help
help > RE: Task-based confounds
Sep 9, 2015 02:09 AM | Alfonso Nieto-Castanon - Boston University
RE: Task-based confounds
Hi John,
For task-designs I typically recommend leaving the main task effects as confounds as well. In connectivity analyses the connectivity task-effects (differences in connectivity between tasks) are typically orthogonal to the "activation" task effects (differences in average BOLD signal between tasks), and the latter are the ones that you would be removing when leaving the "effect of " effects in the list of confounds during Denoising. To clarify a bit, if the BOLD signal response to, for example, a block design, was instantaneous and constant during your block (e.g. a rectangular block response), then the correlation measures within your blocks would already be unaffected by the average BOLD signal within the same blocks (because the correlation between two timeseries is independent of the mean value of the two signals), and the connectivity and "activation" measures would be exactly orthogonal. Unfortunately the BOLD signal response to a block is not like that but it has a few seconds of ramping effects at the beginning and end of the block (modeled by the convolution of your block design with the hemodynamic response function). Because of this, failing to correct for those ramping effects can create spurious correlations between two areas that may simply reflect a common coactivation in response to the task in these areas. On way to correct for this effect is to remove the main BOLD-signal task-related effect before computing the within-block correlation measures, which is what the denoising step will do if you leave the "effect of " effects in the list of confounds.
Hope this helps
Alfonso
Originally posted by John Hutton:
For task-designs I typically recommend leaving the main task effects as confounds as well. In connectivity analyses the connectivity task-effects (differences in connectivity between tasks) are typically orthogonal to the "activation" task effects (differences in average BOLD signal between tasks), and the latter are the ones that you would be removing when leaving the "effect of " effects in the list of confounds during Denoising. To clarify a bit, if the BOLD signal response to, for example, a block design, was instantaneous and constant during your block (e.g. a rectangular block response), then the correlation measures within your blocks would already be unaffected by the average BOLD signal within the same blocks (because the correlation between two timeseries is independent of the mean value of the two signals), and the connectivity and "activation" measures would be exactly orthogonal. Unfortunately the BOLD signal response to a block is not like that but it has a few seconds of ramping effects at the beginning and end of the block (modeled by the convolution of your block design with the hemodynamic response function). Because of this, failing to correct for those ramping effects can create spurious correlations between two areas that may simply reflect a common coactivation in response to the task in these areas. On way to correct for this effect is to remove the main BOLD-signal task-related effect before computing the within-block correlation measures, which is what the denoising step will do if you leave the "effect of " effects in the list of confounds.
Hope this helps
Alfonso
Originally posted by John Hutton:
Hi Alfonso:
I have been wrestling with a task-based (2 states) analysis, which seems to be working better with the latest version of CONN. One basic question (forgive if too basic): in the Denoising step, should I remove my 2 states from the list of confounds? It seems that I should, as they are variables of interest, as opposed to White Matter, CSF, and motion, which are not. Not sure why CONN by default lists them there.
Thank you and best,
John
I have been wrestling with a task-based (2 states) analysis, which seems to be working better with the latest version of CONN. One basic question (forgive if too basic): in the Denoising step, should I remove my 2 states from the list of confounds? It seems that I should, as they are variables of interest, as opposed to White Matter, CSF, and motion, which are not. Not sure why CONN by default lists them there.
Thank you and best,
John
Threaded View
| Title | Author | Date |
|---|---|---|
| John Hutton | Sep 8, 2015 | |
| Alfonso Nieto-Castanon | Sep 9, 2015 | |
| John Hutton | Sep 9, 2015 | |