Hi Ed,
I have the same problem: the brain for the DW images is inflated to match the CSF from the b=0 image. The main problem --I think-- is that eddy_correct uses flirt's default cost function (corratio). Since the contrast in standard DW images (b=1000) is so different from the contrast in the b=0 image, I think you should use normalized mutual information (normmi).
I wrote my own variant of eddy_correct that calls flirt with "-cost normmi". If the default 12 dofs still gives you weird results, I also include a "-dof 6" in my eddy_correct.
HTH,
-Pablo
|