help
help > RE: ICA networks
Oct 12, 2016 07:10 AM | Julia Binnewies
RE: ICA networks
Dear Alfonso,
Thank you so much for your answer!
Julia
Originally posted by Alfonso Nieto-Castanon:
Thank you so much for your answer!
Julia
Originally posted by Alfonso Nieto-Castanon:
Dear
Julia,
There are two slightly different ways that are commonly used to identify/characterize the networks of interest (and both ways are also similarly-commonly used)
1) threshold based on the Z- or T- statistics of a one-sample t-test of all of the subject-level ICA spatial maps (this requires backprojecting first the group-level maps to subject-level maps)
and 2) threshold based on the Z- scores of the group-level ICA spatial maps directly (this only uses the group-level maps, so it does not require backprojection, but instead requires converting the group-level map factor scores to Z-scores directly, which is commonly performed by normalizing the factor score values across all of the voxels; i.e. removing the mean and dividing by the standard deviation across voxels)
In CONN you can, as you describe, implement (1) by going to the ICA.SpatialMaps tab and performing a one-sample t-test there (and thresholding the resulting at the desired level). In CONN you can also implement (2) simply by going to ICA.Summary and choosing a z-score threshold directly on the top-right plot. The differences between (1) and (2) are often subtle and both are perfectly valid/common approaches, so feel free to use whichever method gives you the clearer/cleaner interpretation (I often like (1) the best, because it relates a little bit closer to other between-subject comparisons that I might want to perform on these same subject-level ICA spatial maps, but both are perfectly fine methods to identify/characterize the networks of interest)
Hope this helps
Alfonso
Originally posted by Julia Binnewies:
There are two slightly different ways that are commonly used to identify/characterize the networks of interest (and both ways are also similarly-commonly used)
1) threshold based on the Z- or T- statistics of a one-sample t-test of all of the subject-level ICA spatial maps (this requires backprojecting first the group-level maps to subject-level maps)
and 2) threshold based on the Z- scores of the group-level ICA spatial maps directly (this only uses the group-level maps, so it does not require backprojection, but instead requires converting the group-level map factor scores to Z-scores directly, which is commonly performed by normalizing the factor score values across all of the voxels; i.e. removing the mean and dividing by the standard deviation across voxels)
In CONN you can, as you describe, implement (1) by going to the ICA.SpatialMaps tab and performing a one-sample t-test there (and thresholding the resulting at the desired level). In CONN you can also implement (2) simply by going to ICA.Summary and choosing a z-score threshold directly on the top-right plot. The differences between (1) and (2) are often subtle and both are perfectly valid/common approaches, so feel free to use whichever method gives you the clearer/cleaner interpretation (I often like (1) the best, because it relates a little bit closer to other between-subject comparisons that I might want to perform on these same subject-level ICA spatial maps, but both are perfectly fine methods to identify/characterize the networks of interest)
Hope this helps
Alfonso
Originally posted by Julia Binnewies:
Dear CONN community and Alfonso,
I have a question regarding ICA analysis in CONN, more precisely how I can see which regions the components consist of. I first looked at the networks (which regions they consist of) by selecting 'all subjects' in the spatial section of the ICA networks second-level analysis, with the contrast [1]. In the results explorer I then changed the voxel-wise threshold for each network in order to obtain a nice and clear network.
However, when looking at the components across all subjects in the 'summary' section, for some networks I see very different (or at least much less) clusters for the same components.
How is this possible? What is the difference between these 2 sections? And which one should I use in order to identify the networks I'm interested in?
Any help would be greatly appreciated!
Best whishes,
Julia
I have a question regarding ICA analysis in CONN, more precisely how I can see which regions the components consist of. I first looked at the networks (which regions they consist of) by selecting 'all subjects' in the spatial section of the ICA networks second-level analysis, with the contrast [1]. In the results explorer I then changed the voxel-wise threshold for each network in order to obtain a nice and clear network.
However, when looking at the components across all subjects in the 'summary' section, for some networks I see very different (or at least much less) clusters for the same components.
How is this possible? What is the difference between these 2 sections? And which one should I use in order to identify the networks I'm interested in?
Any help would be greatly appreciated!
Best whishes,
Julia
Threaded View
| Title | Author | Date |
|---|---|---|
| Julia Binnewies | Oct 9, 2016 | |
| Alfonso Nieto-Castanon | Oct 12, 2016 | |
| Julia Binnewies | Oct 12, 2016 | |