help
help > RE: Reviewer not familiar with CONN?
Oct 9, 2015 06:10 PM | Alfonso Nieto-Castanon - Boston University
RE: Reviewer not familiar with CONN?
Hi Jeff and Fran,
Regarding the global signal regression question, I am typically in the exact opposite side of this where I would ask authors for a justification if they DID use GSR instead of one of the many current and arguably better/less-problematic alternatives (CompCor, FIX, AROMA, retroicor, etc.) In addition to your response I would probably just point the reviewer towards a number of publications discussing the issues with GSR (e.g. Murphy et al. 2009, Schölvinck et al. 2010, Chai et al. 2012, Saad et al. 2012, Gotts et al. 2013). Most reviewers should be satisfied if you simply justify your choice even if they do not share your perspective (and if you feel that an additional analyses might help I would suggest to simply enter the gray-matter ROI signal in CONN as an additional confouding effect during denoising and see whether your main results still replicate; even though I personally do not love this approach because of the many confounding effects that this may incurr in)
Regarding the subject-motion response, CONN uses aCompCor (not tCompCor), mainly because tCompCor can just as easily pick-up neural signals from gray matter voxels which one does not want to include as denoising regressors (for the same reasons as in GSR). If you have used ART as part of preprocessing (e.g. in the default CONN preprocessing pipeline) I would mention that as well in the response, since that is one of the main protections agains motion outliers. All considered, it is not a bad idea to perform some additional post-hoc analyses to make sure that potential differences in subject-motion across groups are not influencing your results (and to that end I would probably use some proxy measure of subject-motion degree as an additional second-level covariate and then: a) test whether there are group differences in this measure of subject motion -e.g. in Tools.Calculator-; and b) add this covariate as an additional subject-effect in your main second-level analyses to see whether your results replicate when controlling for the degree of subject-motion)
Last, regarding the "effect of condition", that is simply the regressor associated with your task conditions (e.g. hrf-convolved task blocks), it does not correct for condition-specific global signal effects, but it can remove some portion of the global signal variability that correlates with your task presentation (e.g. coactivation of multiple regions in response to the task).
Hope this helps
Alfonso
Originally posted by Jeff Browndyke:
Regarding the global signal regression question, I am typically in the exact opposite side of this where I would ask authors for a justification if they DID use GSR instead of one of the many current and arguably better/less-problematic alternatives (CompCor, FIX, AROMA, retroicor, etc.) In addition to your response I would probably just point the reviewer towards a number of publications discussing the issues with GSR (e.g. Murphy et al. 2009, Schölvinck et al. 2010, Chai et al. 2012, Saad et al. 2012, Gotts et al. 2013). Most reviewers should be satisfied if you simply justify your choice even if they do not share your perspective (and if you feel that an additional analyses might help I would suggest to simply enter the gray-matter ROI signal in CONN as an additional confouding effect during denoising and see whether your main results still replicate; even though I personally do not love this approach because of the many confounding effects that this may incurr in)
Regarding the subject-motion response, CONN uses aCompCor (not tCompCor), mainly because tCompCor can just as easily pick-up neural signals from gray matter voxels which one does not want to include as denoising regressors (for the same reasons as in GSR). If you have used ART as part of preprocessing (e.g. in the default CONN preprocessing pipeline) I would mention that as well in the response, since that is one of the main protections agains motion outliers. All considered, it is not a bad idea to perform some additional post-hoc analyses to make sure that potential differences in subject-motion across groups are not influencing your results (and to that end I would probably use some proxy measure of subject-motion degree as an additional second-level covariate and then: a) test whether there are group differences in this measure of subject motion -e.g. in Tools.Calculator-; and b) add this covariate as an additional subject-effect in your main second-level analyses to see whether your results replicate when controlling for the degree of subject-motion)
Last, regarding the "effect of condition", that is simply the regressor associated with your task conditions (e.g. hrf-convolved task blocks), it does not correct for condition-specific global signal effects, but it can remove some portion of the global signal variability that correlates with your task presentation (e.g. coactivation of multiple regions in response to the task).
Hope this helps
Alfonso
Originally posted by Jeff Browndyke:
The same question was posed to me about the
global signal regression vs. CompCor. I like your response,
but I suspect it may not satisfy them as it has become hardened in
some that global signal regression is always required. One
thing we did was to collect the global signal intensity values from
our groups during CONN preprocessing (output in MATLAB) and then
compared these values to see if there was a significant difference
between groups and over time. This helps buttress that there
aren't any statistically significant differences between groups,
but I suspect even this may not placate some reviewers.
One question I have is what is the "Effect of Condition" regressor controlling for during the denoising process? Is this a condition specific global signal regression variable?
Jeff
One question I have is what is the "Effect of Condition" regressor controlling for during the denoising process? Is this a condition specific global signal regression variable?
Jeff
Threaded View
| Title | Author | Date |
|---|---|---|
| Fran | Oct 8, 2015 | |
| Jeff Browndyke | Oct 8, 2015 | |
| Alfonso Nieto-Castanon | Oct 9, 2015 | |
| Fran | Oct 21, 2015 | |
| Fran | Oct 9, 2015 | |