Dear Angela,
> we realized this has been a topic before, but we so far haven't found a satisfying procedure.
> We have DTI data with interspersed b0 images (b0 - 20 x b1000 - b0 - 20 x b1000 - b0 - 20 x b1000 - 9 x b0).
> In order to exploit all information we've available, we thought about first eddy correcting by splitting the dataset into three sets, and correct each with the nearest b0 before as a reference.
> For this we would use the old eddy_correct (before doing a BET we'd like to eddy correct, and the new eddy needs a brain mask as input) .
> Then we would merge the sets, and eddy correct the whole set with reference to the average b0 (taking all b0), applying the new eddy function.
> We also thought about whether it's necessary to run a first eddy in the beginning with the first b0 as the reference.
>
> The b0 are all phase encoded in the same direction, so we'd not use the topup-tool.
>
> Does this make sense, is it not necessary or are there better ways to do this?
I think you are making things too complicated. Firstly, the mask in eddy is only used to get a rough idea what voxels are brain for estimating the GP hyperparameters. If in doubt, just make the mask a little smaller to avoid non-brain voxels.
Secondly, for each iteration of eddy it will re-apply the mask to the current location of the different volumes. After a couple of iterations they tend to be quite well aligned, and hence all volumes will ~be in the same space as the mask.
My recommendation would be to extract the first b0 volume and bet that to get a mask and then just run eddy on the whole data set.
Jesper
> Any help is much appreciated!
>
> Best wishes,
> Angela
|