Dear Burkhard, We have not tried to optimise eddy_correct for this situation (limited FOV) although I am surprised that it fails. You might want to try a different cost function. If you take a copy of the file $FSLDIR/bin/eddy_correct then you can modify the call to flirt (near the end) and try other cost functions to see if they do better. Also check that your reference volume is a good choice. All the best, Mark On 3 Aug 2011, at 14:54, Burkhard Maedler wrote: > Hi FSL-forum, > I try to use eddy_correct on a subset of sagital DTI-data. Only the centre of the brain (main part of corpus callosum; 10 slices to either side) has been included. > It looks "eddy_correct" has difficulties with handling the affine transformation of such a subset since it cannot perform a decent ECC for each gradient encoding. > Some transformation matrices are completely screwed, e.g. > > regular (well behaved transform) matrix would look like this: > Final result: > 1.006631 -0.000593 0.000401 0.140801 > 0.002273 1.002451 -0.006194 0.016195 > -0.001417 -0.001287 1.008597 -0.137913 > 0.000000 0.000000 0.000000 1.000000 > > but for some volumes (gradient orientations) I get matrices like this: > Final result: > 0.566810 0.150488 -0.212689 67.358504 > -0.107210 0.875570 -0.004303 8.674744 > 0.138302 0.020314 0.880910 22.970297 > 0.000000 0.000000 0.000000 1.000000 > > which leads to completely useless results. > > I don't know why ECC is not able to correctly register the individual volumes. The susceptibility induced distortions > between all volumes is minor and other registration software can perfectly handle this. When I also use a complete > sagital set of data (60 slices accross the brain) this behaviour does not occur. > > Any help appreciated, as always. > > Burkhard. >