Benjamin Kay (@ScienceBenKay) header
Benjamin Kay (@ScienceBenKay) profile picture

Benjamin Kay

@ScienceBenKay

Neuroimaging scientist at Washington University in St. Louis

 St. Louis, Missouri
Joined December 2022
  • Tweets20
  • Following30
  • Followers80
The good news is that frame censoring does a fantastic job of reducing false positive motion impact score. That means if you motion censor your data after standard processing then you are much less likely to get spurious results due to motion. Highly recommend trying it!
False positive motion impact scores are shown in (a) and false negative scores in (b). Scores below the gray line are not-significant (better). We split up traits into categories (demographic, physical, etc) to make things more interesting.

ALT False positive motion impact scores are shown in (a) and false negative scores in (b). Scores below the gray line are not-significant (better). We split up traits into categories (demographic, physical, etc) to make things more interesting.

2
1
10
10,305
We split our fMRI timeseries and compute separate connectivity matrices for the high- and low-motion halves. We subtract the low-motion matrix from the high-motion matrix. The behavioral connectivity is the same in both halves, so it cancels out, and only motion artifact is left.
We call our approach SHAMAN for Split-Half Analysis of Motion Associated Networks, but tweet us if you think of a catchier name! Here, for illustrative purposes, we can see SHAMAN working its magic on simulated data to generate the connectivity matrix of motion impact score in (e).

ALT We call our approach SHAMAN for Split-Half Analysis of Motion Associated Networks, but tweet us if you think of a catchier name! Here, for illustrative purposes, we can see SHAMAN working its magic on simulated data to generate the connectivity matrix of motion impact score in (e).

1
6
634
And the effect size of residual head motion on functional connectivity is huge -- larger than pretty much any behavioral variable. Yes folks, that's residual head motion, after the standard ABCD study data processing, and even after frame censoring at FD < 0.2 mm!
The black arrow is the effect size (single largest correlation) of head motion, and the gray bars are the effect sizes of all the behavioral traits.

ALT The black arrow is the effect size (single largest correlation) of head motion, and the gray bars are the effect sizes of all the behavioral traits.

1
2
8
1,012
Our motivation for creating the motion impact score came from the realization that lots of behavioral traits (87% in the ABCD study) are, surprisingly, correlated with head motion. Even seemingly unrelated traits like matrix reasoning score were correlated with motion.
Image from Benjamin Kay (@ScienceBenKay)
1
3
9
1,281
Ever wondered if your interesting brain-behavior correlation was affected by head motion, but were afraid to ask? We’ve created a motion impact score for detecting spurious brain-behavior associations. Excited to share my 1st preprint on Science Twitter! biorxiv.org/cgi/content/shor…
You can see how and where residual motion artifact is affecting your connectivity matrix. For example, body mass index (data from the ABCD study) had a mostly false positive motion impact score, indicating that motion inflated brain-behavior correlations. WISC-V matrix reasoning had a mostly false negative motion impact score, indicating that its brain-behavior correlations would actually be bigger if not for motion artifact!

ALT You can see how and where residual motion artifact is affecting your connectivity matrix. For example, body mass index (data from the ABCD study) had a mostly false positive motion impact score, indicating that motion inflated brain-behavior correlations. WISC-V matrix reasoning had a mostly false negative motion impact score, indicating that its brain-behavior correlations would actually be bigger if not for motion artifact!

4
49
6
166
36,564
Load more