Elastix Wrapper

This repository contains the code for the Fiji plugin Elastix for 2D and 3D image registration. The plugin is a wrapper around the image registration suite elastix. Wrapping elastix into Fiji provides the convenience of easy image handling, visualisation and a graphical user interface.

Citation

This plugin:

Tischer, C. (2019). ElastixWrapper: Fiji plugin for 3D image registration with elastix. Zenodo. http://doi.org/10.5281/zenodo.2602549

elastix:

S. Klein, M. Staring, K. Murphy, M.A. Viergever, J.P.W. Pluim, "elastix: a toolbox for intensity based medical image registration," IEEE Transactions on Medical Imaging, vol. 29, no. 1, pp. 196 - 205, January 2010.
D.P. Shamonin, E.E. Bron, B.P.F. Lelieveldt, M. Smits, S. Klein and M. Staring, "Fast Parallel Image Registration on CPU and GPU for Diagnostic Classification of Alzheimer’s Disease", Frontiers in Neuroinformatics, vol. 7, no. 50, pp. 1-15, January 2014.

Learn more

Elastix manual

Installation

Install Fiji and the plugin

Install Fiji: https://fiji.sc/
Add the update site: ElastixWrapper
- The plugins will be available here: [ Plugins > Registration > Elastix ]
- Note: The ElastixWrapper update site has conflicts with the deprecated EMBL-CBA update site, which thus must be disabled.

Install the elastix binary

Windows, MacOS and Linux
- Install elastix: https://elastix.lumc.nl/download.php

Windows specific requirements

Install the corresponding Visual C++: http://www.microsoft.com/en-us/download/details.aspx?id=30679
- See also here: https://github.com/SuperElastix/elastix/wiki/FAQ

Usage instructions

Fiji: [Plugins > Registration > Elastix > Elastix]
Elastix installation directory:
- The folder where you installed elastix, e.g.
- C:\Program Files\elastix_v4.8, or
- /Applications/elastix_macosx64_v4.8/
Fixed image:
- Reference image, may be multi-channel
- The image calibration in x,y,z must be correct!
  - Use [ Image > Properties ] to check this!
Moving image:
- The image to be transformed, must have same number of channels as the fixed image, may have different dimensions and voxel sizes in x,y,z.
Transformation type:
- Translation: 3-D shift
- Euler: 3-D shift and rotation
- Similarity: 3-D shift, rotation, and uniform scaling
- Affine: 3-D shift, rotation, and 3-D scaling
- BSpline: local deformations
- It is recommended to start with a simpler transformation type and then further transform the result (see "Use initial transformation").
Number of iterations:
- Number of iterations to find the best registration in each time-point
- Try 1000 as a starting point
Spatial samples:
- Number of data points that will be used in each iteration for finding the registration
- To use all data point type: full
- Around 3000 often is a good number
- Note that a semicolon separated list can be provided to use a different number of samples for registration at different resolutions (the number of list items must match the number of list items in the below "Gaussian smoothing sigma" parameter). It usually makes sense to use relatively less samples for lower resolutions.
Gaussian smoothing sigma [voxels]
- For each dimension (x,y,z) enter the smoothing sigma.
- Typically, the registration works better with more smoothing.
- You may specify a semicolon separated list of smoothing factors in order to do the registration at different resolutions, e.g. 10,10,10;1,1,1 will first do the registration with a 10 x 10 x 10 down-sampled version of the data and then in a second step at full resolution.
Transformation output file
- Provide a path to a file where the transformation should be saved.
- The file path should end on .txt
- Do NOT save the file in the "Temporary directory for intermediate files" as it may be overwritten there. Save it to some other folder.
- You can then use this transformation in a next run in the below field: "Use initial transformation file".
- Like this you can chain transformations of increasing complexity.
Using masks:
- You can mask either/or the moving or the fixed image.
- The algorithm will then only consider pixels within the mask.
- Using masks can help a lot, because it avoids futile computations on empty pixels.
- The masks must have the same dimensions as your fixed and moving image in x,y,z but also in terms of the number of channels.
- Pixels with a value of 1 in the mask are considered for the registration, pixels with a value of 0 are not considered (the wrapper code is written such that any values >0.1 are converted to 1, thus also ImageJ-style masks with 255 as foreground will work).
Use initial transformation file:
- You can use a transformation that you saved in a previous registration as a starting point. For example, it can make sense to first only register the shift and then go on with an affine transformation.
Weights for multi channel images:
- If your images have multiple channels you can configure how much weight you want to give each of them during the registration, e.g. 0.0,1.0,0.0 would only consider channel 2 for the registration.
- Note that the registration will be applied to all channel, regardless.
- As far as I know the weights do not need to add up to 1.0; e.g. 1.0,1.0,0.0 should be fine to only consider the first two channels.

