help
help > RE: QC_timeseries not in confounds
Apr 10, 2019 12:04 PM | Alfonso Nieto-Castanon - Boston University
RE: QC_timeseries not in confounds
Hi Fatima,
Yes, any first-level covariate named QC_* is not included by default as part of the list of potential confounding effects during denoising, precisely so that CONN can create a few covariates which are mainly useful for quality-control purposes without the risk of modifying the default recommended set of denoising steps. In particular, the QC_timeseries covariate contains the raw (unthresholded) global-signal change and framewise displacement timeseries, which are used in CONN as part of the QA plots step (and they can also be used by users to generate new scrubbing covariates by selecting different outlier threshold values). Yet I would not recommend (typically) to enter this QA_timeseries covariate in the list of potential confounding effects during denoising, mostly because the global signal change covariate contains a mixture of noise and true neural effects, so regressing it out would introduce biases similar to those introduced by global signal regression.
Hope this helps
Alfonso
Originally posted by Fatima Sibaii:
Yes, any first-level covariate named QC_* is not included by default as part of the list of potential confounding effects during denoising, precisely so that CONN can create a few covariates which are mainly useful for quality-control purposes without the risk of modifying the default recommended set of denoising steps. In particular, the QC_timeseries covariate contains the raw (unthresholded) global-signal change and framewise displacement timeseries, which are used in CONN as part of the QA plots step (and they can also be used by users to generate new scrubbing covariates by selecting different outlier threshold values). Yet I would not recommend (typically) to enter this QA_timeseries covariate in the list of potential confounding effects during denoising, mostly because the global signal change covariate contains a mixture of noise and true neural effects, so regressing it out would introduce biases similar to those introduced by global signal regression.
Hope this helps
Alfonso
Originally posted by Fatima Sibaii:
Hey
I really appreciate this forum, I have been able to find answers to questions I've faced several times, and I think that has really helped make CONN a successful and relevant toolbox.
I'm running a task-related connectivity analysis using CONN 18.a. I have noticed that QC_timeseries; which is a first level covariate, is not regressed out of BOLD as a confound in the first level analysis (under the denoising tab, choosing confounds). This surprised me since I thought by default all first level covariates are regressed out in this stage. Since this is the output of ART which includes the raw motion and global signal change timeseries, shouldn't be regressed out? Is it left in the effects because it is regressed out before in the preprocessing step?
Thanks
Fatima
I really appreciate this forum, I have been able to find answers to questions I've faced several times, and I think that has really helped make CONN a successful and relevant toolbox.
I'm running a task-related connectivity analysis using CONN 18.a. I have noticed that QC_timeseries; which is a first level covariate, is not regressed out of BOLD as a confound in the first level analysis (under the denoising tab, choosing confounds). This surprised me since I thought by default all first level covariates are regressed out in this stage. Since this is the output of ART which includes the raw motion and global signal change timeseries, shouldn't be regressed out? Is it left in the effects because it is regressed out before in the preprocessing step?
Thanks
Fatima
Threaded View
| Title | Author | Date |
|---|---|---|
| Fatima Sibaii | Apr 8, 2019 | |
| Alfonso Nieto-Castanon | Apr 10, 2019 | |
| Fatima Sibaii | Apr 11, 2019 | |