help
help > RE: Multisite data harmonisation and CONN
Oct 18, 2021 11:10 AM | Yann Quide
RE: Multisite data harmonisation and CONN
Thanks Alfonso.
However, matlab returns the following error when I use your script:
Error using conn_mtx_write (line 18)
input matrix must be square
Indeed names has 164 entries, and xyz 167 entries.
Not sure I see how to fix this.
Do you have any suggestion?
Thanks again for your time and help.
Yann
Originally posted by Alfonso Nieto-Castanon:
However, matlab returns the following error when I use your script:
Error using conn_mtx_write (line 18)
input matrix must be square
Indeed names has 164 entries, and xyz 167 entries.
Not sure I see how to fix this.
Do you have any suggestion?
Thanks again for your time and help.
Yann
Originally posted by Alfonso Nieto-Castanon:
Hi Yann,
It would be better the use directly the Z variable in the files resultsROI_Condition*.mat files for harmonization, since those are already concatenated across all subjects in your study and they are the ones that the second-level analysis functions will read. If you need that converted to NIFTI format you may use something like the following:
load resultsROI_Condition001.mat Z names xyz;
conn_mtx_write('mydata.nii', Z, names, xyz);
That will create a 4D nifti file with one volume per subject and each volume having a single 2D matrix of "voxels" (each containing one element of the ROI-to-ROI connectivity matrix for the associated subject&condition)
Hope this helps
Alfonso
Originally posted by Yann Quide:
It would be better the use directly the Z variable in the files resultsROI_Condition*.mat files for harmonization, since those are already concatenated across all subjects in your study and they are the ones that the second-level analysis functions will read. If you need that converted to NIFTI format you may use something like the following:
load resultsROI_Condition001.mat Z names xyz;
conn_mtx_write('mydata.nii', Z, names, xyz);
That will create a 4D nifti file with one volume per subject and each volume having a single 2D matrix of "voxels" (each containing one element of the ROI-to-ROI connectivity matrix for the associated subject&condition)
Hope this helps
Alfonso
Originally posted by Yann Quide:
Hi
Alfonso,
Bumping this one up.
Am I correct to think the resultsROI_Subjects0XX_Condition001.mat files are the files to harmonize for RRC analyses?
Thanks in advance.
Yann
Originally posted by Yann Quide:
Bumping this one up.
Am I correct to think the resultsROI_Subjects0XX_Condition001.mat files are the files to harmonize for RRC analyses?
Thanks in advance.
Yann
Originally posted by Yann Quide:
Thanks
Alfonso,
I feared you would say there is no such approach, yet!
Just a quick follow up question: I am fine harmonizing each SBC maps, but what about the RRC matrices?
Do you have any suggestion for this? In particular, what are the files you would recommend to use for these analyses?
Thanks again for your time and help.
Yann
Originally posted by Alfonso Nieto-Castanon:
I feared you would say there is no such approach, yet!
Just a quick follow up question: I am fine harmonizing each SBC maps, but what about the RRC matrices?
Do you have any suggestion for this? In particular, what are the files you would recommend to use for these analyses?
Thanks again for your time and help.
Yann
Originally posted by Alfonso Nieto-Castanon:
Hi Yann,
This is a very interesting suggestion. As far as I know each individual analysis would require a separately harmonisation, e.g. you could harmonise the first-level SBC maps across sites to then perform 2nd-level analyses of SBC maps, or you could harmonise the first-level RRC matrices across sites to then perform 2nd-level analyses of RRC matrices, but I have not seen (yet?) any method that would allow you harmonise the original BOLD signal data or other intermediate sources so that all subsequent connectivity measures derived from these are then automatically balanced across sites (but would love to hear about such approaches). In the absence of that, the simplest / first-order approach that I often recommend is to simply include site as control covariates directly in your second-level analyses (e.g. https://www.nitrc.org/forum/message.php?...), which can of course be equally used in the analysis of any arbitrary connectivity measure
Hope this helps
Alfonso
Originally posted by Yann Quide:
This is a very interesting suggestion. As far as I know each individual analysis would require a separately harmonisation, e.g. you could harmonise the first-level SBC maps across sites to then perform 2nd-level analyses of SBC maps, or you could harmonise the first-level RRC matrices across sites to then perform 2nd-level analyses of RRC matrices, but I have not seen (yet?) any method that would allow you harmonise the original BOLD signal data or other intermediate sources so that all subsequent connectivity measures derived from these are then automatically balanced across sites (but would love to hear about such approaches). In the absence of that, the simplest / first-order approach that I often recommend is to simply include site as control covariates directly in your second-level analyses (e.g. https://www.nitrc.org/forum/message.php?...), which can of course be equally used in the analysis of any arbitrary connectivity measure
Hope this helps
Alfonso
Originally posted by Yann Quide:
Hi Alfonso,
I am working with multisite data and was wondering what would be the best option when to include the scanner site harmonisation step (using the neuroHarmonize tools)? It is usually recommended to harmonise the first level files, but this will not include correlation matrices.
What file(s) would you recommend to be harmonised that will allow any type of subsequent analysis (ROI to ROI, seed based, etc)?
Thanks in advance,
Yann
I am working with multisite data and was wondering what would be the best option when to include the scanner site harmonisation step (using the neuroHarmonize tools)? It is usually recommended to harmonise the first level files, but this will not include correlation matrices.
What file(s) would you recommend to be harmonised that will allow any type of subsequent analysis (ROI to ROI, seed based, etc)?
Thanks in advance,
Yann
Threaded View
| Title | Author | Date |
|---|---|---|
| Yann Quide | Oct 6, 2021 | |
| Alfonso Nieto-Castanon | Oct 10, 2021 | |
| Yann Quide | Oct 10, 2021 | |
| Yann Quide | Oct 14, 2021 | |
| Alfonso Nieto-Castanon | Oct 14, 2021 | |
| Yann Quide | Oct 18, 2021 | |
| Alfonso Nieto-Castanon | Oct 18, 2021 | |
| Yann Quide | Oct 9, 2021 | |