Dear Maria,
>
> Thank you for the useful information! I am looking forward to see the plots. I have some smaller questions:
>
> a. Are these eddy options still supported?:
>
> 1. considering that I lack reversed phase encoding direction acquisition, I'd like to use the --slm=linear option. Is fsl 5.0.7/5.0.8 still supporting this option? I get an error that suggest the option might not be supported.
it should still be supported. Can you please let me know what kind of error message you get?
Having said that, the simulations done by Mark Graham et al. in their paper suggests that it works quite well also without reversed phase encoding and they didn’t see any big advantage using --slm=linear.
>
> 2. I wanted to run eddy with --fwhm=10,0,0,0,0 as per http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/EDDY/UsersGuide/#A--fwhm, just to see whether there is a difference between the motion correction obtained with --fwhm=0 (the default) and the suggested --fwhm=10,0,0,0,0. However, once again, I get an error that suggest the option might not be supported.
Again, it should be supported.
>
> b. I am assuming that one would choose the number of iterations for eddy based on a convergence criterion. Is this the mss displayed for each iteration? If so, is it correct to say that the first series of mss values is related to motion and the second to eddy correction? What are values considered to be sufficient as a stopping criteria? In my case, the mss values from the 1st round of iterations slightly decrease from 3.1 to 2.224 in 5 iterations but for the 2nd round of iterations the total mss remains constant to 247.474. Just as a side remark: I believe that previously when I had run mcflirt before eddy I had gotten much lower total mss value for the 1st round of iterations, i.e., aprox. 0.7.
In all but the worst motion cases have I seen convergence within 5 iteration with --fwhm=0. The recommendation to try --fwhm=10,0,0,0,0 is for cases with really a lot of movement (we use it for babies for example).
You should mostly ignore absolute numbers when it comes to mss cost-functions since many packages will scale their data internally (as does eddy). For eddy it is even hard to assess the relative (between iterations) mss as there are two counteracting things taking place. The generative model gets sharper (which will increase the mss) and the data gets corrected for movement/distortions (which will decrease the mss).
When I say that things have generally converged after 5 iterations that is based on looking at the iteration-to-iteration changes of parameters estimates, the validation we did in the paper and additional validations we have done since based on Mark Graham’s simulations.
> Where could I get the rmsfile=${1}.eddy_movement_rms file from? Eddy is not currently providing it. How is this calculated?
I wasn’t sure if the released version generated those or not, which is why the script checks for its existence. The next version will.
Jesper
>
> Best
> --
> Maria
|