Registration

This tutorial walks through the steps to perform image-based and streamline-based registration using DIPY. Multiple registration methods are available in DIPY.

You can try these methods using your own data; we will be using the data in DIPY. You can check how to fetch the DIPY data.

Image Registration

DIPY’s image registration workflow can be used to register a moving image to a static image by applying different transformations, such as center of mass, translation, and rigid body or full affine (including translation, rotation, scaling and shearing) transformations. During such a registration process, the static image is considered to be the reference, and the moving image is transformed to the space of the static image. Registration methods use some sort of optimization process, and a given metric or criterion (like maximizing the mutual information between the two input images) that is optimized during the process, to achieve the goal.

The DIPY image registration workflow applies the specified type of transformation to the input images, and hence, users are expected to choose the type of transformation that best matches the requirements of their problem. Alternatively, the workflow allows one to perform registration in a progressive manner. For example, using affine registration with progressive set to True will involve center of mass, translation, rigid body and full affine registration; meanwhile, if progressive is set to False for an affine registration, it will include only center of mass and affine registration. The progressive registration will be slower but will improve the quality.

We will first create a directory in which to save the transformed image and the affine matrix (e.g.: image_reg_output):

mkdir image_reg_output

To run the image registration, we need to specify the paths to the static image file, and to the moving image file, followed by the optional arguments. In this case, we will be specifying the type of registration to be performed (transform) and the output directory (out_dir).

To perform center of mass registration, we will call the dipy_align_affine command with the transform parameter set to com e.g.:

dipy_align_affine <path_to_static_file> <path_to_moving_file> --transform "com" --out_dir "image_reg_output"

This command will save the transformed image and the affine matrix to the specified output directory.

If we are to use an affine transformation type during the registration process, we would call the dipy_align_affine command as, e.g.:

dipy_align_affine <path_to_static_file> <path_to_moving_file> --transform "affine" --out_dir "affine_reg_output" --out_affine "affine_reg.txt"

This command will apply an affine transformation on the moving image file, and save the transformed image and the affine matrix to the affine_reg_output directory.

In case you did not specify the output directory, the transformed image file and affine matrix would be saved to the current by default. If you did not specify the name of the output affine matrix, the affine matrix will be saved to a file named affine.txt by default, located in the current directory also by default.

Symmetric Diffeomorphic Registration

Symmetric Diffeomorphic Registration is performed using the Symmetric Normalization (SyN) algorithm proposed by Avants et al.[1] (also implemented in the ANTs software Avants et al.[2]). It is an optimization technique that brings the moving image closer to the static image.

Create a directory in which to save the transformed image (e.g.: syn_reg_output):

mkdir syn_reg_output

To run the symmetric normalization registration method, we need to specify the paths to the static image file, and to the moving image file, followed by optional arguments. In this case, we will be specifying the metric (metric), the output directory (out_dir) and the file name of the output warped image (out_warped). You can use cc (cross correlation), ssd (sum squared differences) or em (expectation-maximization) as metrics.

The symmetric diffeomorphic registration method in DIPY is run through the dipy_align_syn command, e.g.:

dipy_align_syn <path_to_static_file> <path_to_moving_file> --metric "cc" --out_dir "syn_reg_output" --out_warped "syn_reg_warped.nii.gz"

In case you did not specify the output directory, the transformed files would be saved in the current directory by default. If you did not specify the file name of the output warped image, the warped file will be saved as warped_moved.nii.gz by default.

Apply a Transformation

We can apply a transformation computed previously to an image. In order to do so, we need to specify the path of the static image file, moving image file, and transform map file, which is a text(*.txt) file containing the affine matrix for the affine case and a nifti file containing the mapping displacement field in each voxel with this shape (x, y, z, 3, 2) for the diffeomorphic case, followed by optional arguments. In this case, we will be specifying the transform type (transform_type) and the output directory (out_dir).

Create a directory in which to save the transformed files (e.g.: transform_reg_output):

mkdir transform_reg_output

For a diffeomorphic transformation, we would run the command as:

dipy_apply_transform <path_to_static_file> <path_to_moving_file> <path_to_transform_map_file> --transform_type "diffeomorphic" --out_dir "transform_reg_output"

This command will transform the moving image and save the transformed files to the specified output directory.

Streamline-based Registration

Streamline-based registration (SLR) [3] is performed to align bundles of streamlines directly in the space of streamlines. The aim is to align the moving streamlines with the static streamlines.

The workflow for streamline-based registration requires the paths to the static streamlines file, and to the moving streamlines file, followed by optional arguments. In this case, we will be specifying the number of points for discretizing each streamline (nb_pts) and the output directory (out_dir).

Create a directory in which to save the transformed files (e.g.: sl_reg_output):

mkdir sl_reg_output

Then, run the command as:

dipy_slr <path_to_static_file> <path_to_moving_file> --nb_pts 25 --out_dir "sl_reg_output"

This command will perform streamline-based registration and save the transformed files to the specified output directory.

References

[1]

Brian B. Avants, Charles L. Epstein, Murray Grossman, and James C. Gee. Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain. Medical Image Analysis, 12(1):26–41, 2008. Special Issue on The Third International Workshop on Biomedical Image Registration – WBIR 2006. URL: https://doi.org/10.1016/j.media.2007.06.004, doi:10.1016/j.media.2007.06.004.

[2]

Brian B. Avants, Nick Tustison, and Gang Song. Advanced Normalization Tools: V1. 0. The Insight Journal, 2(365):1–35, July 2009.

[3]

Eleftherios Garyfallidis, Omar Ocegueda, Demian Wassermann, and Maxime Descoteaux. Robust and efficient linear registration of white-matter fascicles in the space of streamlines. NeuroImage, 117:124–140, August 2015. URL: https://doi.org/10.1016/j.neuroimage.2015.05.016, doi:10.1016/j.neuroimage.2015.05.016.