Susceptibility Weighted Imaging (CLEAR-SWI)

Published at the ISMRM as CLEAR-SWI. It provides magnetic resonance images with improved vein and iron contrast by weighting a combined magnitude image with a preprocessed phase image. This package has the additional capability of multi-echo SWI, intensity correction, contrast enhancement and improved phase processing. The reason for the development of this package was to solve artefacts at ultra-high field strength (7T), however, it also drastically improves the SWI quality at lower field strength.

Getting Started

Prerequisites

A Julia installation ≥ 1.3 (Official Julia Webpage)

Single-echo or multi-echo Magnitude and Phase images in NIfTI fileformat (4D images with echoes in the 4th dimension)

Installing

Run the following commands in Julia (either interactively in the REPL or as a script)

using Pkg
Pkg.add(PackageSpec(url="https://github.com/korbinian90/CLEARSWI.jl"))

Usage

This is a simple multi-echo case without changing default behavior

using CLEARSWI

TEs = [4,8,12] # change this to the Echo Time of your sequence. For multi-echoes, set a list of TE values, else set a list with a single TE value.
nifti_folder = CLEARSWI.dir("test","testData","small") # replace with path to your folder e.g. nifti_folder="/data/clearswi"
magfile = joinpath(nifti_folder, "Mag.nii") # Path to the magnitude image in nifti format, must be .nii or .hdr
phasefile = joinpath(nifti_folder, "Phase.nii") # Path to the phase image

mag = readmag(magfile);
phase = readphase(phasefile);
data = Data(mag, phase, mag.header, TEs);

swi = calculateSWI(data);
mip = createMIP(swi);

savenii(swi, "<outputpath>/swi.nii"; header=mag.header) # change <outputpath> with the path where you want to save the reconstructed SWI
savenii(mip, "<outputpath>/mip.nii"; header=mag.header)

Available Options

To apply custom options use the following keyword syntax (example to apply 3D high-pass filtering for the phase with the given kernel size and deactivate softplus magnitude scaling):

options = Options(phase_hp_σ=[10,10,5], mag_softplus=false)
swi = calculateSWI(data, options);

All the possible options with the default values are

mag_combine=:SNR
mag_sens=nothing
mag_softplus=true
phase_unwrap=:laplacian
phase_hp_σ=[4,4,0]
phase_scaling_type=:tanh
phase_scaling_strength=4
writesteps=nothing

mag_combine selects the echo combination for the magnitude. Options are
- :SNR
- :average
- :last to select the last echo
- (:CNR => (:gm, :wm)) to optimize the contrast between two selected tissues with the possible tissues classes to select in src\tissue.jl, currently only working for 7T
- (:echo => 3) to select the 3rd echo
- (:closest => 15.3) to select the echo that is closest to 15.3 ms
- (:SE => 15.3) to simulate the contrast that would be achieved using a corresponding single-echo scan with 15.3 ms echo time.
If mag_sens is set to an array, it is used instead of CLEAR-SWI sensitivity estimation. This can also be set to mag_sens=[1] to use the constant sensitivity of 1 and effectively avoid sensitivity correction.
To deactivate scaling of the combined magnitude with the softplus function, use mag_softplus=false.
phase_unwrap is either :laplacian, :romeo, or :laplacianslice (slicewise laplacian unwrapping)
The phase_hp_σ is used for high-pass filtering and is given in voxel for the [x,y,z]-dimension.
phase_scaling_type is the scaling function to create the phase mask and can be :tanh for sigmoidal filtering, or :positive, :negative, and :triangular for traditional SWI application. If the scaling has the wrong sign, the phase input can be simply inverted by a minus sign: phase = -readphase(phasefile);
phase_scaling_strength adjusts the strength of the created phase mask. A higher phase_scaling_strength is a stronger phase appearance. With a traditional SWI phase_scaling_type it corresponds to the power or number of phase mask multiplications.
Set writesteps to the path, where intermediate steps should be saved, e.g. writesteps="/tmp/clearswi_steps". If set to nothing, intermediate steps won't be saved.

License

This project is licensed under the MIT License - see the LICENSE for details

Suggest Category