This repo contains Matlab codes for generating input data and python codes for running INR-QSM.
Here is the paper for INR-QSM:
A subject-specific unsupervised deep learning method for quantitative susceptibility mapping using implicit neural representation
https://www.sciencedirect.com/science/article/abs/pii/S1361841524000987
A subject-specific unsupervised deep learning approach for QSM (quantitative susceptibility mapping) reconstruction using implicit neural representation (INR).
- A new signal representation scheme for QSM recon which maps the susceptbility value as an implicit function of coordinates
- The non-local phase effect is considered in patch-based QSM deep learning method
- The proposed method is subject-specific and unsupervised, indicating that it is free of generalization problems and the requirement of a large dataset, which are two issues in supervised deep learning methods
- Several acceleration strategies are adopted to accelerate training process
- STISuite 3.0
- PyTorch 1.11.0
- Python 3.8
Data preparation
- Generate test data
data.mat
containingphi
,msk
,WG
based on files indata_prep
folder - Adjust the
config.py
by inputting correctvoxel size
,B0_dir
,patch size
, and other parameters
Training and prediction
- Run
main.py
for generating INR-QSM output
Feel free to contact zhangming430424@gmail.com
or mingzhang.bme@sjtu.edu.cn
for questions/discussions/suggestions.