Just an observation.
Running task-analysis with anaticor can be manageable but also very very slow.
Using a typical pre-proc script (where the mask_epi_anat is applied to both 3dDeconvovle and REML) will vary in time depending on if you use band-pass or not.
Only using anaticor (also running 3dDeconvolve to get anaitcor free results):
Not using anaticor but using bandpass (0.01 - 0.17):
Using anaticor combined with bandpass:
Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz
The biggest time-consumer is the final loop of anaticor. I think it is this loop:
++ GLSQ loop:0123456789.0123456789.0123456789.0123456789.0123456789.
It only uses one thread.
Another interesting observation is that anaitcor makes expected results less signifciant.
In this example someone is stroking the right palm of the patient. Activating relevant motor areas on the left side of the brain. We do expect a similar but less significant activation on the opposite site but at the same threshold that is gone in anatcior. Anaticor also misses the activation in visual cortex. The q-scores are also different:
stats: q = 0.00087
reml: q = 0.0121
I’ll grant you that this is one single subject. And lowering the threshold for anaitcor is proabably still significant. The smoothenss might also be affected so that the 3dClustsim results might be different. But it’s to illustrate the difference between the two methods applied to exactly the same data. Any ideas?
Threshold: p < 0.0001 Cluster > 100
Top stats, bottom reml (no bandpass on either)