help
help > RE: do number of ROIs affect FWE correction?
Feb 27, 2015 12:02 AM | Fred Uquillas
RE: do number of ROIs affect FWE correction?
Thank you very much Alfonso. We're all
very lucky to have you, this help forum, and the Conn
toolbox!
I'm hoping to look at the graph theory measures for a data set I have of 12 individuals with reactive aggression, and 17 controls. They don't differ in age, or education level, or handedness or race, but they do differ in IQ. So, I'm hoping to control for this by adding the IQ as a nuisance second level covariate.
They also don't differ in "mean displacement" (mm), which I computed by averaging the last column of the ART .mat covariate files per participant.
In terms of my graph, I keep changing my mind about the amount of ROIs/nodes to include for a whole-brain analysis. I'm running an analysis batch using the 130-something ROIs from the AAL atlas (currently finishing the Denoising step), and I have a template that is a 638 parcellation of the brain, compliments of Nicholas Crossley, that takes the different regions of the brain and makes them equally sized nodes (i.e., a large BA is broken down into smaller parts). The idea was to make the graph nodes as equally sized as possible so as to prevent the node size from influencing the graph measures.
But, there are papers out there using 90x90 (AAL and BA), 168x168 (Harvard-Oxford), and 84x84 (BA) (Conn validation paper) sized graphs. Furthermore, I'm afraid that correcting for multiple comparisons in a graph of 638x638 will kill my stats with a sample of N=29, and I've read that the bigger the graph/the more nodes, the less meaningful local metrics like local efficiency become : /
Any thoughts or possible suggestions here?
Once the 1st-level step is done, I'm hoping to use the ROI-ROI window to get to the graph theory explorer, and start looking at whole-brain measures.
Thank you for any suggestions you may have!!
All the best,
Fred
Originally posted by Alfonso Nieto-Castanon:
I'm hoping to look at the graph theory measures for a data set I have of 12 individuals with reactive aggression, and 17 controls. They don't differ in age, or education level, or handedness or race, but they do differ in IQ. So, I'm hoping to control for this by adding the IQ as a nuisance second level covariate.
They also don't differ in "mean displacement" (mm), which I computed by averaging the last column of the ART .mat covariate files per participant.
In terms of my graph, I keep changing my mind about the amount of ROIs/nodes to include for a whole-brain analysis. I'm running an analysis batch using the 130-something ROIs from the AAL atlas (currently finishing the Denoising step), and I have a template that is a 638 parcellation of the brain, compliments of Nicholas Crossley, that takes the different regions of the brain and makes them equally sized nodes (i.e., a large BA is broken down into smaller parts). The idea was to make the graph nodes as equally sized as possible so as to prevent the node size from influencing the graph measures.
But, there are papers out there using 90x90 (AAL and BA), 168x168 (Harvard-Oxford), and 84x84 (BA) (Conn validation paper) sized graphs. Furthermore, I'm afraid that correcting for multiple comparisons in a graph of 638x638 will kill my stats with a sample of N=29, and I've read that the bigger the graph/the more nodes, the less meaningful local metrics like local efficiency become : /
Any thoughts or possible suggestions here?
Once the 1st-level step is done, I'm hoping to use the ROI-ROI window to get to the graph theory explorer, and start looking at whole-brain measures.
Thank you for any suggestions you may have!!
All the best,
Fred
Originally posted by Alfonso Nieto-Castanon:
Hi Fred,
The number of source/seed ROIs entered at the first-level analysis tab does not affect the FWE stats later on. Those multiple comparison corrections are computed/applied at the second-level analysis step and they will depend on the number of potential target ROIs/voxels (but they do not correct for multiple analyses across potentially multiple source/seed ROIs).
For seed-to-voxel analyses, if you want to restrict the target voxels to a subset smaller than the entire brain (that is the default behavior for FWE corrections) you can do so in SPM by using an a priori mask and/or by using small-volume corrections. For ROI-to-ROI analyses, if you want to restrict the target ROIs to a subset smaller than the entire set of ROIs defined in your CONN project you may do so by selecting the "Analysis results: targets are ..." pulldown menu -in the main CONN second-level results tab GUI- or the "Connectivity matrix: targets are ..." pulldown menu -in the ROI-to-ROI results explorer window-)
Hope this helps (and let me know more details about your specific analyses if you would like me to further clarify how this applies to your case)
Alfonso
Originally posted by Fred Uquillas:
The number of source/seed ROIs entered at the first-level analysis tab does not affect the FWE stats later on. Those multiple comparison corrections are computed/applied at the second-level analysis step and they will depend on the number of potential target ROIs/voxels (but they do not correct for multiple analyses across potentially multiple source/seed ROIs).
For seed-to-voxel analyses, if you want to restrict the target voxels to a subset smaller than the entire brain (that is the default behavior for FWE corrections) you can do so in SPM by using an a priori mask and/or by using small-volume corrections. For ROI-to-ROI analyses, if you want to restrict the target ROIs to a subset smaller than the entire set of ROIs defined in your CONN project you may do so by selecting the "Analysis results: targets are ..." pulldown menu -in the main CONN second-level results tab GUI- or the "Connectivity matrix: targets are ..." pulldown menu -in the ROI-to-ROI results explorer window-)
Hope this helps (and let me know more details about your specific analyses if you would like me to further clarify how this applies to your case)
Alfonso
Originally posted by Fred Uquillas:
Hi Conn community!
Do the number of ROIs entered at the first-level affect the FWE stats later on?
I ran a batch using 638 ROIs, and one using just 2 ROIs, and the FWE stats for two specific ROIs (R and L amygdala) come out the same at the second-level window (and unfortunately very very non-significant (0.998).
Fred
Do the number of ROIs entered at the first-level affect the FWE stats later on?
I ran a batch using 638 ROIs, and one using just 2 ROIs, and the FWE stats for two specific ROIs (R and L amygdala) come out the same at the second-level window (and unfortunately very very non-significant (0.998).
Fred
Threaded View
| Title | Author | Date |
|---|---|---|
| Fred Uquillas | Feb 19, 2015 | |
| Alfonso Nieto-Castanon | Feb 25, 2015 | |
| Fred Uquillas | Feb 27, 2015 | |
| Fred Uquillas | Feb 23, 2015 | |