Further notes

Using Windows OS, there sometimes is an error pop-up window during the running, which you can ignore.

Groovy scripting

Run elastix: https://github.com/embl-cba/elastixWrapper/blob/master/scripts/run_elastix.groovy

Running elastix from command line

Elastix can be called via ImageJ on the command line, as in this example:

/Applications/Fiji.app/Contents/MacOS/ImageJ-macosx --ij2 --headless --run "Elastix" "elastixDirectory='/Applications/elastix_macosx64_v4.8', workingDirectory='/Users/tischer/Desktop/elastix-tmp', fixedImageFile='/Users/tischer/Desktop/elastix-output/muscles.tif-transformed.tif', movingImageFile='/Users/tischer/Desktop/elastix-output/muscles.tif-transformed.tif', elastixParameters='Default',useMask='false',useInitialTransformation='false', transformationType='Translation',numIterations='1',numSpatialSamples='100', gaussianSmoothingSigmas='10,10,10',finalResampler='FinalLinearInterpolator', outputModality='Save output as Tiff'"

The transformed output images will be stored in the specified workingDirectory.

To adapt it to your own application, the following parameters should be adapted:

workingDirectory
- please use another directory for storing your temporary data
fixedImageFile
- the reference image file against which you want to register
movingImageFile
- the image file, which is to be transformed
transformationType
- Translation: 3-D shift
- Euler: 3-D shift and rotation
- Similarity: 3-D shift, rotation, and uniform scaling
- Affine: 3-D shift, rotation, and 3-D scaling
- BSpline: local deformations
  - When using BSpline you need to add another key value pair to the parameter list, e.g.: bSplineGridSpacing='50,50,50'
numSpatialSamples
- around 3000 for 3D images can be enough
numIterations
- around 1000 typically is often enough
finalResampler
- FinalLinearInterpolator
  - for intensity images
- FinalNearestNeighborInterpolator
  - for binary or label masks
gaussianSmoothingSigmas
- specify how much you want to smooth in pixels in 3D (my experience is that rather smoothing more is better than less)
- you may specify a semicolon separated list of smoothing factors in order to do the registration at different resolutions, e.g. "10,10,10;1,1,1" will first do the registration with a 10x10x10 downsampled version of the data and then in a second step at full resolution.
outputModality
- Save transformed image as Tiff
  - This will save the transformed output image into the workingDirectory

And just a reminder: Elastix works in physical units and it is thus important that your images are properly calibrated.

Running elastix on EMBL slurm cluster

/g/almf/software/Fiji.app/ImageJ-linux64 --ij2 --headless --run "Elastix" "elastixDirectory='/g/almf/software/elastix_v4.8',workingDirectory='/g/almf/software/elastix-test/elastix-tmp',fixedImageFile='/g/almf/software/elastix-test/muscles.tif',movingImageFile='/g/almf/software/elastix-test/muscles.tif',elastixParameters='Default',useMask='false',useInitialTransformation='false', transformationType='Translation',numIterations='1',numSpatialSamples='100', gaussianSmoothingSigmas='10,10,10',finalResampler='FinalLinearInterpolator', outputModality='Save transformed image as Tiff'"

Above command should run on all cluster nodes; please adapt parameters as explained above.

Running transformix from command line

Transformix can be called via ImageJ on the command line, as in this example:

/Applications/Fiji.app/Contents/MacOS/ImageJ-macosx --ij2 --headless --run "Transformix"  "elastixDirectory='/Applications/elastix_macosx64_v4.8', workingDirectory='/Users/tischer/Desktop/elastix-tmp', inputImageFile='/Users/tischer/Documents/detlev-arendt-clem-registration--data/data/prospr-new/muscles.tif',transformationFile='/Users/tischer/Documents/detlev-arendt-clem-registration/transformations-new/TransformParameters.Similarity-3Channels.0.txt',outputDirectory='/Users/tischer/Desktop/elastix-output',outputModality='Save as Tiff stack'"

The parameters, given as a comma separated key value pair list, are:

elastixDirectory
- installation directory of elastix
workingDirectory
- temporary directory
inputImageFile
transformationFile
outputDirectory
outputModality

Running transformix on EMBL Slurm cluster

Executable command

/g/almf/software/Fiji.app/ImageJ-linux64 --ij2 --headless --run "Transformix" "elastixDirectory='/g/almf/software/elastix_v4.8', workingDirectory='/g/almf/software/elastix-test/tmp', inputImageFile='/g/almf/software/elastix-test/muscles.tif',transformationFile='/g/almf/software/elastix-test/TransformParameters.RotationPreAlign.0.txt',outputFile='/g/almf/software/elastix-test/out',outputModality='Save as BigDataViewer .xml/.h5'"

Above command should run on all cluster nodes.