This program is designed to help line up the major axes of an
  anatomical volume with those of the volumetric field of view in
  which it sits.  A benefit of this would be to have slices that are
  more parallel to standard viewing planes and/or a bit more
  uniform/standardized across a group.  This may be useful, for
  example, if you want to use an anatomical for registration of other
  data (e.g., diffusion data going to be processed using TORTOISE, esp
  if coloring by local orientation), and you want *that* data to be
  more regularly oriented for viewing, as well.

  This program works by registering (affinely, 9 DOF) the input volume
  to some user-defined reference image, but in the end then only
  applying the translation+rotation parts of the registration to the
  input volume.  Before the registration is done, some polishing of
  the input volume is performed, in order to remove outliers, but none
  of these steps are applied to the output volume itself. Thus, the
  idea is to distort the original brain as little as possible (NB:
  smoothing will occur as part of this process, e.g., due to rotation
  and any regridding), just to reorient it in space.  The output
  volume can be up/downsampled at the same time, if desired.

  You probably *wouldn't* want to use this if your anatomical data set
  really fills up its volume (i.e., has no space to rotate/resituation
  itself).  In that case, you might want to pre-zeropad the volume?


  Ver. 4.3 (PA Taylor, Feb 12, 2019)



  This script has two *required* arguments ('-inset ...' and '-refset
  ...'), and the rest are optional:

  $ fat_proc_axialize_anat  \
        -inset   IN_FILE                      \
        -refset  REF_FILE                     \
        {-mode_t2w | -mode_t1w}               \
        -prefix PPP                           \
        {-workdir  WWW}                       \
        {-out_match_ref}                      \
        {-do_ceil_out}                        \
        {-extra_al_wtmask WT }                \
        {-extra_al_cost CC}                   \
        {-extra_al_inps II}                   \
        {-extra_al_opts SS}                   \
        {-focus_mask MMM}                     \
        {-focus_by_ss}                        \
        {-remove_inf_sli II}                  \
        {-pre_align_center_mass}              \
        {-pre_center_mass}                    \
        {-post_lr_symm}                       \
        {-no_pre_lr_symm}                     \
        {-no_clean}                           \
        {-qc1_ulay_range UMIN UMAX}           \
        {-no_qc_view}                         \
        {-qc_prefix QCP}


  -inset  IN_FILE  :is the full name of the input anatomical volume.
  -refset REF_FILE :is the full name of the reference volume, such as
                    TT or MNI or something (probably you want to match
                    the contrast of your anatomical INFILE, whether
                    there is a skull or not, etc.).

  -prefix  PPP     :output prefix for files and snapshots (required).

  -out_match_ref   :switch to have the final output volume be in the same
                    'space' (FOV, spatial resolution) as the REF_FILE.
                    Might be useful for standardizing the reference
                    output across a group, or at least centering the brain
                    in the FOV. (This applies a '-master REF_FILE' to the
                    final 3dAllineate in the script.)

  -mode_t2w        :switch to alter some intermediate steps (turn off
                    unifizing and raise voxel ceiling threshold).
                    This is particularly useful (= essential,
                    probably) when dealing with a (adult) T2w image,
                    which tends to be bright in the CSF and darker in
                    other tissues; default options are for dealing
                    with (adult) T1w brains, where the opposite is the
  -mode_t1w        :similar to the preceding option, but specifying an
                    image with (human, adult) t1w-like contrast
                    has been input.
     NB ---> one of these 'mode_*' setting MUST be picked.

  -do_ceil_out     :for the final output volume, apply a ceiling based
                    on the 98%ile value from within an automasked volume
                    of the dset.  This reduces the influence of (sometimes
                    very) large spiky voxels.  Seems like it might be
                    useful for later alignments.

  -extra_al_wtmask WT:
                    Axialization is generally based on an overall
                    whole brain alignment.  If you want, however, you
                    can add extra emphasis to part of the weight mask WT
                    for deciding what is good alignment.  For example,
                    you might make a WB mask of values ~1 and a make
                    a subcortical volume have larger values ~5 or so, so
                    that that part of the brain's alignment carries more
                    weight (in this example, behaving more like AC-PC
                    alignment, potentially).

 -extra_al_cost CC :specify a cost function for 3dAllineate to work
                    with (default is 'lpa'; one might investigate
                    'lpc', esp. if contrasts differ between the
                    IN_FILE and REF_FILE, or 'nmi').
 -extra_al_inps II :specify extra options when *calculating* the warp
                    with 3dAllineate.  These could be any
                    option+argument combination from the 3dAllineate
                    helpfile (except the cost function would be done
                    with "-extra_al_cost CC").
 -extra_al_opts SS :specify extra output options when *applying* the
                    warp with 3dAllineate at the end.  One I could see
                    being useful would be having "-newgrid X", where X
                    is desired final resolution of the data.

