Maximum Spherical Mean Value for Shadow Reduction (mSMV)

Contents

Whole brain atlas construction
Ex vivo susceptibility
Whole brain susceptibility

Summary

Here, an algorithm based on the maximum corollary of Green’s theorem is proposed to remove shadows in quantitative susceptibility mapping while preserving the edge of the brain. This method is referred to as maximum Spherical Mean Value, or mSMV.

Residual background field is a major source of shadow artifacts in QSM. The algorithm filters large field magnitude values near the border, where the extreme values of the harmonic background field are located. Further details can be found in Publications.

Installation

Clone the repository with: git clone https://github.com/agr78/mSMV.git

Function arguments

For ease of use with the MEDI Toolbox (included in mSMV/code/dependencies/MEDI_functions/), the function mSMV accepts the following arguments:
in_file Input file containing the field map after background field removal (by PDF, VSHARP, LBV, etc.)
out_file Output file containing mSMV filtered field map
radius Prefilter radius (default 5mm)
maxk Maximum number of iterations with minimum kernel radius (default 5)
vr Frangi filter vessel radius, see MATLAB's fibermetric function for further details
pf Optional disabling of the prefilter step (typically used with SHARP, RESHARP, VSHARP, etc.)

Prerequisites

All necessary toolboxes are included in mSMV/code/dependencies/. If these toolboxes are already installed, mSMV/code/mSMV_functions/ can be added to the MATLAB path. To update an existing MEDI installation, see prerelease.

Examples

If an installation of MEDI already exists, pull the msmv branch and modify the QSM reconstruction as follows:

QSM = MEDI_L1('filename', 'RDF.mat', 'lambda', 1000, 'merit', 'msmv');

If starting from a simple clone of this repository, run:

% Import complex field data
[iField,voxel_size,matrix_size,CF,delta_TE,TE,B0_dir,files]=Read_DICOM('DICOM');

% Generate the magnitude image 
iMag = squeeze(sqrt(sum(abs(iField).^2,4)));

% Compute brain mask
Mask = BET(iMag,matrix_size,voxel_size);

% Estimate noise and normalize field
if (~exist('noise_level','var'))
    noise_level = calfieldnoise(iField, Mask);
end
iField = iField/noise_level;

% Estimate the frequency offset in each of the voxel using complex fitting 
[iFreq_raw,N_std] = Fit_ppm_complex(iField);

% Fit R2* map
R2s = arlo(TEk,abs(iField));

% CSF zero-reference
Mask_CSF = extract_whole_CSF(R2s,Mask,voxel_size);

% Unwrap phase using ROMEO
iFreq = romeo(iFreq_raw, iMag, Mask);

% Background field removal
[RDF,shim] = PDF(iFreq,N_std,Mask,matrix_size,voxel_size,B0_dir,0,100);

% Save local field generated by PDF
save RDF.mat RDF iFreq iFreq_raw iMag N_std Mask matrix_size...
     voxel_size delta_TE CF B0_dir Mask_CSF R2s;

% Reconstruct QSM
QSM = MEDI_L1('filename', 'RDF.mat', 'lambda', 1000, 'merit', 'msmv', 5);

Notes

• The vessel mask requires an $R_2^*$ map for generation, if this variable is missing in in_file, this step will be skipped.
• The default phase unwrapping algorithm is ROMEO, called from a MATLAB mex file compiled on Windows 10. On different operating systems, unwrapPhase.m will be used.
• The mrm branch contains code needed to reproduce figures in the paper.

Publications

If this code is used, please cite the following:

Magnetic Resonance in Medicine Article: A. G. Roberts et al., "Maximum Spherical Mean Value (mSMV) Filtering for Whole Brain Quantitative Susceptibility Mapping," Magnetic Resonance in Medicine, 2024, DOI: 10.1002/mrm.29963

Preprint: A. G. Roberts et al., "Maximum Spherical Mean Value (mSMV) Filtering for Whole Brain Quantitative Susceptibility Mapping," arXiv pre-print server, 2023-04-22 2023, arxiv:2304.11476

BibTex

@article{Roberts_mSMV_2024,
   author = {Roberts, Alexandra G. and Romano, Dominick J. and Şişman, Mert and Dimov, Alexey V. and Nguyen, Thanh D. and Kovanlikaya, Ilhami and Gauthier, Susan A. and Wang, Yi and Spincemaille, Pascal},
   title = {Maximum spherical mean value filtering for whole‐brain QSM},
   journal = {Magnetic Resonance in Medicine},
   volume = {91},
   number = {4},
   pages = {1586-1597},
   ISSN = {0740-3194},
   DOI = {10.1002/mrm.29963},
   url = {https://dx.doi.org/10.1002/mrm.29963},
   year = {2024},
   type = {Journal Article}
}

Contact

Please direct questions to Alexandra Roberts at agr78@cornell.edu.