). A few blips are visible, but smaller. The MB8 and BSF117 examples below were downsampled to match the MB0 TR; note that these are the same MB8 movement regressors as above: after downsampling the oscillations no longer tightly match the respiration, but are still more prominent than the others.