help
help > RE: activation in conn.15d
Mar 9, 2020 05:03 AM | tiffanyk
RE: activation in conn.15d
Hi Alfonso,
I'm hoping you can give me some advice. I have some task-based data that I'd like to analyse, both for functional connectivity (PPI) and activation. Since the data is already preprocessed in conn19.b , I'd like to keep it in conn to analyse the activation data, if feasible (if this info is useful, I've saved the data as two different conn files, one for analyzing activation, the other for analysing FC, this way I can denoise the FC data, but not the activation data, and otherwise not confuse the two types of analyses).
My question is this: I have hypotheses about specific ROIs--if I follow the steps you've outlined below, but keep the ROIs that I'm interested in as sources in the first-level analysis, will this give me activation, limited to my ROIs? Alternatively, would it be better/more accurate to take my data out of conn, and analyze it using e.g., MarsBaR software?
Thanks,
Tiffany
Originally posted by Alfonso Nieto-Castanon:
I'm hoping you can give me some advice. I have some task-based data that I'd like to analyse, both for functional connectivity (PPI) and activation. Since the data is already preprocessed in conn19.b , I'd like to keep it in conn to analyse the activation data, if feasible (if this info is useful, I've saved the data as two different conn files, one for analyzing activation, the other for analysing FC, this way I can denoise the FC data, but not the activation data, and otherwise not confuse the two types of analyses).
My question is this: I have hypotheses about specific ROIs--if I follow the steps you've outlined below, but keep the ROIs that I'm interested in as sources in the first-level analysis, will this give me activation, limited to my ROIs? Alternatively, would it be better/more accurate to take my data out of conn, and analyze it using e.g., MarsBaR software?
Thanks,
Tiffany
Originally posted by Alfonso Nieto-Castanon:
Dear
Daniel,
Yes, you can (although typically connectivity analyses are orthogonal to activation-based analyses so you have to "trick" CONN a bit into performing those activation-based analyses instead). The way to do these analyses in a way that is most similar to the FSL or SPM activation-analysis output should be something like the following:
1) in Setup.Denoising remove all of the effects from the 'confounding effects' list, and enter in 'band-pass filter' the same filter that you used in FSL (e.g. a high-pass filter only)
2) in first-level analysis select as 'sources' all of the 'effect of task' effects plus your 'realignment' covariates (and any other covariates you may have entered into your first-level models in FSL; do not select there any of the ROI sources), then select the 'multivariate regression' measure, set the 'weighting' to 'none', press 'Done' and select there only the 'rest' condition before pressing 'Ok'
3) in second-level analysis select in the 'sources' list your task effect(s) and enter the desired contrast
Those results should be most similar to the activation-based analyses you have run in FSL. That said, there are going to be some subtle and some not-so-subtle differences, some arising from differences between FSL and SPM (e.g. FSL uses a different canonical hrf base than SPM) and some arising from CONN's focus on connectivity analyses and not really activation-based analyses (e.g. typically first-level activation-based models are fit separately for each session while the steps above will fit your task activation effects jointly across sessions). In addition in the steps above we are skipping all 'Denoising' steps (and enter some of those effects in your first-level analysis model instead) to mimic the way one would typically perform activation-based analyses, but there is a good reason why the Denoising step is necessary for connectivity-based analyses (e.g. residual motion and physiological effects in activation-based analyses do not typically bias your first-level task-related measures but they will bias any first-level connectivity-based measures so there is a considerably stronger focus in connectivity-based analyses to remove those effects using aCompcor, scrubbing, etc.)
Hope this helps
Alfonso
Originally posted by Daniel Yang:
Yes, you can (although typically connectivity analyses are orthogonal to activation-based analyses so you have to "trick" CONN a bit into performing those activation-based analyses instead). The way to do these analyses in a way that is most similar to the FSL or SPM activation-analysis output should be something like the following:
1) in Setup.Denoising remove all of the effects from the 'confounding effects' list, and enter in 'band-pass filter' the same filter that you used in FSL (e.g. a high-pass filter only)
2) in first-level analysis select as 'sources' all of the 'effect of task' effects plus your 'realignment' covariates (and any other covariates you may have entered into your first-level models in FSL; do not select there any of the ROI sources), then select the 'multivariate regression' measure, set the 'weighting' to 'none', press 'Done' and select there only the 'rest' condition before pressing 'Ok'
3) in second-level analysis select in the 'sources' list your task effect(s) and enter the desired contrast
Those results should be most similar to the activation-based analyses you have run in FSL. That said, there are going to be some subtle and some not-so-subtle differences, some arising from differences between FSL and SPM (e.g. FSL uses a different canonical hrf base than SPM) and some arising from CONN's focus on connectivity analyses and not really activation-based analyses (e.g. typically first-level activation-based models are fit separately for each session while the steps above will fit your task activation effects jointly across sessions). In addition in the steps above we are skipping all 'Denoising' steps (and enter some of those effects in your first-level analysis model instead) to mimic the way one would typically perform activation-based analyses, but there is a good reason why the Denoising step is necessary for connectivity-based analyses (e.g. residual motion and physiological effects in activation-based analyses do not typically bias your first-level task-related measures but they will bias any first-level connectivity-based measures so there is a considerably stronger focus in connectivity-based analyses to remove those effects using aCompcor, scrubbing, etc.)
Hope this helps
Alfonso
Originally posted by Daniel Yang:
Dear Dr. Nieto-Castanon,
I used to run FSL-based and am new to the conn.15d and SPM.
There are many steps involved in conn.15d such as the default steps in the denoising.
Is it possible to view activation (not functional-connectivity)-based results so I can see if I replicate the FSL results?
Many thanks!
Daniel Yang
I used to run FSL-based and am new to the conn.15d and SPM.
There are many steps involved in conn.15d such as the default steps in the denoising.
Is it possible to view activation (not functional-connectivity)-based results so I can see if I replicate the FSL results?
Many thanks!
Daniel Yang
Threaded View
| Title | Author | Date |
|---|---|---|
| Daniel Yang | Aug 10, 2015 | |
| Alfonso Nieto-Castanon | Aug 10, 2015 | |
| Athena Demertzi | Jan 28, 2021 | |
| Alfonso Nieto-Castanon | Jan 29, 2021 | |
| Luis Martinez Agulleiro | Mar 5, 2025 | |
| Clas Linnman | Aug 4, 2021 | |
| tiffanyk | Mar 9, 2020 | |
| Daniel Yang | Aug 10, 2015 | |
| Alfonso Nieto-Castanon | Aug 10, 2015 | |
| Daniel Yang | Aug 10, 2015 | |