Dear Arnaud,
Dear Arnaud,
> Correct me if I'm wrong: looking at the FSL mailing list I've seen that if you acquire an odd number of slice (z axis) you need to remove one slice (top or bottom) in order to use topup.
>
> There is one other solution which is to modify the Subsampling level 2,2,2,2,2,1,1,1,1 to 1,1,1,1,1,1,1,1,1. It is way slower than removing one slice and I'm not sure the result will be as good as removing one slice.
>
> My question is: can I consider adding a slice of zeros instead of removing one slice if my FoV is too tight ?
I think your best solution is to do what you suggest with the subsampling levels. If you also change the --lambda values for the first 5 steps you should get just as good results, and all you pay will be execution time.
I've run topup with and without removing the top slice. I've got 3b0s ap and 9 b0s pa. It took 1h (after removing one slice) and 4h (without removing the top slice) to run topup. The results were similar (even slices gives smoother results) but the resulting images are cropped top and bottom (+/- 1-2 slices). Is this "normal" ? Is there a better way to run topup ?it sounds like some serious overkill to run topup on 3+9 volumes. My experience from any half decent SNR data is that 1+1 is usually enough, possibly 2+2 if you are pressed for SNR. That would reduce your execution time a lot.
The cropping comes from topup only considering the estimated field valid for voxels that fall within the FOV for all volumes in the input. Consider for example if you have two volumes (one each in AP and PA) and the subject slides down (translation in the z-direction) two slices between the AP and the PA. That means that in the PA we have not acquired the two bottom slices that we acquired in the AP, and hence we have no way of knowing what the field is in those two slices. topup will then set the output to zero for those two slices.
For getting the --lambda values ~right I suggest you do a test run with 2,2,2,2,2,1,1,1,1 and 1,1,1,1,1,1,1,1,1 for a subject where you have removed one slice and use trial and error to change --lambda so that you get ~the same reduction in SSD for the first five steps in both cases.
Thank you for your suggestion, I think it will be complicated to change lambdas since I've no idea of how each of them will have an impact on the resulting images. Especially if each trial takes 4h to run.I will run some tests so I can make a .cnf file also for the case when people have an odd number of slices.
Jesper
Thank you again for your helpArnaud
Jesper
>
> Thank you in advance for your help.
>
> Arnaud Bore
--