help
help > RE: Functional Network Connectivity for ROI-to-ROI 2nd level analysis
Apr 19, 2023 10:04 PM | Alfonso Nieto-Castanon - Boston University
RE: Functional Network Connectivity for ROI-to-ROI 2nd level analysis
Dear Chris
Yes, the MVPA omnibus test here uses the exact same SVD-based procedure as the fc-MVPA method to perform an omnibus test encompassing multiple connections (in this case all connections within each cluster; the procedure here reduces to first performing a singular value decomposition of the connections-by-subjects matrix of connectivity-strength values within each cluster, and then performing a multivariate test evaluating your 2nd-level analyses on the resulting components-by-subjects matrix of component scores, all of this separately within each cluster)
Regarding the number of ROIs you are right that this procedure is often used in cases where the number of ROIs is relatively large (e.g. hundreds or larger), while for relatively fewer ROIs (e.g. tens or lower) performing directly connection-level inferences (e.g. using the "alternative settings for connection-based inferences" option) may suffice, as the total number of connections evaluated will not be terribly constraining (this method simply computes a separate test for each individual connection and then applies an FDR correction across all connections evaluated)
Hope this helps
Alfonso
Originally posted by Christopher Hyatt:
Yes, the MVPA omnibus test here uses the exact same SVD-based procedure as the fc-MVPA method to perform an omnibus test encompassing multiple connections (in this case all connections within each cluster; the procedure here reduces to first performing a singular value decomposition of the connections-by-subjects matrix of connectivity-strength values within each cluster, and then performing a multivariate test evaluating your 2nd-level analyses on the resulting components-by-subjects matrix of component scores, all of this separately within each cluster)
Regarding the number of ROIs you are right that this procedure is often used in cases where the number of ROIs is relatively large (e.g. hundreds or larger), while for relatively fewer ROIs (e.g. tens or lower) performing directly connection-level inferences (e.g. using the "alternative settings for connection-based inferences" option) may suffice, as the total number of connections evaluated will not be terribly constraining (this method simply computes a separate test for each individual connection and then applies an FDR correction across all connections evaluated)
Hope this helps
Alfonso
Originally posted by Christopher Hyatt:
Dear CONN users,
I'm working with 51 user-defined ROIs in CONN, doing an ROI-to-ROI analysis only, with a 2nd level analysis using the 'standard settings for cluster-based inferences #1: Functional Network Connectivity' option.
For this FNC option, there is the second setting for 'cluster-level p-FDR corrected (MVPA omnibus test)'.
Exactly how is MVPA (omnibus test) used for this ROI-to-ROI FNC? Is it the same SVD-based mathematical formulation as voxel-to-voxel fc-MVPA, or a different algorithm?
Also, is 51 ROIs a large enough sample to use cluster-level methods for mult comparison correction? It might be that 51 ROIS (1275 pairs of connections) is too few for a proper cluster-level correction method.
Thanks for your help,
Chris
I'm working with 51 user-defined ROIs in CONN, doing an ROI-to-ROI analysis only, with a 2nd level analysis using the 'standard settings for cluster-based inferences #1: Functional Network Connectivity' option.
For this FNC option, there is the second setting for 'cluster-level p-FDR corrected (MVPA omnibus test)'.
Exactly how is MVPA (omnibus test) used for this ROI-to-ROI FNC? Is it the same SVD-based mathematical formulation as voxel-to-voxel fc-MVPA, or a different algorithm?
Also, is 51 ROIs a large enough sample to use cluster-level methods for mult comparison correction? It might be that 51 ROIS (1275 pairs of connections) is too few for a proper cluster-level correction method.
Thanks for your help,
Chris
Threaded View
| Title | Author | Date |
|---|---|---|
| Christopher Hyatt | Apr 14, 2023 | |
| Alfonso Nieto-Castanon | Apr 19, 2023 | |
| Christopher Hyatt | Apr 21, 2023 | |