Scrubbing in CONN toolbox

Hi Experts,

Anyone knows how the “scrubbing” is achieved in CONN toolbox’s Denoising?



Scrubbing generally is defined as removing volumes that are motion contaminated. The criteria for defining an outlier are usually based on head motion (framewise displacement) and/or change in bold signal attributed to head motion (dvars). Your threshold should be chosen based on past literature and considerations from your sample (how many subjects do you want In denoising, would subject age or clinical diagnosis lead to more inherent motion, etc.). A bit more about implementation: in CONN, each outlier gets its denoising vector to regress out, which is all 0s except for a 1 at the index of the bad outlier. So for example if you had two outliers, one at the beginning and one at the end of the scan, your two vectors would be [1, 0, 0…0] and [0, 0, … 0, 1]. These are regressed out at the denoising state of CONN. You can make these vectors yourself from information like the confounds.tsv from fMRIPrep, or you can have CONN create these for you with the built in ART toolbox. The ART toolbox will look for outliers based on FD and dvars as I said earlier. The toolbox has different settings of aggressiveness in denoising, which as suggested earlier should be chosen in consideration with your sample. Might have over answered a bit, but happy to explain more if something I said didn’t make sense or the real answer you were looking for got lost in the weeds.


Thanks for explaining! Steven. Super clear!
But i am not sure if CONN toolbox has already regressed out the outliers after the “Denoising” step or this will be done in first level analysis in CONN ?

No problem! Regression of scrubbed outliers happens during the denoising step, so before defining first-level analyses.

As an aside, you may also want to define subjects that are high motion (perhaps too many volumes were scrubbed to have enough signal to work with meaningfully), and if you want to exclude them, that would be a 2nd-level covariate that you define in setup, and can look at in your 2nd level analyses.


Thanks! Steven. Well explained.