mri_synthmorph-register

NAME
        mri_synthmorph-register - register 3D brain images without preprocessing

SYNOPSIS
        mri_synthmorph register [options] moving fixed

DESCRIPTION
        SynthMorph is a deep-learning tool for symmetric, acquisition-agnostic
        registration of brain MRI with any volume size, resolution, and
        orientation. The registration is anatomy-aware, removing the need for
        skull-stripping, and you can control the warp smoothness.

        SynthMorph registers a moving (source) image to a fixed (target) image.
        Their geometries can differ. The options are as follows:

        -m model
                Transformation model (joint, deform, affine, rigid). Defaults to
                joint. Joint includes affine and deformable but differs from
                running both in sequence in that it applies the deformable step
                in an affine mid-space to guarantee symmetric joint transforms.
                Deformable assumes prior affine alignment or initialization with
                -i.

        -o image
                Save moving registered to fixed.

        -O image
                Save fixed registered to moving.

        -H
                Update the voxel-to-world matrix instead of resampling when
                saving images with -o and -O. For matrix transforms only. Not
                all software supports headers with shear from affine
                registration.

        -t trans
                Save the transform from moving to fixed, including any
                initialization.

        -T trans
                Save the transform from fixed to moving, including any
                initialization.

        -i trans
                Apply an initial matrix transform to moving before the
                registration.

        -M
                Apply half the initial matrix transform to moving and (the
                inverse of) the other half to fixed, for symmetry. This will
                make running the deformable after an affine step equivalent to
                joint registration. Requires -i.

        -j threads
                Number of TensorFlow threads. System default if unspecified.

        -g
                Use the GPU in environment variable CUDA_VISIBLE_DEVICES or GPU
                0 if the variable is unset or empty.

        -r lambda
                Regularization parameter in the open interval (0, 1) for
                deformable registration. Higher values lead to smoother warps.
                Defaults to 0.5.

        -n steps
                Integration steps for deformable registration. Lower numbers
                improve speed and memory use but can lead to inaccuracies and
                folding voxels. Defaults to 7. Should not be less than 5.

        -e extent
                Isotropic extent of the registration space in unit voxels (192,
                256). Lower values improve speed and memory use but may crop the
                anatomy of interest. Defaults to 256.

        -w weights
                Use alternative model weights, exclusively. Repeat the flag to
                set affine and deformable weights for joint registration, or the
                result will disappoint.

        -h
                Print this help text and exit.

ENVIRONMENT
        The following environment variables affect mri_synthmorph-register:

        CUDA_VISIBLE_DEVICES
                Use a specific GPU. If unset or empty, passing -g will select
                GPU 0. Ignored without -g.

        FREESURFER_HOME
                Load model weights from directory FREESURFER_HOME/models.
                Ignored when specifying weights with -w.

EXAMPLES
        Joint affine-deformable registration, saving the moved image:
                # mri_synthmorph register -o out.nii mov.nii fix.nii

        Joint registration at 25% warp smoothness:
                # mri_synthmorph register -r 0.25 -o out.nii mov.nii fix.nii

        Affine registration saving the transform:
                # mri_synthmorph register -m affine -t aff.lta mov.nii.gz
                fix.nii.gz

        Deformable registration only, assuming prior affine alignment:
                # mri_synthmorph register -m deform -t def.mgz mov.mgz fix.mgz

        Deformable step initialized with an affine transform:
                # mri_synthmorph reg -m def -i aff.lta -o out.mgz mov.mgz
                fix.mgz

        Rigid registration, setting the output image header (no resampling):
                # mri_synthmorph register -m rigid -Ho out.mgz mov.mgz fix.mgz