help > RE: Nuisance regress via compcor
May 19, 2015  01:05 PM | Alfonso Nieto-Castanon - Boston University
RE: Nuisance regress via compcor
Hi Frank,

Yes, that is perfectly possible. If you have already fully preprocessed your data simply enter your preprocessed functionals in Setup.Functional and your preprocessed structurals (in same space as your functional volumes, e.g. normalized if you have normalized your functionals) in Setup.Structural. If you have already segmented your structural volumes you may enter those gray/white/csf masks (again in the same space as your functional and structural volumes) in Setup.ROIs (or leave those empty otherwise and CONN will run that segmentation). In Setup.Options check the 'create confound-corrected timeseries' checkbox, the 'same space as functionals' option in 'analysis space', the 'implicit masking' option in 'analysis mask', and (if your data is band-pass filtered) select the 'raw units' option in 'analysis units'. Then skip the 'Preprocessing' step and run directly the Setup step. Last, in the Denoising step select the desired denoising options (e.g. enter only the white/csf masks and remove other regressors as well as filtering and detrending if you only want to apply CompCor) and run that step. After this you will find the denoised files in conn_*/results/preprocessing/niftiDATA_Subject*_Condition*.nii (one 4f nifti file per subject/condition). You will also find the eroded masks in the same folders as your original structural or segmentation files (eroded filenames prepend 'e' to the original gray/white/csf mask filenames)

Let me know if that is what you had in mind

EDIT: on second thoughts, if your functional data is already band-pass filtered the 'implicit masking' option is going to have trouble identifying the mask of "in-brain" voxels form the functional data alone (since that mask is computed from the average BOLD signal across your entire timeseries) so I would use instead 'explicit mask' and enter there an appropriately liberal mask (e.g. if your data is normalized to MNI space leaving the default brainmask option will probably suffice since that is quite liberal)

Originally posted by Frank Hillary:
Is there a way to access CompCor for regression of CSF and white matter signal only?  That is, for this specific question, all other preprocessing is handled elsewhere, but we want to use CompCor to analyze signal from CSF and white matter as regressors of no interest. How might this limited nuisance regression be conducted within CONN? Also, can one create an "eroded" mask for each within CONN?
Much appreciated,

Threaded View

Frank Hillary May 18, 2015
RE: Nuisance regress via compcor
Alfonso Nieto-Castanon May 19, 2015
Moa Peter May 15, 2018