help
help > RE: First-level ICC output
Aug 5, 2017 12:08 AM | Alfonso Nieto-Castanon - Boston University
RE: First-level ICC output
Hi Leah,
If the 'normalization' checkbox (in the voxel-to-voxel first-level analyses tab, when defining your ICC analyses) is unchecked, then the values computed for each subject/condition at each voxel in your ICC analyses represent the rms (root mean square) of the correlation values between that same voxel and all of the voxels in the brain (i.e. square root of the average r-squared across all other voxels). If, on the other hand, the 'normalization' checkbox is checked then these values are normalized to follow a N(0,1) distribution (this normalization is performed separately for each subject/condition).
Regarding your second question, the entire voxel-to-voxel correlation matrix (for each subject & condition) is implicitly computed but never explicitly stored (it would take a ridiculously large amount of space to do so, as each matrix contains many billions of elements). Nevertheless, if I am understanding correctly (please correct me if I am wrong) you would like to compute the average of these ICC values within individual clusters or regions that you have pre-computed (i.e. regions that showed significant differences between the two groups), and then you would like to see whether these average values are correlated with some independent subject-level measure (e.g. symptom scale). There are many different ways to do this directly from the computed ICC maps. Perhaps one simple way would be the following:
1) in your second-level ICC results tab define a new correlation analysis looking at the association between ICC and your behavioral-scores variable (e.g. select 'AllSubjects' and 'BehavioralScores' and enter a [0 1] between-subjects contrast), and click on 'results explorer'
This will look at any areas where the ICC values are correlated across subjects with their behavioral scores. If, in addition, you would like to limit these analyses only to your pre-defined ROIs then you could:
2) click on the 'plot effects' button, change the option that reads 'cluster of interest in current analysis' to 'other clusters of interest (select mask/ROI file)', and then select your pre-defined ROIs mask, and then click 'Ok'
That will show you one effect-size and statistic associated with each of your pre-defined ROIs showing the association/correlation between average ICC values within these ROIs and symptom scales across subjects.
Hope this helps
Alfonso
Originally posted by Leah Fleming:
If the 'normalization' checkbox (in the voxel-to-voxel first-level analyses tab, when defining your ICC analyses) is unchecked, then the values computed for each subject/condition at each voxel in your ICC analyses represent the rms (root mean square) of the correlation values between that same voxel and all of the voxels in the brain (i.e. square root of the average r-squared across all other voxels). If, on the other hand, the 'normalization' checkbox is checked then these values are normalized to follow a N(0,1) distribution (this normalization is performed separately for each subject/condition).
Regarding your second question, the entire voxel-to-voxel correlation matrix (for each subject & condition) is implicitly computed but never explicitly stored (it would take a ridiculously large amount of space to do so, as each matrix contains many billions of elements). Nevertheless, if I am understanding correctly (please correct me if I am wrong) you would like to compute the average of these ICC values within individual clusters or regions that you have pre-computed (i.e. regions that showed significant differences between the two groups), and then you would like to see whether these average values are correlated with some independent subject-level measure (e.g. symptom scale). There are many different ways to do this directly from the computed ICC maps. Perhaps one simple way would be the following:
1) in your second-level ICC results tab define a new correlation analysis looking at the association between ICC and your behavioral-scores variable (e.g. select 'AllSubjects' and 'BehavioralScores' and enter a [0 1] between-subjects contrast), and click on 'results explorer'
This will look at any areas where the ICC values are correlated across subjects with their behavioral scores. If, in addition, you would like to limit these analyses only to your pre-defined ROIs then you could:
2) click on the 'plot effects' button, change the option that reads 'cluster of interest in current analysis' to 'other clusters of interest (select mask/ROI file)', and then select your pre-defined ROIs mask, and then click 'Ok'
That will show you one effect-size and statistic associated with each of your pre-defined ROIs showing the association/correlation between average ICC values within these ROIs and symptom scales across subjects.
Hope this helps
Alfonso
Originally posted by Leah Fleming:
Hi all,
I am currently running a voxel-to-voxel ICC analysis across two different drug conditions. I want to see whether there is a correlation between ICC and symptom scales given to the participants in the regions which were found to be different across drug conditions. First of all, what is the statistical output for the first-level analysis with ICC? Is this just an average r-value or r-squared across all other voxels?
And second, is there a way that I can access a matrix of this first-level statistic across all participants from the voxel-to-voxel results? Or could I do a seed-to-voxel analysis using the voxels pulled out by ICC to get this information and somehow calculate the averaged global connectivity from that?
Thanks in advance for your help!
Best,
Leah
I am currently running a voxel-to-voxel ICC analysis across two different drug conditions. I want to see whether there is a correlation between ICC and symptom scales given to the participants in the regions which were found to be different across drug conditions. First of all, what is the statistical output for the first-level analysis with ICC? Is this just an average r-value or r-squared across all other voxels?
And second, is there a way that I can access a matrix of this first-level statistic across all participants from the voxel-to-voxel results? Or could I do a seed-to-voxel analysis using the voxels pulled out by ICC to get this information and somehow calculate the averaged global connectivity from that?
Thanks in advance for your help!
Best,
Leah
Threaded View
| Title | Author | Date |
|---|---|---|
| Leah Fleming | Aug 1, 2017 | |
| Alfonso Nieto-Castanon | Aug 5, 2017 | |
| Leah Fleming | Aug 16, 2017 | |
| Alfonso Nieto-Castanon | Aug 19, 2017 | |
| Leah Fleming | Aug 19, 2017 | |
| Leah Fleming | Aug 8, 2017 | |
| Leah Fleming | Aug 5, 2017 | |