bbfrederick
commented
5 years ago Hi Daniel,
Cool application!
There is sometimes a little bit of a learning curve dialing in settings for a new application, and that sounds sufficiently different from anything I've done so far that the defaults probably aren't appropriate. A couple thoughts:
0) You should probably use rapidtide2x - at the moment, it's much more advanced than rapidtide2 - there are a number of things I haven't yet moved into the main branch. 1) As you suspect, motion correction is a very good idea. Eye motion is likely to be a problem - don't know exactly the best way to deal with that. If you have some estimate of the eye motion, rapidtide2x has some routines to remove motion timecourses from the data using a GLM procedure (see the --motionfile option). Those might help. 2) You should do slice time correction, otherwise you're going to see stripes in the slice select direction. While these carry information (like, whether you have your slice order right!), it's not interesting to your application. 3) I typically do some spatial filtering (in rapidtide). For whole head high res multi band fMRI, I usually use 1.5-2mm of smoothing (-f 1.5 to -f 2). I don't know what would be best for your data - maybe start with -f 0.5?
The main problem with restricted field of view data with heterogenous tissue (eyes, and the muscle surrounding the eyes) is that there may be multiple very distinct blood supplies with different delays. This can make deriving the initial regressor fail. Do you have any way to generate an eye mask? If so, use that for the --meanmask and --corrmask - that will limit the initial regressor estimation, and further refinement, to the areas you have masked off. Also use the --pickleft option, which will select the leftmost (shortest delay) delay peak as the zero time if the delay histogram is multipeaked.
See if that helps. If you're still having issues, if you send a dataset I'd be happy to play with it a bit to see if I can help optimize it.
Blaise
Dear Blaise,
Thanks for sharing your package. I would like to used rapidtide2 to estimate blood arrival times in the visual cortex. I have a 7T task dataset with relatively high temporal (0.7s) and spatial (0.8mm iso) resolution, but with limited FOV (15 coronal slices covering 12mm). For preprocessing I've only motion corrected the data. Running
rapidtide2 in outon my dataset worked, but results look very much like random noise.I take from the documentation in the repository that rapidtide doesn't care whether we're dealing with task or resting datasets:
But I wonder what the optimal preprocessing would be, and whether there is an optimal set of acquisition parameters that I should use to improve the estimation. Perhaps an additional resting-state scan with a larger brain coverage would be more appropriate, or even collecting an external regressor?
I couldn't find pointers to issues with acquisition parameters in the documentation. Any pointers would be welcome.