Our methods are detailed in Nature Communications, and Matlab code for computing motion impact score is available on GitHub. Try it on your favorite brain-behavior association and let us know what you find!
doi.org/10.1038/s414...
github.com/DosenbachGre...

The good news is that frame censoring does a fantastic job of reducing motion overestimation scores. That means if you motion censor your data after standard processing then you are much less likely to get spurious results due to motion.

Over 40% of brain-behavior effects were likely overestimated due to motion. There was no correlation between a trait’s a priori correlation with motion and its motion overestimation score. Calculating the motion impact score was necessary to discover the problematic brain-behavior effects!

The principle behind the motion impact score is simple! We know head motion varies from second-to-second while behavioral traits are stable over time. We just compare high- and low-motion portions of the fMRI scans. The effect of behavior is the same and cancels out, leaving the motion impact.

We realized that seemingly-unrelated behavioral traits like matrix reasoning ability are, surprisingly, correlated with in-scanner head motion. Could brain-behavior associations of more motion-correlated traits be more impacted by residual head motion? We developed a tool to find out!

Unfortunately, even with fancy processing pipelines and motion censoring, the effect size of residual head motion on functional connectivity can still be larger than the effect of many behavioral traits.

That's an intriguing question! We analyzed data from the ABCD n-back task as if it were rest and found essentially the same stimulant-related differences in FC as we did for "pure" rest, see Supplemental Figure 4.

Touché... although I've never once had a program officer tell me that I've been publishing too much 😅

I appreciate your offer of help! We're always looking for ways to extract the most quality out of our precision studies, so I may need to take you up on that at some point!

This is not meant to be directed at @ckorponay.bsky.social, but I take issue with reviewers in general asking for large data sets to be reprocessed using their favorite pipeline without strong theoretical justification. Reprocessing 40,000+ scans takes a lot of taxpayer $$$!

Let's suppose, for the sake of argument, that our stimulant-related differences in FC are due entirely to decreased sLFO inflation in children taking stimulants. If sLFO inflation is driven by diminishing arousal, this would still support our central hypothesis that stimulants increase arousal.

Snark aside, we've appreciated non-stationarity of fMRI BOLD going back to the early days of Friston and Volterra kernels. Every denoising pipeline has grappled with this to some extent, and I'm excited to see the insights from @ckorponay.bsky.social's RIPTiDe get incorporated into future pipelines.

We're confident stimulants modulate arousal, but how much of what we see is "pure" neuronal signal and how much is filtered through physiology is an open question for all fMRI studies. It's terrific having tools like RIPTiDe, and your paper raises awareness about collecting physiological covariates!

Yes, loved the fresh perspective of your Nat Hum Behav article, more and more we're realizing how much arousal and associated physiological changes affect rs-fMRI. I'm sure you know Ryan Raut's work doi.org/10.1126/scia... and Catie Chang's recent Nat Neurosci paper doi.org/10.1038/s415...

That's a really interesting hypothesis! We think norepinephrine drives the arousal boost and dopamine drives the salience boost. Whether high levels of dopamine cause psychosis through the salience network or through other brain targets is an open question, see Winton-Brown: doi.org/10.1016/j.ti...

This is the point of our story: the biggest differences in connectivity are in arousal and salience networks (to which PMN is closely yoked). Contrary to what people expect, the differences in "top down" attention networks are relatively smaller (FPN) or basically not existent (DAN).

You are reading those matrices correctly! The decreased within-network FC in primary systems is what we see in subjects who are more alert/better rested. The between-network differences are significant for SAL and PMN. There may be a trend for DMN and FPN, but it didn't reach significance.

You're right, the link between amphetamines and sleep dates back to Gordon Alles 1927. The link with salience goes back to at least Robbins (Nature) 1976. That hasn't stopped claims that MPH acts directly on attention. Our study shows the greatest effects are on arousal and salience, not attention.