  -focus_by_ss     :make a mask by simply skullstripping input data
                    set, which gets applied early on to focus the
                    processing+alignment; the final, axialized volume
                    will not have the mask applied, it's just used to
                    help get rid of non-brain garbage.  Might be very
                    useful if the input volume has lots of non-brain
  -focus_mask MMM  :input a mask of the inset that gets applied early
                    on to focus the processing+alignment; the final,
                    axialized volume will not have the mask applied,
                    it's just used to help get rid of non-brain
                    garbage.  Note: before application, MMM gets binarized
                    to 1 where MMM>0 and 0 elsewhere.

-remove_inf_sli II :sometimes data is acquired with lots of nonbrain
                    volume in the FOV, particularly neck and things like
                    that.  While necks are important, they also might
                    move the center of mass estimate of the brain
                    far lower than it should be.  You can get rid of this
                    by applying this option, to remove 'II' number of
                    slices from the inferior part of the FOV.

  -pre_align_center_mass :
                    pre-align the centers of mass of the volumes; might help
                    in conjunction with '-remove_inf_sli ...' above, when
                    there is a large amount of non-brain material
  -pre_center_mass :pre-recenter input center of mass to (0, 0, 0); probably
                    the '-pre_align_center_mass' would be more useful, but
                    this can be used similarly.

  -no_pre_lr_symm  :a pre-alignment left-right symmetrization is
                    performed by default, but you can turn it off if you
                    desire (probably wouldn't want to in most cases,
                    unless *weird* stuff were happening).
  -post_lr_symm    :a post-alignment left-right symmetrization can be
                    added, if desired.

  -workdir WWW     :the name of the working subdirectory in the output
                    directory can be specified
                    (default: __WORKING_axialize_anat).

  -no_clean        :is an optional switch to NOT remove working
                    directory '__WORKING_axialize_anat';
                    (default: remove working dir).

  -no_cmd_out      :by default, a copy of the command and file location
                    from which it is run is dumped into the WORKDIR
                    (file name: 'PREFIX*_cmd.txt').
                    If you don't want this to happen, then use this

  -qc1_ulay_range UMIN UMAX
                   :provide a min (UMIN) and max (UMAX) range for
                    underlay grayscale bar (black=UMIN; white=UMAX).
                    For QC visualization only-- does not affect the
                    actual MRI data files.

  -no_qc_view      :turn off default+automatic QC image saving/viewing
                    (whyyy would you do this?).
  -qc_prefix QCP   :provide a prefix for the QC stuff, separate from
                    the PREFIX above.



    PREFIX.nii.gz   :an anatomical data set that is *hopefully*
                     regularly situated within its FOV volume.  Thus,
                     the axial slices would sit nicely within a given
                     view window, etc.

    WORKDIR         :the working directory with intermediate files, so
                     you can follow along with the process and possibly
                     troubleshoot a bit if things go awry (what are the
                     odds of *that* happening?).



   fat_proc_axialize_anat  \
        -inset  SUB001/ANATOM/T1.nii.gz                            \
        -refset /somewhere/mni_icbm152_t1_tal_nlin_sym_09a_MSKD.nii.gz \
        -mode_t1w                                                  \
        -extra_al_opts "-newgrid 1.0"                              \
        -focus_by_ss                                               \
        -prefix t2w_axlz


   fat_proc_axialize_anat  \
     -inset  SUB001/ANATOM/T2.nii.gz                            \
     -refset /somewhere/mni_icbm152_t2_tal_nlin_sym_09a.nii.gz  \
     -extra_al_wtmask mni_icbm152_t2_relx_tal_nlin_sym_09a_ACPC_wtell.nii.gz \
     -mode_t2w                                                  \
     -prefix t2w_axlz


    + When analyzing adult T1w data, using the following option might
      be useful:
         -extra_al_inps "-nomask"
      Using this, 3dAllineate won't try to mask a subregion for
      warping/alignment, and I often find this helpful for T1w volumes.

    + If the input volume has lots of non-brain material (lots of neck or
      even shoulders included), then consider "-focus_by_ss" or
      "-focus_mask ..".  It helps when trying to roughly align vols initially
      with center of mass, esp. if using "-pre_align_center_mass".  Also, for
      T1w vols this might be particularly effective.

    + For centering data, using the '-out_match_ref' switch might be
      useful; it might also somewhat, veeeery roughly help standardize
      a group of subjects' data in terms of spatial resolution, centering
      in FOV, etc.

    + To try to get something closer to AC-PC alignment, one can add in a
      weight mask with '-extra_al_wtmask ...' that has the ~subcortical
      region given extra weight.