chi-separation toolbox (χ-separation, x-separation)

• ⭐ New version of the toolbox is now available for test! Don't hesitate to contact us for bugs and feedbacks: snu.list.software@gmail.com

• ⭐ A new error that a saved nii file has revsered left-right images in some dicom input files has been reported and has not be fixed yet. Users are advised to double check this issue in their data.

• ⭐ This toolbox is developed for in-vivo human 3T datasets. Other field stengths, high resolution ex-vivo or nonhuman datasets (< 0.6 mm) are not fully tested and may require additional processing.

• ⭐ For 7T processing (> 0.6 mm), we have a new method coming up soon [link]. Please contact us if you want to use this method.

• ⭐ E-mail us if you need data processing other than in-vivo human 3T datasets. We may be able to help you with scan protocol and processing.

• ⭐ If you have both GRE and SE data, you have data processing options of conventional optimization (MEDI, iLSQR) and neural network (chi-sepnet-R2', better quality).

• ⭐ If you only have GRE data, neural network (chi-sepnet-R2*) will deliver high quality χ-separation maps.

• ⭐ Neural networks can process resolution > 0.6 mm. This makes high resolution ex-vivo or rodent data processing difficult if you only have GRE data.

• The χ-separation toolbox includes the following features:

  • DICOM/NIFTI data compatibility
  • QSMnet: Quantitative susceptibility mapping (QSM) reconstruction algorithm based on deep neural network (QSMnet; J. Yoon et al., Neuroimage, 2018)
  • χ-separation using R₂' (or R₂* ): Magnetic susceptibility source separation algorithms based on convex optimization (χ-separation; H. Shin et al., Neuroimage, 2021) that share similar contrasts and optimization parameters with either MEDI+0 (Liu et al., MRM, 2018) or iLSQR (Li et al., Neuroimage, 2015) algorithms. The toolbox also provides the option to use pseudo R₂ map if R₂ measurement is not availabe (using R₂' is reconmmanded for accurate estimation).
  • χ-sepnet using R₂' (or R₂* ): A U-Net-based neural network that reconstructs COSMOS-quality χ-separation using R₂' and phase. In case R₂ is not measured, another neural network is trained to estimate χ-separation maps from R₂* and phase.
  • The toolbox also supports phase preprocessing (e.g. phase unwrapping and background removal) powered by MEDI, STI Suite, SEGUE, and mritools toolboxs (see the Chisep_script.m file for details).

• Last update: June-11-2024 (Sooyeon Ji, Hyeong-Geol Shin, Jun-Hyeok Lee, Minjun Kim, Kyeongseon Min)

Requirements

• Common requirement
  • MATLAB (tested in R2019a-R2021a. Toolbox will fail any version before R2019a.)
• Additional requirements
  • For QSMnet and χ-sepnet, Deep Learning MATLAB Toolbox Converter for ONNX Model Format (https://www.mathworks.com/matlabcentral/fileexchange/67296-deep-learning-toolbox-converter-for-onnx-model-format)
  • For DICOM/NIFTI read and phase processing, see https://www.dropbox.com/sh/3zafav50bfnruuu/AABVVYpdsznsRXKy8YKK4ybla?dl=0

Recommendation for acquisition

• Multiecho GRE data for in-vivo
  • TR = 33 ms; TE1 = 5 ms; Echo spacing = 6 ms; Number of echoes = 5; flip angle = 15° Matrix size (AP LR HF) = 256 x 176 x 144 Resolution = isotropic 1 mm Parallel imaging factor = 2; elliptical k-space shutter Acquisition time = 6 mins.
  • Follow QSM consensus paper (https://doi.org/10.1002/mrm.30006)
• Multiecho SE data for in-vivo
  • TBA
• Multiecho GRE data for high resolution ex-vivo
  • Highly recommend to acquire multi-orientation data (relative to B0).

Recommendation for data analysis

• χ-separation template and regions of interest are available here [link]
• K Min et al. A human brain atlas of χ-separation for normative iron and myelin distributions. NMR Biomed, 2024 online [link]

Reference

• H. Shin, J. Lee, Y. H. Yun, S. H. Yoo, J. Jang, S.-H. Oh, Y. Nam, S. Jung, S. Kim, F. Masaki, W. Kim, H. J. Choi, J. Lee. χ-separation: Magnetic susceptibility source separation toward iron and myelin mapping in the brain. Neuroimage, 2021 Oct; 240:118371.
• For more papers related to χ-separation, please refer to [link]

Contacts

• snu.list.software@gmail.com
• sin4109@gmail.com (Hyeong-Geol Shin, PhD)