help
help > RE: graph theory measures questions
Jun 1, 2017 02:06 AM | Alfonso Nieto-Castanon - Boston University
RE: graph theory measures questions
Hi Alain,
Regarding (1) yes, that makes sense. If you want to do that within CONN you may simply enter your design (e.g. select 'AllSubjects' and your 'Behav' covariate and enter a between-subjects contrast [0 1]) in the second-level results tab before clicking on 'graph theory' and the corresponding stats will report the association between the graph theory measures and your behavioral variable directly (rather than having to do this manually from the exported data).
Regarding (2), just in the same way that you would do for other connectivity analyses simply select the two group covariates (e.g. 'young' and 'old') and enter a [1 -1] between-subject contrast, then click on 'graph theory' and define there your desired ROIs and graph-forming thresholds, and that will perform a two-sample t-test comparing the network metrics between the two groups (and apply a FDR correction across the multiple ROIs)
Regarding (3), yes, that happens for AveragePathLength and ClusteringCoefficient measures, because these measures are derived from the path-length between two nodes/ROIs, and the path-length between two nodes that are not connected by any path takes the value inf. When that happens the exported data simply reports NaN for the corresponding measures. When using the approach in (1) and (2) above, those cases will be automatically disregarded (treated as missing values) by CONN in your second-level analyses, and you will notice that because the dof's reported for will be smaller than expected (only for the ROIs that are involved). Other measures, like global/local efficiency, are not affected by this condition because inverse-path-length is simply zero in these cases, so those can be incorporated normally into the analyses.
Hope this helps
Alfonso
Originally posted by Alain Imaging:
Regarding (1) yes, that makes sense. If you want to do that within CONN you may simply enter your design (e.g. select 'AllSubjects' and your 'Behav' covariate and enter a between-subjects contrast [0 1]) in the second-level results tab before clicking on 'graph theory' and the corresponding stats will report the association between the graph theory measures and your behavioral variable directly (rather than having to do this manually from the exported data).
Regarding (2), just in the same way that you would do for other connectivity analyses simply select the two group covariates (e.g. 'young' and 'old') and enter a [1 -1] between-subject contrast, then click on 'graph theory' and define there your desired ROIs and graph-forming thresholds, and that will perform a two-sample t-test comparing the network metrics between the two groups (and apply a FDR correction across the multiple ROIs)
Regarding (3), yes, that happens for AveragePathLength and ClusteringCoefficient measures, because these measures are derived from the path-length between two nodes/ROIs, and the path-length between two nodes that are not connected by any path takes the value inf. When that happens the exported data simply reports NaN for the corresponding measures. When using the approach in (1) and (2) above, those cases will be automatically disregarded (treated as missing values) by CONN in your second-level analyses, and you will notice that because the dof's reported for will be smaller than expected (only for the ROIs that are involved). Other measures, like global/local efficiency, are not affected by this condition because inverse-path-length is simply zero in these cases, so those can be incorporated normally into the analyses.
Hope this helps
Alfonso
Originally posted by Alain Imaging:
Hi evryone and Alfonso,
i would like to use the graph theory analysis in conn in order to answer some of my research questions.
In particular, I am trying to understand if a certain network of brain regions change with age and if its relationship with cognition change with age.
To do that I had:
1) selected my ROIs
2) computed the graph theory measures using correlation coefficient and tresholding theedges to .15
3) exported the data.
In the exported data I have one line per subject with the measures for the network overall and for each node.
I have some questions about this:
1) does it make any sense to correlate network-wise and single node measures with behavioral data ?
2) how would I compare networks for different age groups ? Should I use some multivariate analysis (e.g. MANOVA including all the network measures as DV and the group as IV) ?
3) for some of the subjects some measures are missing (NaN), why does this happen ? Can I just remove these subjects ?
Thanks a lot for any answer!
Alain
i would like to use the graph theory analysis in conn in order to answer some of my research questions.
In particular, I am trying to understand if a certain network of brain regions change with age and if its relationship with cognition change with age.
To do that I had:
1) selected my ROIs
2) computed the graph theory measures using correlation coefficient and tresholding theedges to .15
3) exported the data.
In the exported data I have one line per subject with the measures for the network overall and for each node.
I have some questions about this:
1) does it make any sense to correlate network-wise and single node measures with behavioral data ?
2) how would I compare networks for different age groups ? Should I use some multivariate analysis (e.g. MANOVA including all the network measures as DV and the group as IV) ?
3) for some of the subjects some measures are missing (NaN), why does this happen ? Can I just remove these subjects ?
Thanks a lot for any answer!
Alain
Threaded View
| Title | Author | Date |
|---|---|---|
| Alain Imaging | May 17, 2017 | |
| Alfonso Nieto-Castanon | Jun 1, 2017 | |
| Jeff Browndyke | Jun 1, 2017 | |
| Alfonso Nieto-Castanon | Jun 1, 2017 | |
| Alain Imaging | May 29, 2017 | |