You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, if --tedort is used, certain behaviors are a little confusing:
The mixing matrix that is written out in the standard location (desc-ICA_mixing.tsv) is the un-orthogonalized version, and the orthogonalized version is written out to its own location (desc-ICAOrth_mixing.tsv).
Regression outputs (desc-denoised_bold.nii.gz, desc-optcomRejected_bold.nii.gz, and desc-optcomAccepted_bold.nii.gz) are generated using the full, but orthogonalized, mixing matrix (equivalent to non-aggressive denoising with the orthogonalized mixing matrix), which probably doesn't make sense. If we have orthogonalized the rejected components, we should just regress those out of the optcom data to denoise it.
The component weight maps that are written out (desc-ICA_components.nii.gz, desc-ICA_stat-z_components.nii.gz, desc-ICAAccepted_components.nii.gz, desc-ICAAcceptedZ_components.nii.gz) are all created from the orthogonalized mixing matrix.
The component figures are created from the original mixing matrix. I don't know whether we want the figures to use the orthogonalized mixing matrix (but see Figures are generated for orthogonalized components if tedort is used #243) or to write out the unorthogonalized component weight maps to NIfTIs, but we really should make those two consistent.
I am personally leaning toward making the following changes:
Use the unorthogonalized components for the figures and the weight map NIfTIs.
Optionally write out the orthogonalized component weight maps to NIfTIs if verbose is enabled.
Use non-aggressive denoising with the non-orthogonalized mixing matrix even if tedort is enabled.
If tedort is enabled, write out an additional denoised output using aggressive denoising with the orthogonalized mixing matrix to a new file (desc-denoised+orth_bold.nii.gz?).
Additional information
This is related to #1356, as we are unable to load the desc-ICA_components.nii.gz and use those weight maps for the component figures as long as the NIfTI and figures use different components.
Summary
Currently, if
--tedortis used, certain behaviors are a little confusing:desc-ICA_mixing.tsv) is the un-orthogonalized version, and the orthogonalized version is written out to its own location (desc-ICAOrth_mixing.tsv).desc-denoised_bold.nii.gz,desc-optcomRejected_bold.nii.gz, anddesc-optcomAccepted_bold.nii.gz) are generated using the full, but orthogonalized, mixing matrix (equivalent to non-aggressive denoising with the orthogonalized mixing matrix), which probably doesn't make sense. If we have orthogonalized the rejected components, we should just regress those out of the optcom data to denoise it.desc-ICA_components.nii.gz,desc-ICA_stat-z_components.nii.gz,desc-ICAAccepted_components.nii.gz,desc-ICAAcceptedZ_components.nii.gz) are all created from the orthogonalized mixing matrix.I am personally leaning toward making the following changes:
desc-denoised+orth_bold.nii.gz?).Additional information
This is related to #1356, as we are unable to load the
desc-ICA_components.nii.gzand use those weight maps for the component figures as long as the NIfTI and figures use different components.