bionic (1) WarpImageMultiTransform.1.gz

Provided by: ants_2.2.0-1ubuntu1_amd64 bug

NAME

       WarpImageMultiTransform - part of ANTS registration suite

DESCRIPTION

       Usage:

       WarpImageMultiTransform   ImageDimension   moving_image   output_image    -R   reference_image   --use-NN
       SeriesOfTransformations--(See Below)

              SeriesOfTransformations --- WarpImageMultiTransform can apply,  via  concatenation,  an  unlimited
              number  of  transformations  to  your  data  .   Thus,  SeriesOfTransformations  may be  an Affine
              transform followed by a warp  another affine and then another warp.

       Inverse affine transformations are invoked by calling
              -i MyAffine.txt

       InverseWarps are invoked by passing the InverseWarp.nii.gz
              filename (see below for a note about this).

              Example 1: Mapping a warped image into the reference_image domain by applying abcdWarp.nii.gz  and
              then abcdAffine.txt

       WarpImageMultiTransform 3 moving_image output_image -R reference_image abcdWarp.nii.gz abcdAffine.txt

              Example  2:  To  map the fixed/reference_image warped into the moving_image domain by applying the
              inversion of abcdAffine.txt and then abcdInverseWarp.nii.gz .

       WarpImageMultiTransform   3   reference_image   output_image   -R   moving_image    -i     abcdAffine.txt
       abcdInverseWarp.nii.gz

              Note  that the inverse maps (Ex. 2) are passed to this program in the reverse order of the forward
              maps (Ex. 1).

              This makes sense, geometrically ... see ANTS.pdf for visualization of this syntax.

              Compulsory arguments:

              ImageDimension: 2 or 3 (for 2 or 3 Dimensional registration)

              moving_image: the image to apply the transformation to

              output_image: the resulting image

              Optional arguments:

       -R: reference_image space that you wish to warp INTO.

       --tightest-bounding-box: Computes the tightest bounding box using all the affine transformations. It will
              be overrided by -R reference_image if given.

       --reslice-by-header:  equivalient  to  -i  -mh,  or  -fh  -i  -mh  if  used together with -R. It uses the
              orientation matrix and origin encoded in the image file header.

              It can be used together with -R. This is typically not used together with any other transforms.

       --use-NN: Use Nearest Neighbor Interpolation.

       --use-BSpline: Use 3rd order B-Spline Interpolation.

       --use-ML sigma: Use anti-aliasing interpolation for multi-label  images,  with  Gaussian  smoothing  with
              standard deviation sigma.

              Sigma  can  be  specified  in  physical  or  voxel units, as in Convert3D. It can be a scalar or a
              vector.

       Examples:
              --use-ML 0.4mm    -use-ML 0.8x0.8x0.8vox

       -i: will use the inversion of the following affine transform.

              Other Example  Usages:  Reslice  the  image:  WarpImageMultiTransform  3  Imov.nii.gz  Iout.nii.gz
              --tightest-bounding-box   --reslice-by-header   Reslice   the   image   to   a   reference  image:
              WarpImageMultiTransform  3  Imov.nii.gz   Iout.nii.gz   -R   Iref.nii.gz   --tightest-bounding-box
              --reslice-by-header

              Important  Notes:  Prefixname  "abcd"  without  any  extension  will  use ".nii.gz" by default The
              abcdWarp   and   abcdInverseWarp   do   not   exist.   They   are   formed   on   the   basis   of
              abcd(Inverse)Warp.nii.gz when calling WarpImageMultiTransform, yet you have to use them as if they
              exist.