EPI registration

[JW1]

Dear AFNI experts,

Below is the script we used to preprocess my EPI (task-based) data. We are looking for pre-processed but non-statistically analyzed data so we removed regress options.

afni_proc.py
-anat_has_skull yes
-blocks despike tshift align tlrc volreg mask scale regress
-copy_anat mprage/mprage.{subj}.{cond}+orig
-dsets /Users/lab/Desktop/AES2016_FNT/{subj}/{subj}.{cond}/FNT/fnt.{subj}.{cond}+orig \ -out_dir Purcell_result_without_regress_glt \ -script py_result_{subj}.{cond}.script \ -scr_overwrite \ -subj_id {subj}
-verb 2
-align_opts_aea -deoblique off -giant_move -Allineate_opts -usetemp
-tlrc_base MNI152_T1_2009c+tlrc
-tlrc_opts_at -ok_notice -dxyz 1 -pad_base 100 -init_xform AUTO_CENTER
-tlrc_suffix _MNI152
-volreg_align_to third
-volreg_interp -Fourier
-volreg_opts_vr -twopass -noclip -dfile volreg.{subj}.{cond}.1d
-volreg_tlrc_warp
-volreg_warp_dxyz 3

When I checked the data quality, I see some discrepancies between the volreg data and MNI space (see attached picture). Could you please let me know how to address this gap?

Sorry, I accidentally posted two same messages. Could you please delete the above?

Best,
JW

[ptaylor]

Hi, JW-

I think your alignment from anatomical to standard space is just being done with linear affine registration, since you have not specified any nonlinear alignment option. It will give reasonable alignment, but it simply can’t get a really detailed fit between two different subject brains (i.e., your subj and the MNI template in this case). Nonlinear alignment is needed for the higher quality fit.

Running @SSwarper before using afni_proc.py and providing the warp results would be the method of choice, at current. You can also see the afni_proc.py help for this opt:

-tlrc_NL_warp

… but really, using @SSwarper seems to give really nice results in general.

Note that at present, if you don’t include the “regress” block, you won’t get the fun QC HTML, which will help you trouble shoot anything. You might find it useful to still include, therefore, even if your regression/modelling is done elsewhere.

–pt