help
help > RE: CONN denoise output
Feb 9, 2021 01:02 AM | Alfonso Nieto-Castanon - Boston University
RE: CONN denoise output
Dear Lorenzo,
Right, the difference is that conn_module does not mask nor re-scale the data, while the CONN GUI (if you use the default settings) will do so (see Setup.Options tab). In those distribution plots you are basically seeing the spatial distribution of the average BOLD signal (since the voxel-to-voxel variability in BOLD signal is orders of magnitude larger than the temporal variability in BOLD signal). Those average BOLD signals include a lot of voxels outside of the brain in the "conn_module" results (hence the peak in values around zero), which are not present in the "CONN GUI" results. Also the scale of the "CONN GUI" results is around 100 (as a product of the PSC unit re-scaling), while the scale of the "CONN module" results is maintained from the original BOLD data.
If you compare both of these results to those of fsl_glm, the main difference there I believe is going to be that CONN (since release 19b, see "change.log" for details) adds back the average BOLD signal to the resulting denoised timeseries. The reasons for this is that otherwise the average BOLD signal will naturally be exactly zero (high-pass removes the 0-frequency), and we often found that this was a constant source of problems in earlier versions when users wanted to use the denoised functional data in different software packages (e.g. SPM often uses the average BOLD signal at each voxel to determine a reasonable "brainmask", or to determine a reasonable percent-signal-change unit scale, etc. so using zero-average functional data was rather tricky; adding back that constant BOLD signal value at each voxel to the denoised data fixed these issues)
Hope this helps
Alfonso
Originally posted by Lorenzo Pasquini:
Right, the difference is that conn_module does not mask nor re-scale the data, while the CONN GUI (if you use the default settings) will do so (see Setup.Options tab). In those distribution plots you are basically seeing the spatial distribution of the average BOLD signal (since the voxel-to-voxel variability in BOLD signal is orders of magnitude larger than the temporal variability in BOLD signal). Those average BOLD signals include a lot of voxels outside of the brain in the "conn_module" results (hence the peak in values around zero), which are not present in the "CONN GUI" results. Also the scale of the "CONN GUI" results is around 100 (as a product of the PSC unit re-scaling), while the scale of the "CONN module" results is maintained from the original BOLD data.
If you compare both of these results to those of fsl_glm, the main difference there I believe is going to be that CONN (since release 19b, see "change.log" for details) adds back the average BOLD signal to the resulting denoised timeseries. The reasons for this is that otherwise the average BOLD signal will naturally be exactly zero (high-pass removes the 0-frequency), and we often found that this was a constant source of problems in earlier versions when users wanted to use the denoised functional data in different software packages (e.g. SPM often uses the average BOLD signal at each voxel to determine a reasonable "brainmask", or to determine a reasonable percent-signal-change unit scale, etc. so using zero-average functional data was rather tricky; adding back that constant BOLD signal value at each voxel to the denoised data fixed these issues)
Hope this helps
Alfonso
Originally posted by Lorenzo Pasquini:
dear Alfonso and other CONN users,
I am really perplexed by what is going on here.
For whatever reason the denoised output from the CONN_modules (first row in screenshot) and CONN GUI (second row) do not correspond.
The outputs are very different in terms of distributions (histograms) and spatial patterns (masked to skull stripped brain).
I also regressed out the covariates manually using fsl_glm and these maps look again pretty different.
Below an extract for the code used to generate the bdswuaf.nii:
conn_module( 'preprocessing', ...
'structurals', {'/Volumes/My_Book/test/anat.nii'}, ...
'functionals', {'/Volumes/My_Book/test/f.nii'}, ...
'steps', {'default_mni', 'functional_bandpass','functional_regression'}, ...
'sliceorder', 'ascending', ...
'fwhm', 6, ...
'reg_names', {'realignment','scrubbing','White Matter','CSF'}, ...
'reg_dimensions',[inf, inf, 5, 5], ...
'reg_deriv', [1, 0, 0, 0], ...
'bp_filter', [0.008 0.09],...
'reg_filter', [1 0 1 1])
Is the effect of rest and detrending implemented in the default
conn_module available online?
Any idea on what I may be doing wrong?
Any help would be greatly appreciated.
Thanks
Lorenzo
I am really perplexed by what is going on here.
For whatever reason the denoised output from the CONN_modules (first row in screenshot) and CONN GUI (second row) do not correspond.
The outputs are very different in terms of distributions (histograms) and spatial patterns (masked to skull stripped brain).
I also regressed out the covariates manually using fsl_glm and these maps look again pretty different.
Below an extract for the code used to generate the bdswuaf.nii:
conn_module( 'preprocessing', ...
'structurals', {'/Volumes/My_Book/test/anat.nii'}, ...
'functionals', {'/Volumes/My_Book/test/f.nii'}, ...
'steps', {'default_mni', 'functional_bandpass','functional_regression'}, ...
'sliceorder', 'ascending', ...
'fwhm', 6, ...
'reg_names', {'realignment','scrubbing','White Matter','CSF'}, ...
'reg_dimensions',[inf, inf, 5, 5], ...
'reg_deriv', [1, 0, 0, 0], ...
'bp_filter', [0.008 0.09],...
'reg_filter', [1 0 1 1])
Any idea on what I may be doing wrong?
Any help would be greatly appreciated.
Thanks
Lorenzo
Threaded View
| Title | Author | Date |
|---|---|---|
| Lorenzo Pasquini | Feb 9, 2021 | |
| Alfonso Nieto-Castanon | Feb 9, 2021 | |
| Lorenzo Pasquini | Feb 9, 2021 | |