help
help > RE: Implementation of graph theory in task fMRI
Dec 6, 2019 03:12 PM | Alfonso Nieto-Castanon - Boston University
RE: Implementation of graph theory in task fMRI
Dear Ofir
Sorry the current graph measures analyses implemented in CONN are somewhat limited in flexibility/options. In your case the optimal first-level analysis is probably gPPI given the relatively short length of your blocks, but you are right that this would force you to use some external toolbox like BCT (https://www.nitrc.org/projects/bct/) in order to analyze those connectivity matrices, since CONN only works with undirected graphs. Alternatively you may also try instead simply using weighted-GLM first-level analyses and the default bivariate correlation measures, which would then allow you to use CONN's graph theory metrics and methods. Given that 12.5s is not terribly short, weighted GLM and gPPI could be expected to produce similar results so this may be a simpler approach. Last, regarding the number of ROIs, typically graph properties are easier to identify over large networks (particularly so when using unweighted graphs) because that minimizes the impact of discretization issues in small samples. That said, I am not aware of any recommended minimum number of nodes/ROIs in the literature (and if I had to make up a "rule of thumb" I would probably recommend switching to weighted graphs for improved sensitivity if you have graphs with fewer than ~10 ROIs)
Hope this helps
Alfonso
Originally posted by Ofir Shany:
Sorry the current graph measures analyses implemented in CONN are somewhat limited in flexibility/options. In your case the optimal first-level analysis is probably gPPI given the relatively short length of your blocks, but you are right that this would force you to use some external toolbox like BCT (https://www.nitrc.org/projects/bct/) in order to analyze those connectivity matrices, since CONN only works with undirected graphs. Alternatively you may also try instead simply using weighted-GLM first-level analyses and the default bivariate correlation measures, which would then allow you to use CONN's graph theory metrics and methods. Given that 12.5s is not terribly short, weighted GLM and gPPI could be expected to produce similar results so this may be a simpler approach. Last, regarding the number of ROIs, typically graph properties are easier to identify over large networks (particularly so when using unweighted graphs) because that minimizes the impact of discretization issues in small samples. That said, I am not aware of any recommended minimum number of nodes/ROIs in the literature (and if I had to make up a "rule of thumb" I would probably recommend switching to weighted graphs for improved sensitivity if you have graphs with fewer than ~10 ROIs)
Hope this helps
Alfonso
Originally posted by Ofir Shany:
Dear CONN experts,
I'm interested in conducting a graph theory analysis on data form a block-designed fMRI task, and I have a few questions about this issue:
1. A general question -I rarely encounter studies that implement graph theory in context of task-fMRI. Why is this the case, and when can it be suitable to do so (i.e. what would be an "optimal" task design allowing implementation of graph theory)?
2. Specifically, in my task there are 4 conditions that appear in blocks of 12.5s (5 X TR 2.5s), where each condition repeats 3 times – can I conduct a graph theory analysis on such a design?
3. If I intend to do graph theory in a task, should I still use gPPI in the 1st level analysis? Or should I switch to basic functional connectivity (weighted GLM)?
4. Furthermore, if gPPI is performed in the 1st level, how does CONN deal with the fact that the correlations between two nodes differ according to their definition as target vs. seed ROIs? Does this mean that the graph is directed? And which methods are appropriate in this case?
5. Is there any minimum recommended number of ROIs for performing a graph analysis?
Any answer will be greatly appreciated,
Best regards,
Ofir
I'm interested in conducting a graph theory analysis on data form a block-designed fMRI task, and I have a few questions about this issue:
1. A general question -I rarely encounter studies that implement graph theory in context of task-fMRI. Why is this the case, and when can it be suitable to do so (i.e. what would be an "optimal" task design allowing implementation of graph theory)?
2. Specifically, in my task there are 4 conditions that appear in blocks of 12.5s (5 X TR 2.5s), where each condition repeats 3 times – can I conduct a graph theory analysis on such a design?
3. If I intend to do graph theory in a task, should I still use gPPI in the 1st level analysis? Or should I switch to basic functional connectivity (weighted GLM)?
4. Furthermore, if gPPI is performed in the 1st level, how does CONN deal with the fact that the correlations between two nodes differ according to their definition as target vs. seed ROIs? Does this mean that the graph is directed? And which methods are appropriate in this case?
5. Is there any minimum recommended number of ROIs for performing a graph analysis?
Any answer will be greatly appreciated,
Best regards,
Ofir
Threaded View
| Title | Author | Date |
|---|---|---|
| Ofir Shany | Nov 30, 2019 | |
| Alfonso Nieto-Castanon | Dec 6, 2019 | |
| Ofir Shany | Jan 13, 2020 | |
| Sadam Hussain | Dec 1, 2019 | |