Conventional and deep learning based approaches
We have the following advanced Processing Tools for Quantitative Susceptibility Mapping (QSM) analysis using input of MR magnitude and phase data acquired using gradient echo (GRE) sequence.
JHU/KKI QSMToolbox: https://godzilla.kennedykrieger.org/QSM/
Two packages are currently available:
(i) JHU/KKI QSM Toolbox V3.0: A Matlab based software package with GUI developed by Dr. Xu Li, Dr. Jiri van Bergen, Dr. Lijun Bao and others for doing QSM. It has different pipelines developed by us and other groups for doing multi-step QSM , including phase unwrapping (Laplacian/Path/Nonlinear fitting), background field removal (PDF/VSHARP/LBV/iRSHARP) and QSM dipole inversion (TKD/iTKD/iLSQR/MEDI/SFCR).
- Supported Data Types: GE, Philips, Siemens, Par/Rec, DICOM, Enhanced DICOM.
- A PDF manual is included.
- Two example datasets (1 mm isotropic, whole brain coverage, 6echoes with TE1/deltaTE/TR=6/6/46ms) acquired at 3T Philips scanner are included for testing.
Note: this toolbox uses a processing routine from FSL, thus FSL must be installed.
(ii) iRSHARP Background Field Removal Tool: Matlab scripts developed by Dr. Lijun Bao for background field removal with an example data set.
(1) Bao L, Li X, Cai C, Chen Z, van Zijl PC. Quantitative Susceptibility Mapping Using Structural Feature Based Collaborative Reconstruction (SFCR) in the Human Brain. IEEE Trans Med Imag. 2016 Sep;35(9):2040-50.
(2) Fang J, Bao L, Li X, van Zijl PCM, Chen Z. Background field removal using a region adaptive kernel for quantitative susceptibility mapping of human brain. J Magn Reson. 2017 Aug;281:130-140
(3) Fang J, Bao L, Li X, van Zijl PCM, Chen Z. Background field removal for susceptibility mapping of human brain with large susceptibility variations. Magn Reson Med. 2019 Mar;81(3):2025-2037.
Deep Learning Based Analysis:
LP-CNN QSM: https://github.com/Sulam-Group/LPCNN
The LPCNN approach was developed by Kuo-Wei Lai, Dr. Xu Li and Dr. Jeremias Sulam, for solving the ill-posed dipole deconvolution problem in Quantitative Susceptibility Mapping (QSM) as detailed in (4). By integrating proximal gradient descent with deep learning, it is the first deep learning based QSM method that can handle an arbitrary number of phase input measurements. The official implementation of LPCNN network can be found on GitHub, which includes QSM training datasets (n=8, with local phase data acquired at 7T and 4-5 orientations COSMOS) and PyTorch implementations of LPCNN that offer the following functions:
- Create default training dataset including patched local phase image and QSM target pairs.
- Conduct single or multiple orientation dipole deconvolution training using LPCNN.
- Reconstruct QSM maps with user’s own data using trained LPCNN model
(4) Lai KW, Aggarwal M, van Zijl P, Li X, Sulam J. Learned Proximal Networks for Quantitative Susceptibility Mapping. Med Image Comput Comput Assist Interv. 2020 Oct;12262:125-135.