help
help > RE: ART clarification
Mar 23, 2016 05:03 PM | Alfonso Nieto-Castanon - Boston University
RE: ART clarification
Hi Annchen,
Yes, your interpretation is perfectly correct, CONN will simply censor/scrub those time points but it will not change any of the raw data (e.g. it does not try to interpolate the "bad" scans BOLD signal). Censoring is actually implemented by entering additional covariates (one covariate per outlier scan) in the Denoising regression step to remove the influence of those outlier scans on the analyses (again similar to what most scrubbing techniques do).
Regarding your batch scripts, those look perfectly fine, I would only suggest to change the deriv field to:
batch.Preprocessing.confounds.deriv={0,0,0,1,0,1};
This field determines whether, for each confounding covariate, you would like to add, in addition to the raw covariate, its first- (or higher order) temporal derivatives as additional covariates. Typically it is a good idea to do so for the subject movement covariates (so that the BOLD signal covarying with the amount of scan-to-scan movement is also regressed out), but it is not necessary to do so for the outlier covariates (since the raw covariates in this case already remove entirely the influence of those identified outlier scans).
Hope this helps
Alfonso
Originally posted by Annchen Knodt:
Yes, your interpretation is perfectly correct, CONN will simply censor/scrub those time points but it will not change any of the raw data (e.g. it does not try to interpolate the "bad" scans BOLD signal). Censoring is actually implemented by entering additional covariates (one covariate per outlier scan) in the Denoising regression step to remove the influence of those outlier scans on the analyses (again similar to what most scrubbing techniques do).
Regarding your batch scripts, those look perfectly fine, I would only suggest to change the deriv field to:
batch.Preprocessing.confounds.deriv={0,0,0,1,0,1};
This field determines whether, for each confounding covariate, you would like to add, in addition to the raw covariate, its first- (or higher order) temporal derivatives as additional covariates. Typically it is a good idea to do so for the subject movement covariates (so that the BOLD signal covarying with the amount of scan-to-scan movement is also regressed out), but it is not necessary to do so for the outlier covariates (since the raw covariates in this case already remove entirely the influence of those identified outlier scans).
Hope this helps
Alfonso
Originally posted by Annchen Knodt:
Hi
Alfonso,
I found this post quite helpful in understanding the use of ART outliers in CONN, which seems to be the best way within the toolbox to get at doing the data "scrubbing" that seems to be essential these days. To be sure I do understand it correctly, when you say that CONN "disregard"s identified time points, that's essentially the same as censoring those time points as the scrubbers do, correct? (and you say that "the original time series is left intact" just to clarify that none of the raw data are changed by the toolbox?)
Also, I want to make sure that I am setting up my batch properly so that CONN does disregard those art outlier time points (if that's indeed what it does). So, I'm importing the 6 SPM realignment / movement parameters as "realignment_params" and the ART outliers (from art_regression_outliers_*.mat) as "art_outliers" using batch.Setup.covariates. Then, for the confounds step I have:
batch.Preprocessing.confounds.names={'GM','WM','CSF','realignment_params','art_outliers','Effect of rest'};
batch.Preprocessing.confounds.dimensions={1,3,3,6,[],[]};
batch.Preprocessing.confounds.deriv={0,0,0,0,[],1};
I'm not sure what to put for dimensions and deriv with the art outliers, so I left that blank. I don't know if this is right or if I totally understand what it's doing, since it seems like somehow those ART flags should get treated differently from the other covariates if they really result in the volumes to being "disregarded" rather than covaried for.
If you could let me know whether I'm on the right track, I'd really appreciate it! thanks!!
Annchen
Originally posted by Alfonso Nieto-Castanon:
I found this post quite helpful in understanding the use of ART outliers in CONN, which seems to be the best way within the toolbox to get at doing the data "scrubbing" that seems to be essential these days. To be sure I do understand it correctly, when you say that CONN "disregard"s identified time points, that's essentially the same as censoring those time points as the scrubbers do, correct? (and you say that "the original time series is left intact" just to clarify that none of the raw data are changed by the toolbox?)
Also, I want to make sure that I am setting up my batch properly so that CONN does disregard those art outlier time points (if that's indeed what it does). So, I'm importing the 6 SPM realignment / movement parameters as "realignment_params" and the ART outliers (from art_regression_outliers_*.mat) as "art_outliers" using batch.Setup.covariates. Then, for the confounds step I have:
batch.Preprocessing.confounds.names={'GM','WM','CSF','realignment_params','art_outliers','Effect of rest'};
batch.Preprocessing.confounds.dimensions={1,3,3,6,[],[]};
batch.Preprocessing.confounds.deriv={0,0,0,0,[],1};
I'm not sure what to put for dimensions and deriv with the art outliers, so I left that blank. I don't know if this is right or if I totally understand what it's doing, since it seems like somehow those ART flags should get treated differently from the other covariates if they really result in the volumes to being "disregarded" rather than covaried for.
If you could let me know whether I'm on the right track, I'd really appreciate it! thanks!!
Annchen
Originally posted by Alfonso Nieto-Castanon:
Hi Fred,
Yes, ART will simply identify the outlier scans and output that information into a art_*.mat file but it will not modify and/or remove anything in the original BOLD timeseries (which is left intact).
That new .mat file is simply input into CONN as a first-level covariate which, when entered as a Confounding effect during denoising, will make CONN disregard those identified time-points from any subsequent analysis (by adding a column for each outlier scan -dummy coding the offending timepoint- to the design matrix used during denoising to regress out all effects of no interest)
Hope this helps clarify
Alfonso
Originally posted by Fred Uquillas:
Yes, ART will simply identify the outlier scans and output that information into a art_*.mat file but it will not modify and/or remove anything in the original BOLD timeseries (which is left intact).
That new .mat file is simply input into CONN as a first-level covariate which, when entered as a Confounding effect during denoising, will make CONN disregard those identified time-points from any subsequent analysis (by adding a column for each outlier scan -dummy coding the offending timepoint- to the design matrix used during denoising to regress out all effects of no interest)
Hope this helps clarify
Alfonso
Originally posted by Fred Uquillas:
Hi NITRC community!
I just want to verify that the automatic preprocessing done via ART through the Conn toolbox is not physically removing any outlier time points from the time-series.
The word 'Srubbing' in this case is not synonymous with the 'Scrubbing' done with other artifact toolboxes such as ArtRepair, so that phrase has made us want to verify that no physical time points are being removed.
Thank you so much, and happy new year!!
Fred
I just want to verify that the automatic preprocessing done via ART through the Conn toolbox is not physically removing any outlier time points from the time-series.
The word 'Srubbing' in this case is not synonymous with the 'Scrubbing' done with other artifact toolboxes such as ArtRepair, so that phrase has made us want to verify that no physical time points are being removed.
Thank you so much, and happy new year!!
Fred
Threaded View
| Title | Author | Date |
|---|---|---|
| Fred Uquillas | Jan 15, 2015 | |
| Alfonso Nieto-Castanon | Jan 16, 2015 | |
| Fatima Sibaii | Apr 10, 2019 | |
| Annchen Knodt | Mar 22, 2016 | |
| Alfonso Nieto-Castanon | Mar 23, 2016 | |
| Annchen Knodt | Apr 15, 2016 | |
| Fred Uquillas | Jan 20, 2015 | |