Hello CONN users,

I've been reading some articles on artifact reduction in connectivity through BOLD signal (these three in particular):
https://www.ncbi.nlm.nih.gov/pmc/article...
https://www.ncbi.nlm.nih.gov/pubmed/2795...
http://www.sciencedirect.com/science/art...

Curiously I've noticed that some of the authors have suggested smoothing the EPI images within specified mask regions (i.e. 6mm FWHM on a grey matter mask, same on a white matter mask using 3dBlurInMask) on the residuals of the EPI data once the nuisance regressors have been regressed out.

I was wondering how (if possible) one could do this in CONN.

Is it as simple as asking a residual file to be written, smoothing it with the specified parameters, and then placing the relevant files back into the appropriate matlab directory?

If so, may I ask where exactly this area in the directory would be?

Hi Thomas,

Regarding the order of operations (smoothing before or after regression) both orders should result in exactly the same results (see this thread and linked posts for additional details/discussion). Smoothing within a mask is not terribly complex, you can accomplish that by:
1) creating a new volume by multiplying your mask by your functional data
2) smoothing the new masked functional volume
3) smoothing your original mask
4) dividing the results of (2) by the results of (3)
(e.g. steps (1) and (4) can be done using spm_imcalc and steps (2) and (3) would be normal smoothing/convolution operations)
Since the order of operations is irrelevant, if you want to perform masked-smoothing I would probably suggest you apply that to your original functional data (after preprocessing, before denoising), and then continue with the normal steps in CONN (i.e. denoising, first-level analyses, etc.)

Alphonso,

Thank you for the quick reply and your input.

If I may ask though, why does the original mask need to be smoothed?

I've actually read some methods here and there suggesting that the WM and CSF masks be eroded a bit. Something I've played with in CONN but can't seem to find enough consensus in the literature to find concrete suggestions.