help > How is scrubbing implemented in conn?
Showing 1-4 of 4 posts
Mar 22, 2017 10:03 AM | Sascha Froelich
How is scrubbing implemented in conn?
Dear all,
can anyone tell me how scrubbing is implemented in conn? As far as I understand, scrubbing refers to the removal of frames and their replacement by interpolated data. Based on which measure are frames scrubbed? And how is said interpolation computed? Does scrubbing use the spectral decomposition of the remaining uncensored data to interpolate the "replacement data"?
Where in the preprocessing-folder can I see which volumes were scrubbed?
EDIT: I realize that scrubbing is based on the ART-toolbox (https://www.nitrc.org/projects/artifact_...). However, I cannot seem to find out what method the toolbox uses for scrubbing. The manual does not elaborate on that.
Thanks!
Cheers,
Sascha
can anyone tell me how scrubbing is implemented in conn? As far as I understand, scrubbing refers to the removal of frames and their replacement by interpolated data. Based on which measure are frames scrubbed? And how is said interpolation computed? Does scrubbing use the spectral decomposition of the remaining uncensored data to interpolate the "replacement data"?
Where in the preprocessing-folder can I see which volumes were scrubbed?
EDIT: I realize that scrubbing is based on the ART-toolbox (https://www.nitrc.org/projects/artifact_...). However, I cannot seem to find out what method the toolbox uses for scrubbing. The manual does not elaborate on that.
Thanks!
Cheers,
Sascha
Mar 22, 2017 01:03 PM | Jeff Browndyke
RE: How is scrubbing implemented in conn?
Hi, Sascha.
My understanding is that CONN does not "scrub" data in the traditional sense. What you are describing sounds like frame-wise displacement to correct for bad data points. CONN, though use of ART detect, actually accounts for the bad data points and then includes those bad data point time courses and movement time courses as nuisance regressors during the denoising procedure. Those bad data points via CONN processing are probably better described as being "censored" from the data as opposed to "scrubbed." It is an important distinction because data is not being inserted or interpolated with CONN, which makes decisions about things like the maximal amount of total signal noise allowable important or the ratio of good to bad time points allowable for a subject.
Hope this helps,
Jeff
My understanding is that CONN does not "scrub" data in the traditional sense. What you are describing sounds like frame-wise displacement to correct for bad data points. CONN, though use of ART detect, actually accounts for the bad data points and then includes those bad data point time courses and movement time courses as nuisance regressors during the denoising procedure. Those bad data points via CONN processing are probably better described as being "censored" from the data as opposed to "scrubbed." It is an important distinction because data is not being inserted or interpolated with CONN, which makes decisions about things like the maximal amount of total signal noise allowable important or the ratio of good to bad time points allowable for a subject.
Hope this helps,
Jeff
Mar 23, 2017 09:03 AM | Sascha Froelich
RE: How is scrubbing implemented in conn?
Dear Jeff,
thanks!
What I thought is that CONN completely removes volumes of high motion and then replaces them with interpolated data. But apparently this is not the case. So if I understood you correctly, CONN does not remove these volumes, but uses the bad time points to create regressors for nuisance regression, is that correct?
However, I am still a bit confused. I thought the terms "censoring" and "scrubbing" both describe the same procedure, so what is the difference?
Cheers,
Sascha
thanks!
What I thought is that CONN completely removes volumes of high motion and then replaces them with interpolated data. But apparently this is not the case. So if I understood you correctly, CONN does not remove these volumes, but uses the bad time points to create regressors for nuisance regression, is that correct?
However, I am still a bit confused. I thought the terms "censoring" and "scrubbing" both describe the same procedure, so what is the difference?
Cheers,
Sascha
Apr 15, 2019 07:04 PM | Mary Newsome
RE: How is scrubbing implemented in conn?
Originally posted by Sascha Froelich:
Dear Jeff,
thanks!
What I thought is that CONN completely removes volumes of high motion and then replaces them with interpolated data. But apparently this is not the case. So if I understood you correctly, CONN does not remove these volumes, but uses the bad time points to create regressors for nuisance regression, is that correct?
However, I am still a bit confused. I thought the terms "censoring" and "scrubbing" both describe the same procedure, so what is the difference?
Cheers,
Sascha
thanks!
What I thought is that CONN completely removes volumes of high motion and then replaces them with interpolated data. But apparently this is not the case. So if I understood you correctly, CONN does not remove these volumes, but uses the bad time points to create regressors for nuisance regression, is that correct?
However, I am still a bit confused. I thought the terms "censoring" and "scrubbing" both describe the same procedure, so what is the difference?
Cheers,
Sascha
Hi Sascha,
I am currently learning more about this myself.
My general understanding is that Conn uses ART to identify the
outlier scans based on parameters you select and then uses those
scans as nuisance regressors in the first-level analysis. ART also
produces a mask of the outliers that can be used as an explicit
mask to avoid any influence of the outliers in the first level
analysis. (I got this information from the ART code included in the
ART download. It is attached) I don't see that ART does any
interpolation.
In Spiegel et al (2014), scrubbing means the
same thing as censoring, and it means "applying temporal masks to
remove high motion volumes", and "In motion censoring, volumes
in which head motion exceeded a threshold (a)re withheld from
GLM estimation." It sounds like Conn (through ART) does both motion
regression (if you enter the regressors into the analysis) and
motion censoring, i.e., applies a temporal mask. However, to make
the temporal mask, in the commented out part of the file with the
code, it states "(in SPM you will also need to modify the defaults
in order to skip the implicit masking operation, e.g. set
defaults.mask.thresh = -inf)
Of course any or all of this could be wrong. I
would appreciate any comments!
Best,
Mary