It's difficult to know exactly what the issue is. Of course, after transformation, data that had been transformed will be shifted relative to the original template.

The cerebellum is a particular problem ecause its relative position and relative size vary so much across subjects, yet the alignment uses an affine transformation based on the whole brain . If you have specific interest in the cerebellum, you might consider alignment of just a cerebellar mask region to a cerebellar template extracted from one of the templates.

Regarding the method to transform data from Talairach space to MNI, there are several ways to do this. The first way is the way you tried with:

3dWarp -tta2mni -prefix dset_mni dset+tlrc

This method uses the Brett transform to transform the data using a two step procedure. Our TT_avg152T1 template was created by the transforming the MNI template using this method (-mni2tta). The "-newgrid" option takes one number to be used as the cubic voxel size of the output. For a 3mm^3 size, use -newgrid 3, for example. Use a size that makes sense for the data you have. We typically do something like the minimum dimension of the input dataset.

Another way is to use the affine transformations more recently recommended by Lancaster and Fox. I have stored a copy of the affine transformation matrices here:

[afni.nimh.nih.gov]

To transform from TLRC space to MNI space, use the MNI FSL transformation matrix (mni_fsl2tal.1D) like this.

3dWarp -matvec_out2in mni_fsl2tal.1D -prefix dset_mni dset+tlrc

There are affine transformations stored there for MNI_SPM and MNI_OTHER transformations if you need those also.

Another way is to use the 12-piece manual transformation that was used to transform the N27 dataset to Talairach space. This applies the reverse transformation and puts the output data onto a template grid. You will need to create the template dataset (maybe using one of the previous methods)

3dfractionize -input dset+tlrc -warp ~/abin/TT_N27+tlrc -prefix dset_mni -template template_mni+tlrc

Yet a fourth way is to align everything again to the template in the space you want. This way is probably the most accurate. In this case, align an anatomical image to the MNI_152 template and then align the epi to which all the other datasets are aligned using align_epi_anat.py. By using the "-child_epi" and "-tlrc_apar" option with your mask, you should be able to get that mask to the MNI space. The interpolation there will be the default. You migh consider applying the transformation separately (take out the -child_epi part then).

@auto_tlrc -base ~/abin/MNI_avg152T1+tlrc -input sb23_mpra+orig
align_epi_anat.py -anat sb23_mpra+orig -epi epi_r03+orig \
-epi_base 6 -child_epi mask_*+orig.HEAD \
-epi2anat -prep_off -suffix _al2anat_mni \
-tlrc_apar sb23_mpra_at+tlrc

3dAllineate -final NN -1Dmatrix_apply epi_r03_al2anat_tlrc_mat.aff12.1D \
-prefix mask_mni mask+orig

Also you might want to take a look at this previous post regarding the transformations between MNI and TLRC spaces.

<[afni.nimh.nih.gov];