Hi,
I'm interested in create denoised fMRI volumes for analyses in other software.
These are created in niftiDATA_SubjectXXX_Condition000.nii in the .../result/preprocesing folder. However, they are also free from the Effect of the covariates that I'm interested in. I would like to take these denoised images and add the effect of interest back. I suppose I can do this by:
y* = y+X*beta (for each voxel)
where y* is the "un-regressed" signal, y is the denoised signal in niftiData...., X is the design submatrix of regressors of interest (which I suppose I can get from .../data/COV/COV_SubjectXXX_SessionXXX.mat) and beta is the column vector of corresponding effect sizes, which I suppose are stored in BETA_denoising_SubjectXXX.nii.
Does this make sense? Is there any filtering or whitening operation that I must take into account?
Besides, the BETA images are 4d Niftis with variable amount of effects per subject. This makes sense of course because each subject has its own amount of regressors (scrubbing, etc). But I find no file referring to which 3D image (effect) corresponds to which specific regressor. In fact, identifying the betas for my covariates of interests is already very informative!

Hi,

Here are a couple of suggestions which may be of help:

1. You could go to the "data" folder and pick up the matc files. Then use the conn_matc2nii function to create NIfTI files. This would give you the time series without denoising (and also would retain the effects of all covariates).

2. An alternate is to run denoising once again and simply disable.remove the effects that you do not want to remove. This would be a preferred choice in case you want selected operations (like band pass filtering) to be applied while retaining the effect of other covariates.

Hope this helps


Best
Pravesh

P.S: You might want to post the question to Conn forum

Dear Pravesh,
Thank you for the tips.

I actually think I figured out what I wanted. But feedback is always appreciated.
I can indeed denoise after removing the 'Effect of...' covariates from the nuisance list.
But I tried something else.
Since I wanted to take advantage of CONN way of creating nuisance time series, and at the same time of SPM other functionalities, like more flexible HRFs, Volterra kernels..., I created the design matrix using conn_designmatrix (after loading the roi, cov and confounds and filters) and removed the conditions from this (very easy since the names of the variables are the last output of this function).
Then I created the appropiate regressors in the SPM matlabbatch for the model specification.
Similarly I also exported the conditions to the SPM matlabbatch.

Any thought??

PS: I thought I was posting! How do I do that!