Dear all,

I was wondering whether it is possible not to provide the skull stripped EPI into PESTICA since I am used to skullstrip my subjects in the first step of the preprocessing while running PESTICA later when my subjects are already skullstripped. Could I at least trick anyhow PESTICA by feeding the skullstripped image and using the same image as the mask? or any other solutions? 
I really wish to avoid to skullstrip my image once again or changing my pipeline just for physiological correction.

By the way, can anyone clarify why the skullstripped image is required by PESTICA?

Thanks in Advance
R.

errata corrige

can anyone clarify why is the mask required by pestica?

Hi Rocco,
it is indeed possible to bypass the skull strip stage and use your own.  In order to do this, first create the subdirectory _pestica/ as appropriate for your input data (the filename prefix you give to run_pestica.sh -d ), and then 3dcopy your brain mask into this dir as .brain+orig.  That will bypass the skull stripping stage and use your skull stripped brain instead.

PESTICA requires a skull-stripped brain for two reasons: 1) a mask file reduces the number of voxels fed to the ICA procedure, and 2) the coregistration to MNI space works well only with a skull-stripped brain.  It sometimes works without stripping, but sometimes with affine coregs if you're coregistering a non-stripped image containing fat and other signal outside brain to a stripped image, it shrinks the whole non-stripped image so it fits inside the stripped image, artificially shrinking it.  This isn't a problem with rigid-body coregs, but most people's brains differ from the MNI template by (to 1st order) an affine transform.
Erik

Thx Erik for your quick reply,

So there is no way to bypass the skullstrip stage in the sense that I have to provide at least my mask instead of providing nothing.
Does all this imply that if I provide my skullstripped EPI as -d and the same EPI renamed EPI.brain+orig. I will gather nothing in the end?
If the algorithm simply subtracts, this means that all the voxels will be deactivated. correct?

thx
R.

I think my description of PESTICA is confusing, so I'll summarize.  PESTICA requires the temporal information, so the -d option should be the 3D+time dataset.  If you use the skull stripped brain as input, the ICA won't run as there will be no temporal data.  The first three stages of PESTICA are estimation steps that result in two vectors matched to your 3D+time dataset input, these should match up with a parallel monitored pulse oximetry and respiratory bellows signals.  The PESTICA algorithm produces these signals for you when you don't have a pulse ox or respiration measurement built into your pulse sequence or it failed during the scan.  Then you can use these estimators (when everything works right and these would match up with pulse ox/respiration its perhaps not appropriate to call them estimators, they're more "derived pulse trace and respiration trace" signals) in any of the other pulse ox/respiration signal correction methods, such as RETROICOR.  Steps 4-5 of run_pestica.sh actually run RETROICOR (if you specify the -r flag) and IRF-RETROICOR, and 6 produces QA plots.  If you have pulse ox and respiration signals for each scan, then its better to use those.  We created PESTICA specifically because many sites don't have pulse ox/respiration in their pulse sequences.  Does that make sense?
erik

Hi, I am wondering about two points here: I read that using PESTICA is better than using pulse ox and respiration signals, even if pulse and respiration signals were collected. And 2. If i do not specify  the -r flag PESTICA will not run retroicor and irf-retroicor? Is that coeect? 
 
Thanks. Jasmin

Hi Jasmin,

Actually, I consider monitored pulse and respiration to be the gold standard. If its monitored, properly synched to the acquisition and there aren't any large artifacts in the PMU and respiration belt signals, then those are the best things to use to model physiologic noise. PESTICA should be used when those aren't available or the PMU/respiration belt monitoring failed.

PESTICA is equivalent to monitored pulse and respiration if all goes well, but it can't be better unless the pulse/respiration monitoring failed in part of the acquisition. If there aren't any finger wiggle artifacts, the respiration belt was on okay, and the monitoring is good, then I'd go with monitored over PESTICA. There are times when you just don't see much physiologic noise coupling in the data, and in those instances, PESTICA's signals suffer. But then, there is less physiologic noise in the data in those instances. Validation is a tricky (and unsolved) problem for physiologic noise removal. I've seen more than a few times where in two scans of the same person, taken right after each other, one scan had lots of physiologic noise artifact in a seeded connectivity analysis but not in the other, mostly due to cardiac. Out of curiosity, where did you read that?

If you do not specify the -r flag, then PESTICA will not run RETROICOR in addition to IRF-RETROICOR. It always runs IRF-RETROICOR. So to reiterate (thats confusing), without -r, it will output one file, with -r it will output two files, one cleaned with RETROICOR and the other with IRF-RETROICOR.

Erik