Hi FSL users,
I am running some tractography with Fdt, using a seed and multiple target masks, which are in native subject T1 space. I made the revelant xfms files using a 6 DOF flirt as instructed at http://www.fmrib.ox.ac.uk/fsl/fdt/fdt_reg.html. However, when viewing the generated tracts, there is quite a degree of mismatch between the tractography and the white matter of the subject's T1 structural.
I suspect that imperfect registration between subjects' T1 and nodif_brain images in the xfms is to blame. Attached is a jpeg of a nodif_brain overlaid with the structural2diffusion image generated using the struct2diff xfm, and there is noticeable inaccuracy in registration, particularly around the ventricles. The subjects are neurodegenerative patients with large ventricles and considerable atrophy, and it seems that a linear transformation as used in flirt is just not able to achieve good registration between structural and diffusion space.
A nonlinear transformation in fnirt (using the .mat obtained in the previous flirt) appears to do the job nicely - however, fnirt does not output a .mat transformation matrix to use in Fdt when transforming seed and target masks to diffusion space.
The only option I can see is to apply the fnirt warps to all of my masks manually, taking them accurately into diffusion space before using them in Fdt. Then, after probtrackx has run, to use the inverse fnirt warp to take any tractography results back accurately into structural space. This seems a little long-winded and I wonder if there is an alternative? Have any other FSL users conducting tractography on neurodegenerative patients encountered a similar problem with imperfect xfms generated using linear registration?
I would be most grateful for any thoughts on the best way to proceed.
Many thanks,
Charlotte Rae
--------------------------------------------------------
Charlotte Rae
MRC Cognition and Brain Sciences Unit
Cambridge, UK
[log in to unmask]
--------------------------------------------------------
|