i run flirt with 12 dof to register diffusion datasets to their T1 datasets and i noticed that for the brainstem (pons) the results are not as good as for the brain. does non linear (fnirt) improve the registration results ? or is there any other mean adapted for this purpose ?