ANTs

ANTs (Advanced Normalization Tools) is a software package that can be used to perform image registration with variable transformations (elastic, diffeomorphic, diffeomorphisms, unbiased) and similarity metrics (landmarks, cross-correlation, mutual information, etc). The source code can be found at: https://github.com/ANTsX/ANTs

Installation

Installation instructions can be found here: https://github.com/ANTsX/ANTs/wiki/Compiling-ANTs-on-Linux-and-Mac-OS

The installation takes a while (could be a few hours).

Post installation, the environment variables PATH and ANTSPATH have to be set. On linux, this can be done by adding the following lines to ~/.bash_profile:

export ANTSPATH=/opt/ANTs/bin/ 
export PATH=${​​​ANTSPATH}​​​​​​​​​​:$PATH

To update the changes type this command in terminal:

source ~/.bash_profile

Registering an Image 

antsRegistration -d 3 \                                       #3D image
-r [anatomical.nii,meanb0.nii,1] \
-m MI[anatomical.nii,meanb0.nii,1,32] \                       #similarity metric: mutual information​ 
-t Affine[0.1] \                                              #Transform 1: Affine transform, step size = 0.1​ 
-c 10000x10000x10000x10000x10000 \                            #convergence factors 
-s 0.8x0.6x0.4x0.2x0mm \                                      #smoothing sigmas 
-f 5x4x3x2x1 \                                                #shrink factors 
-l 1 \
-r [anatomical.nii,meanb0.nii,1] \
-m MI[anatomical.nii,meanb0.nii,1,32] \
-t SyN[0.1] \                                                 #Transform 2: Symmetric diffeomorphic registration​ 
-c 50x35x15 \
-f 3x2x1 \
-s 0.4x0.2x0mm \
-n BSpline \                                                  #3rd order B-spline interpolation​ 
-u true \
-l 1 \
-o [transform,b0_warped.nii.gz,b0_inversewarped.nii.gz]       #output: "transform" produces 3 registration transform files: transform0GenericAffine.mat, transform1Warp.nii.gz, transform1InverseWarp.nii.gz

#the registered images are: b0_warped.nii.gz and b0_inversewarped.nii.gz (which is the anatomical registered to the b0)
  • Smoothing sigmas (-s option) may be scaled according to the DWI resolution (for the code above the resolution was 0.2x0.2x0.5 mm).
  • The resolution of the output registered images will be the same as anatomical image.

Applying Registration Transforms

This is an example to register DWI data (4D image) to the anatomical data by applying the registration transforms acquired with antsRegistration:

antsApplyTransforms -d 3 \
-e 3 \                                     #specifies input image type: the default is scalar (0); the options are 0/1/2/3/4 which corresponds to scalar/vector/tensor/time-series/multichannel
-i DWI.nii.gz \                            #input image
-r meanb0.nii \                            #reference image: the output image will have the same resolution as the reference
-o DWI_reg.nii.gz \                        #output image: registered DWI
-n BSpline \
-t [transform1Warp.nii.gz] \               #transforms to be applied
-t [transform0GenericAffine.mat,0]
  • ANTs applies transforms from the end to the beginning of the command line, so in this case, the Affine transform will be applied first
  • To deform the anatomical to the mean b0 space, the inverse warp nifti file is used and the affine transform command is changed to:
    • -t [transform0GenericAffine.mat,1]
  • Identity Transform: If the -t option is omitted in antsApplyTransforms, the “identity transform” is applied, which can be useful if only the resolution is being changed to match the resolution of the “reference image.”
  • Similarity metrics (such as MI, CC, etc.) of the registered image can be computed via the MeasureImageSimilarity command:
MeasureImageSimilarity -d 3 -m MI[anatomical.nii,DWI_reg.nii.gz,1,32]