Dear Amelia,

Dear FSL and eddy experts,
We are trying using the new version of eddy. Out sequence is a multishell forward-reverse sequence with 33 directions @ b=700, 64 directions @ b=1000 and 100 directions @ b=2500, plus 13 b0s for a total of 210 P>>A volumes and 210 A>>P volumes. Our gradient table is a q-ball table.
In our CENTOS OS 6, the eddy_openmp seems to work fine and here are the outputs:
And also:
As per previous posts, we used the -- resamp=lsr option to average the 210 P>>A and the 210 A>>P. So our fslnvols EC.nii is now 210. However, the EC.eddy_rotated_bvecs  still has 420 directions. Can you confirm that the best option available now will be to average the 2 main components of this merged and corrected bvec file?


In this regard, future version of eddy are going to offer ?better? solution for this?

No, what would a better solution be? If there has been a large rotation in between the AP-PA pair, then one is clearly combining apples and pears by using the --lsr option. BUT, a typical average solid angle between neighbouring diffusion gradients would be ~15 degrees so if the rotation between the images in the pair is ~a couple of degrees it still makes sense, and one will recover some resolution from distorted regions. If on the other hand the rotation between the the images in the pair is ~ten degrees it will not make sense and it is better to keep all the 420 volumes using jac resampling.

Also, we tested the ‘standard’ version of eddy (described as part of FSL 5.0.8) that comes with FSL 5.0.9 package. With this version, we have been able to run the  -- resamp=lsr option, but, of course not the –fep option. However, along with EC.nii, EC.eddy_outlier_report and EC.eddy_parameters, we found an extra image:

Yes, this is an image that is always produced and which contains the original images (i.e. no correction for movement, EC or anything). It is mainly there as a way to check on the outlier report. 

With eddy 5.0.8, the lsr option did not seem to work (i.e., 420 volumes); however EPI distortions were effectively reduced in the EC.nii, but not in the EC.eddy_outlier_free_data.nii. The name of this second output suggests the outliers  reported in EC.eddy_outlier_report had been removed. Is this correct? Is there a way to test this? We tried a simple subtraction, but given that the 1st, but not the 2nd , image is ‘corrected’, it was not possible to make any inference. Also, does the –fep option (in eddy_openmp) ‘fill empty planes in x- or y- directions’ after having removed the outliers? In other words, outliers are already removed in the EC.nii output of eddy_openmp?

No, in this version we are not removing the outliers for the output. The outlier detection/replacement has been in a state of flux for some time so we weren’t sure of the best parameters, or even if it was a really good idea or not. We have recently been getting access to simulated data from UCL and that has really helped us refine the outlier detection and to get it right. So for the next release we will feel confident to include that feature.


Any help will be greatly appreciated! Thanks a lot!! Amelia
Amelia Versace, MD
Loeffler Building, room 215
121 Meyran Avenue
Phone: 412.383.8131