help
help > RE: Using SPM to process second level
Mar 22, 2015 03:03 AM | Alfonso Nieto-Castanon - Boston University
RE: Using SPM to process second level
Hi Julian,
That is perfectly fine, there are no particular issues with mixing CONN first-level volumes with SPM for further analyses. Assuming that you use indirect normalization (normalize each subject structural volume and then apply that transformation to the corr_* or BETA_* images) then the results should be almost identical to what you would get by first normalizing in a similar way your original functional volumes and then processing your normalized functional volumes (some subtle differences may arise from the difference between smoothing in MNI space and smoothing in subject-space, since the two spaces are not linearly related). That said, I would recommend using the BETA_Subject#_Condition#_Source#.nii images instead of the corr_Subject#_Condition#_Source#.nii images for further second-level analyses. The latter represent correlation coefficients, while the former represent Fisher-transformed correlation coefficients, which typically are preferred because the resulting second-level model residuals are most likely to follow an approximately-normal distribution.
Hope this helps
Alfonso
Originally posted by Julian Cheng:
That is perfectly fine, there are no particular issues with mixing CONN first-level volumes with SPM for further analyses. Assuming that you use indirect normalization (normalize each subject structural volume and then apply that transformation to the corr_* or BETA_* images) then the results should be almost identical to what you would get by first normalizing in a similar way your original functional volumes and then processing your normalized functional volumes (some subtle differences may arise from the difference between smoothing in MNI space and smoothing in subject-space, since the two spaces are not linearly related). That said, I would recommend using the BETA_Subject#_Condition#_Source#.nii images instead of the corr_Subject#_Condition#_Source#.nii images for further second-level analyses. The latter represent correlation coefficients, while the former represent Fisher-transformed correlation coefficients, which typically are preferred because the resulting second-level model residuals are most likely to follow an approximately-normal distribution.
Hope this helps
Alfonso
Originally posted by Julian Cheng:
Hi all,
Our group is rather unique in that our subject level data is not normalized to MNI before feeding it into CONN for first level analysis.
As such, the correlation maps are in subject space and not MNI space, so we can't use CONN to do group level.
What we have been doing is using SPM to normalize the corr_* images to MNI space, and then using SPM to do straight-forward T-tests.
My question is does this seem valid to you? Are there caveats we should pay attention to when we mix CONN processing with SPM processing like this?
Thanks!
Our group is rather unique in that our subject level data is not normalized to MNI before feeding it into CONN for first level analysis.
As such, the correlation maps are in subject space and not MNI space, so we can't use CONN to do group level.
What we have been doing is using SPM to normalize the corr_* images to MNI space, and then using SPM to do straight-forward T-tests.
My question is does this seem valid to you? Are there caveats we should pay attention to when we mix CONN processing with SPM processing like this?
Thanks!
Threaded View
| Title | Author | Date |
|---|---|---|
| Julian Cheng | Mar 18, 2015 | |
| Alfonso Nieto-Castanon | Mar 22, 2015 | |
| Julian Cheng | Mar 24, 2015 | |