help
help > RE: 2nd level analysis - ROIs lack symmetry
Jun 14, 2016 04:06 PM | Alfonso Nieto-Castanon - Boston University
RE: 2nd level analysis - ROIs lack symmetry
Dear Rob,
When using bivariate-correlation weighted-GLM analyses the connectivity measures will be symmetrical (the connectivity/correlation between ROI-A and ROI-B will be the same as that between ROI-B and ROI-A). The same does not hold true if one uses bivariate-regression, multivariate-regresssion, semipartial-correlation, and/or PPI/gPPI analyses. In all of those cases the connectivity values will be non-symmetrical because the underlying models/measures are also non-symmetrical. In the case of gPPI, for example, when using ROI-A as source and ROI-B as target the underlying model is:
X_B = X_A*k1 + H*k2 + (X_A*H)*k3 + Noise_B
where X_A and X_B are the BOLD signal within ROIs A and B, respectively, H are your psychological factor(s), and k1-k3 are the model parameters estimated (k3 is the psychological modulation factor of interest that will be entered in your second-level analyses). In contrast, when using ROI-B as source and ROI-B as target the underlying model is:
X_A = X_B*j1 + H*j2 + (X_B*H)*j3 + Noise_A
and the factor j3 of interest will typically be different than the k3 factor estimated in the previous model. Conceptually, in a gPPI model the modulatory factor of interest represents the change in effective connectivity between the "source" region and the "target" region in the presence/absence of the psychological factor (and contrasting with functional connectivity, effective connectivity is a directional construct, as it measures the effect that one region exerts on another). You may want to check the SPM-list and/or the gPPI forum here at NITRC for additional resources on this topic, but as far as I know there are no clear guidelines beyond the above to best interpret differences between ROI-A to ROI-B vs. ROI-B to ROI-A effects in gPPI analyses.
Hope this helps
Alfonso
Originally posted by Rob McCutcheon:
When using bivariate-correlation weighted-GLM analyses the connectivity measures will be symmetrical (the connectivity/correlation between ROI-A and ROI-B will be the same as that between ROI-B and ROI-A). The same does not hold true if one uses bivariate-regression, multivariate-regresssion, semipartial-correlation, and/or PPI/gPPI analyses. In all of those cases the connectivity values will be non-symmetrical because the underlying models/measures are also non-symmetrical. In the case of gPPI, for example, when using ROI-A as source and ROI-B as target the underlying model is:
X_B = X_A*k1 + H*k2 + (X_A*H)*k3 + Noise_B
where X_A and X_B are the BOLD signal within ROIs A and B, respectively, H are your psychological factor(s), and k1-k3 are the model parameters estimated (k3 is the psychological modulation factor of interest that will be entered in your second-level analyses). In contrast, when using ROI-B as source and ROI-B as target the underlying model is:
X_A = X_B*j1 + H*j2 + (X_B*H)*j3 + Noise_A
and the factor j3 of interest will typically be different than the k3 factor estimated in the previous model. Conceptually, in a gPPI model the modulatory factor of interest represents the change in effective connectivity between the "source" region and the "target" region in the presence/absence of the psychological factor (and contrasting with functional connectivity, effective connectivity is a directional construct, as it measures the effect that one region exerts on another). You may want to check the SPM-list and/or the gPPI forum here at NITRC for additional resources on this topic, but as far as I know there are no clear guidelines beyond the above to best interpret differences between ROI-A to ROI-B vs. ROI-B to ROI-A effects in gPPI analyses.
Hope this helps
Alfonso
Originally posted by Rob McCutcheon:
Dear Alfonso,
I was hoping you might be able to help me with a query regarding second-level analysis in CONN.
I have conducted a gPPI analysis (bivariate regression). I have a patient group, and a control group, and a block design task. I am aiming to look at differences in connectivity between the patient and control group during the task.
My question is this:
When I select ROI-A i obtain a beta and a p value for the connectivity with ROI-B. However, when I select ROI-B the values are different. I have checked the forum and one explanation I read explains why FDR corrected p values might be different. However I am seeing differences in the uncorrected values.
Any explanation would be most helpful!
Many Thanks,
Rob
I was hoping you might be able to help me with a query regarding second-level analysis in CONN.
I have conducted a gPPI analysis (bivariate regression). I have a patient group, and a control group, and a block design task. I am aiming to look at differences in connectivity between the patient and control group during the task.
My question is this:
When I select ROI-A i obtain a beta and a p value for the connectivity with ROI-B. However, when I select ROI-B the values are different. I have checked the forum and one explanation I read explains why FDR corrected p values might be different. However I am seeing differences in the uncorrected values.
Any explanation would be most helpful!
Many Thanks,
Rob
Threaded View
| Title | Author | Date |
|---|---|---|
| Rob McCutcheon | Jun 14, 2016 | |
| Alfonso Nieto-Castanon | Jun 14, 2016 | |
| Rob McCutcheon | Jun 14, 2016 | |
| Alfonso Nieto-Castanon | Jun 15, 2016 | |
| Rob McCutcheon | Jun 15, 2016 | |