 |
 |
Hi, If you have substantial distortion in your images then it will be difficult/impossible to get a very good registration with a rigid registration. However, you could try using a cost weighting image that is one in all voxels except those in *and around* the ventricles (you need to include some of the surrounding tissue to get the border) where the values should be large (say 1000, although try 100, 10000 and 100000 too). This will have the effect of making the registration concentrate much more on the ventricle alignment, at the expense of cortical alignment. If you can't get a good registration this way then you probably have too much distortion in the images, and really need a fieldmap. To run FLIRT with the cost function weighting image you need to make the image yourself (by hand in FSLView or via fslmaths and segmentation tools) and then take your original FLIRT results (the matrix) and use this as the -init argument, together with -nosearch and either -refweight (if the weighting image is aligned with the reference image) or -inweight (if it is aligned with the input image). I would also recommend going from DTI to T1 space, and then inverting the output if you need to transform in the opposite direction. My experience is that the T1 works better as the reference image, especially when using the (default) correlation ratio cost function. All the best, Mark On 8 Feb 2012, at 14:04, Franklin, Crystal G wrote: > I have ran rigid registration in flirt and I am still getting misalignment > around the ventricles. We would like to keep the DTI in native space. > > Thank you, > Crystal > > On 2/7/12 5:54 PM, "Cunningham, Dustin" <[log in to unmask]> > wrote: > >> Hi Crystal, >> >> Usually you can get satisfactory registration between T1 and b0 in the >> same subject using rigid registration in flirt. May I ask why you want to >> register the T1 to DTI and not the other way around? Also, the ventricles >> should look very similar in shape in the T1 and b0 image, which is >> essentially T2 weighted. >> >> Dustin >> ________________________________________ >> From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of >> Franklin, Crystal G [[log in to unmask]] >> Sent: Tuesday, February 07, 2012 4:52 PM >> To: [log in to unmask] >> Subject: [FSL] T1 to DTI Registration >> >> Dear FSL, >> >> I am trying to register my subject's T1 (voxel dimensions of .8x.8x.8) to >> their DTI (voxel dimensions of 1.719x1.719x3). Unfortunately I do not >> have a T2 image or field maps. Do you suggest using the FA image or the >> b0? I have tried using both in flirt. The registration looks pretty >> good until I get around the ventricles. What is the best way to go about >> correcting this misalignment or the best procedure for registering the T1 >> to DTI? >> >> Thank you, >> Crystal >