 |
 |
Re-posting my questions from last Thursday in the hope of getting some pointers on how to progress with this... Best, Matthias --------------------------------------------------------------------------------------------------------------- Dear Experts, I've started using the cuda implementation of eddy (FSL 5.0.11 with slice-to-volume corrections) on various data sets collected previously in our lab (mostly single shell data collected without multi-band factor) with generally great results in terms of improved CNR (as measured by the very useful eddy_qc tools) and improved waytotals when running probtrackx2. I ran some tests comparing the performance of eddy_cuda with different settings/flags (different inputs for --niter and --s2v_niter) on a data set of autistic adolescents that contains a lot of motion and ended up with the same settings as recommended for "a data set with lots of movement..." in the eddy User's Guide (--niter=8 --fwhm=10,8,4,2,0,0,0,0 --repol --s2v_niter=5 --s2v_lambda=1 --s2v_interp=trilinear -> I ran additional iterations and found that the mss did not significantly go down after the number of iterations specified here.) My specific questions are the following: Do you have any recommendations for eddy_cuda settings to use for a) a pediatric data set with a fair amount of motion, though less than what one would expect in a data set with "lots of movement"? b) typical adult data sets of disordered populations that might display more than the typical amount of motion (e.g., individuals with a fluency disorder such as stuttering)? ... other than using the default settings where available (--s2v_niter=5)? Based on my reading of the Andersson et al. 2017 paper, a reasonable approach might be to always use slice-to-volume correction with the --mporder flag set to 2 (for modeling movement with 18 degrees of freedom -> 6 rigid-body parameters * 3) and possibly using --mporder=1 for multi-band data (-> which would equate modeling movement with 12 DOF)? I've seen some recommendations for using fwhm=10,0,0,0,0 when one expects some large movement in the data set. Does using such a big kernel for the first iteration pose any risk of introducing variance in subjects with very little motion? Finally, should one remove any interspersed b0 images with pronounced slice dropouts due to motion (e.g., for the subject with the worst motion in afore-mentioned autism data set) prior to running eddy or does the quality of the b0 images not have much of an effect on eddy? Thank you very much for any advice on this! Best, Matthias ######################################################################## To unsubscribe from the FSL list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1