help
help > RE: parcellation based on voxel-to-voxel analysis
May 1, 2014 05:05 AM | Alfonso Nieto-Castanon - Boston University
RE: parcellation based on voxel-to-voxel analysis
Dear Hong,
Some thoughts on your questions below
Best
Alfonso
Originally posted by Jui Yang Hong:
If you use a explicit mask then only within-thalamus connectivity will be considered for voxel-to-voxel analyses. If you are interested in thalamo-cortical connectivity then keep the original whole-brain mask and use later small volume correction in the resulting voxel-to-voxel results to focus on thalamic regions.
2) Since I have only one group, can I use either one of the analysis methods (ILC, RCC, ICC or RSC)?
Yes, you may use any of the ILC/RCC/ICC/RSC measures (but not the MVPA measures, since those are aimed at between-group or between-condition comparisons)
3) At first level results, why did ICC and ILC only generate 1 component (1 contrast) and for RCC and RSC, they generate 4 component (3 contrasts)? what do they mean?
ICC and ILC are scalar measures, while RCC and RSC are vector measures. For each vector measure your get three maps (components 1 to 3) associated with each spatial coordinate (x,y, and z, respectively) and the additional map (component 0) represents the norm of the associated vector. ILC and RCC are local metrics characterizing the local connectivity between each voxel and its neighbors, while ICC and RSC and global measures characterizing the global connectivity between each voxel and the rest of the brain (see the manual or the the toolbox reference for a more detailed description). For parcellation I would probably suggest focusing on the RSC measure, since that may be the measure most likely to highlight potential regional boundaries (see for example Cohen et al. 2008)
4) how does the kernel size affect the results
For local measures (ILC and RCC), the kernel size defines the extent of the neighborhood considered when computing local connectivity patterns (by default this is 8mm), while for global measures the kernel size controls whether you want to aggregate the resulting measures across neighboring voxels (by default it is set to 0mm, meaning no aggregation)
5) Which files in the firtlevel results should I look at (BETA map?) for the subregions of thalamus based on voxel-to-voxel analysis at individual level and group level?
The files named BETA_Subject#_Condition#_Measure#_Component#.nii in the conn*/results/firstlevel/ folder store the subject-level measures (for each subject/condition; see the file _list_measures.txt to see what each measure number represents). When you perform a second-level analysis on any of these measures a new folder will be created and the group-level results will be stored in that folder (within conn*/results/secondlevel) and named beta_#.img and con_*.img (the standard SPM convention; they will store your beta coefficients and contrast values, respectively, and their interpretation depends on your second-level design -for a simple one-sample t-test they represent the group-level averages)
Hope this helps
Alfonso
Some thoughts on your questions below
Best
Alfonso
Originally posted by Jui Yang Hong:
At the Setup, should I change the explicit mask
with only thalamus mask? To do so, are the connectivity patterns
only within the thalamus not with rest of the brain? However, if I
don't use the thalamus mask, how to get only the subregions of
thalamus connectivity patterns (or subclusters of thalamus)?
If you use a explicit mask then only within-thalamus connectivity will be considered for voxel-to-voxel analyses. If you are interested in thalamo-cortical connectivity then keep the original whole-brain mask and use later small volume correction in the resulting voxel-to-voxel results to focus on thalamic regions.
2) Since I have only one group, can I use either one of the analysis methods (ILC, RCC, ICC or RSC)?
Yes, you may use any of the ILC/RCC/ICC/RSC measures (but not the MVPA measures, since those are aimed at between-group or between-condition comparisons)
3) At first level results, why did ICC and ILC only generate 1 component (1 contrast) and for RCC and RSC, they generate 4 component (3 contrasts)? what do they mean?
ICC and ILC are scalar measures, while RCC and RSC are vector measures. For each vector measure your get three maps (components 1 to 3) associated with each spatial coordinate (x,y, and z, respectively) and the additional map (component 0) represents the norm of the associated vector. ILC and RCC are local metrics characterizing the local connectivity between each voxel and its neighbors, while ICC and RSC and global measures characterizing the global connectivity between each voxel and the rest of the brain (see the manual or the the toolbox reference for a more detailed description). For parcellation I would probably suggest focusing on the RSC measure, since that may be the measure most likely to highlight potential regional boundaries (see for example Cohen et al. 2008)
4) how does the kernel size affect the results
For local measures (ILC and RCC), the kernel size defines the extent of the neighborhood considered when computing local connectivity patterns (by default this is 8mm), while for global measures the kernel size controls whether you want to aggregate the resulting measures across neighboring voxels (by default it is set to 0mm, meaning no aggregation)
5) Which files in the firtlevel results should I look at (BETA map?) for the subregions of thalamus based on voxel-to-voxel analysis at individual level and group level?
The files named BETA_Subject#_Condition#_Measure#_Component#.nii in the conn*/results/firstlevel/ folder store the subject-level measures (for each subject/condition; see the file _list_measures.txt to see what each measure number represents). When you perform a second-level analysis on any of these measures a new folder will be created and the group-level results will be stored in that folder (within conn*/results/secondlevel) and named beta_#.img and con_*.img (the standard SPM convention; they will store your beta coefficients and contrast values, respectively, and their interpretation depends on your second-level design -for a simple one-sample t-test they represent the group-level averages)
Hope this helps
Alfonso
Threaded View
| Title | Author | Date |
|---|---|---|
| Jui Yang Hong | Apr 29, 2014 | |
| Alfonso Nieto-Castanon | May 1, 2014 | |
| Jui Yang Hong | May 2, 2014 | |
| Alfonso Nieto-Castanon | May 8, 2014 | |