Hi Franck,

using a CSF-suppressed acquisition would be a good option, but strongly limits your slices number, and reduces signal-to-noise ratio. If you use just the CSF-suppressed b0 image with identical sequence parameters before you run the DW series, you might observe head motion effects (but you said you use head fixation, maybe it's neglegible, or can better be handled than distortions from eddy currents).
I didn't make good experiences when using the averaged b1000-image, probably for good reasons: the averaged image contains all geometric distortions, so borders are poorly defined.

I have run the sequence with a phantom where "CSF" is not a problem, corrected with the b0-image as reference, and identified the DW directions that showed the least difference between uncorrected and corrected outlines. However, distortions depend on distance from coil centre, slices tilting, actual head position, coil load etc. so there won't be _the_ best reference volume.

Cheers,
Jochen



>>> Franck Lamberton <[log in to unmask]> 10.05.07 18.42 Uhr >>>
Thanks a lot for your response

Jochen Hirsch a écrit :
> Hi Franck,
>
> I haven't found a hint in your e-mail which image volume you specify as
> reference image.
>   
I used the following command :

eddy_correct input_colume_4D output_volume_4D 0

So, I use the b0-volume as a reference.

I understand that the high CSF signal can lead to wrong corrections. I 
have already noticed that the external edge of the brain in 
b1000-volumes (and thus, edge of the white matter) are moved  toward the 
edge of the CSF in b0-volume (and thus, edge of CSF).
Do you suggest to use another dedicated acquisition, with the same 
parameters as the b0-volume but with an additional pre-inversion in 
order to suppress signal from CSF ?
Or maybe using the mean volume of all b1000-volumes as a reference...

> But correction for eddy currents distortions will probably fail if you
> choose the b0-volume due to the high CSF signal which is not present in
> b1000-images. 
>
> Regards,
> Jochen
>
>
>
>   
>>>> Lamberton Franck <[log in to unmask]> 10.05.07 11.13 Uhr >>>
>>>>         
> Hi,
>
> I worked with DTI a few weeks ago and used FDT in order to compute FA
> (last
> version of FSL running on Linux FC4).
> Recently, I acquired a set of diffusion weigthed images on a subject
> with
> minimal head movement (head control) :
>         40 slices
>         1 b=0 s/mm2 + 32 different encoding directions at b=1000 s/mm2
>         resolution = 2x2x2 mm3
>
> Using the 'Movie mode' of fslview shows that very little distortion
> occured
> between successive acquisitions (they seem  to be related to
> eddy-currents).
> After applying 'eddy_correct', significant geometrical distortion
> appeared
> in the se ries that were not present in the first place.
> I also used avwmaths++ command with the option -Tstd in order to compute
> the
> standard deviation of the series. Signifi cant geometrical distortion
> introduced by eddy_correct induced high std values all arround the
> brain.
>
> I edited the script 'eddy_correct' in order to test different options of
> FLIRT, cost function (-searchcost) and no ang ular research limitation
> (-nosearch). 
> In all cases, spurious geometrical distortions remain the same after
> applying the modified 'eddy_correct'.
>
> I also used the option "-interp nearestneighbour" in order to better
> compare
> uncorrected and corrected series (less blurring). To do this, I used
> 'avwmaths++' with '-Tstd' option in order to compute the std volume of
> only
> corrected diffusion weighetd volumes. After correction, greater values
> appear at the periphery of the brain.
>
> Does anyone have already seen a similar effect of 'eddy_correct' ?
> What about using  -refweight or -inweight of FLIRT ? How can I use this
> two
> options ?
>
> Thanks for any response
>
> Franck
>
>