thanks Eugene for your answers...
i still have some doubts...i realized, reasoning and reading further posts, that final FEAT analysis (PPI or RSFC) must be done over res4d_meaned (adding to original residual a fix value or the mean image ). and that res4d is the output of a nuisance regression step which model and regress out the CSF, WM and global signals.
what i do not understand is how FEAT manage motion correction. The 3-stages pipeline of http://www.ccn.ucla.edu/wiki/index.php/Run_PPI_in_FSL makes me think that CSF/WM/global signal should be extracted from a functional image already corrected for motion. their 3 stage pipeline is:
1) motion correction: input: original EPI, do : add motion param the model
2) nuisance regression: input: filtered func data from step1; model: 2/3 EV single column with time series of CSF, WM and global confounds
3) PPI: input: res4d from stage 2, after re-meaning to 10000 (or to mean image as elsewhere suggested): model: 3 ev for PPI, while one EV (???), the roi timeserie for RSFC
you told me that filtered_func_data after a simple MCFLIRT and "add motion param the model" FEAT, is NOT corrected for motion. if so, that pipeline do not correct for motion, as stage 2 uses filtered func data from stage1 (not res4d) and do not use motion params in any other pipeline step.
trying to understand it, i read the feat log where it appears that filtered_func_data is the post-processed output of MCFLIRT which should correct for motion. accordingly, considering that res4d of stage 1 should be the remaining signal not explained by the 6 motion's EV model, it should be motion corrected. well, If I add the mean_func to the res4d, i get the filtered_func_data.
according instead to what you told me (filtered_func_data is not motion corrected), in stage 2 I should use:
a) the res4d created in stage 1 (after having meaned it). inserting 3 EVs for CSF/WM/global confounds?
b) the filtered func data created in stage 1. inserting 3 EVs for CSF/WM/global confounds and an additional confound file with 6 columns
which is the correct way?
reading further post, I read that all the final FEAT are made with just the ROI timeserie EV, i don't need to set a 1s column ? (which would substitute the 1st EV of PPI analysis)
I still have to understand why most of the post suggest using FEAT, while Christian here : https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1203&L=fsl&D=0&1=fsl&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=278621
suggests fsl_glm and randomize, but this is not a priority now.
thanks in advance
Alberto
|