This code was downloaded from https://www.neuroimaging.at/pages/qsm.php and adapted to run on newer versions of python, nibabel by yanarof

QSM reconstruction using Total Generalized Variation (TGV-QSM) Kristian Bredies and Christian Langkammer Version from June 2016 www.neuroimaging.at

NEW

• It is a python package now which simplifies installation
• Orientation fixed: image data does NOT have to be transversal any longer (thx Daniel!)

Installation

Linux requirements:

sudo apt install python3.12-dev build-essential

Python requirements:

pip install nibabel numpy Cython

Compile the Cython

cd src ; python setup.py build_ext --inplace ; cd ..

Command line options

usage:

tgv_qsm [-h] -p PHASE -m MASK [-o OUTPUT_SUFFIX]
               [--alpha ALPHA ALPHA | --factors FACTORS [FACTORS ...]]
               [-e EROSIONS] [-i ITERATIONS [ITERATIONS ...]]
               [-f FIELDSTRENGTH] [-t ECHOTIME] [-s] [--ignore-orientation]
               [--save-laplacian] [--output-physical] [--no-resampling]
               [--vis] [-v]

remarks for options: -t TE in seconds -f fieldstrength in Tesla -s autoscaling for SIEMENS phase data

test data:

python src/qsm_tgv_main.py -p test_data/epi3d_test_phase.nii.gz -m test_data/epi3d_test_mask.nii.gz -f 2.89 -t 0.027 -o epi3d_test_QSM

bet brain masking

• for high res data (~0.5 mm iso) this does quite a good job:
• bet magni.nii.gz mask -n -m -R -f 0.1 -g 0.0
• (in case something is lost, vary parameter g minimally)

This code was run and tested with the following

Ubuntu 24.04 LTS
python-3.12
nibabel==5.2.1
numpy==1.26.4
Cython==3.0.10