This repo contains the official PyTorch implementation for the paper WaveSep: A Flexible Wavelet-based Approach for Source Separation in Susceptibility Imaging, at MLCN 2023
by Zhenghan Fang, Hyeong-Geol Shin, Peter van Zijl, Xu Li, and Jeremias Sulam
Create and activate a new conda environment
conda create -n wavesep python==3.10
conda activate wavesep
Install necessary python packages
pip install -r requirements.txt
python wavesep/qsm_sep.py --data <yml of input data>
The yml file contains the input data for QSM source separation. See data/yml/template_qsm.yml for more details. See data/yml/example_qsm.yml for an example.
🔄 Update (2025/04/23): Support for different Dr values for para- and dia-magnetic maps (Dr_pos ≠ Dr_neg)
In this case, the second term in fQSM in Eq. (3) in the paper is changed from
1/2 * || R2' / Dr - (x_pos - x_neg) ||_2^2
to
1/2 * || R2' / Dr_pos - (x_pos - x_neg * Dr_neg / Dr_pos) ||_2^2.
where the units are:
R2': HzDr,Dr_pos,Dr_neg: Hz/ppm
python wavesep/sti_sep.py --data <yml of input data>
The yml file contains the input data for STI source separation. See data/yml/template_sti.yml for more details. See data/yml/example_sti.yml for an example.
If you have any questions, please contact me at
Zhenghan Fang
Email: zfang23@jhu.edu