MyoQMRI

This project is an open-source effort to put together tools for quantitative MRI of the muscles. Current features include:

• multi echo spin echo sequence, written in PyPulseq
• fast water T2 mapping from multi echo spin echo images
• fat water separation from 2 echo gradient echo data

Setup

Virtual environment

We recomment creating a virtual environment.
Do this with

python -m venv /path/to/new/virtual/environment

If you are on Windows, this will look something like this:

python -m venv C:\path\to\new\virtual\environment

Depending on your distribution, you may need to use python3 instead of python as command.
If there are any problems concerning the creation of the virtual environment or its use, refer to: https://docs.python.org/3/library/venv.html

To activate the virtual environment, run:

source /path/to/new/virtual/environment/bin/activate

On Windows:

C:\path\to\new\virtual\environment\Scripts\activate.bat

To deactivate the virtual environment, run:

deactivate

Installation

Install via PyPI

MyoQMRI is on PyPI, you can easily install it via

pip install myoqmri

if you have cuda, also install the optional dependencies

pip install myoqmri[cuda]

Download and install via GitHub

Alternatively, you can install MyoQMRI directly from GitHub. The git repository contains the compiled pulseq multi echo spin echo sequence, all analysis tools and example data.
Download it via

git clone https://github.com/bammri/MyoQMRI

or manually via the webpage.
Install it with

pip install -e /path/to/folder/MyoQMRI

If you want to use cuda for GPU support, install the optional dependencies via

pip install -e /path/to/folder/MyoQMRI[cuda]

On Windows:

pip install -e C:\path\to\folder\MyoQMRI

Multi-Echo Spin Echo Sequence

For instructions on how to install the pulseq interpreter and bring the sequence to the scanner, see https://github.com/pulseq/tutorials.

The sequence was developed for imaging of the thighs. In principle, however, it is possible to easily adapt it for other body regions.
For this,

• Copy the write_pulseq_mese_leg.py file
• Rename the file and the seq_filename variable within
• Edit the fov_x and fov_y variables (and the Nx, Ny accordingly to keep the resolution the same)
• Execute the code in the terminal with: python write_pulseq_your_seqname.py

The .seq and .json file are written into the current working directory.

Reconstruction

Data from the mese_leg.seq is reconstructed offline.

• Transfer the raw data from the scanner
• rename the .dat file to the patient token of your choice
• Copy mese_leg.json and mese_leg.seq into the same folder
• Run raw2nii_pulseq_mese_leg -p path/to/folder/containing/the/data

The code outputs a .nii.gz and an according .json file into an /mr-anat folder.

Prepare DICOM data

Dicom files need to be converted to the ormir-mids format.
For this, we use the ormir-mids converter (more information on https://github.com/ormir-mids/ormir-mids). If you are working with the enhanced dicom format and are converting Dixon data, see additional step below.

The conversion is done with:

dcm2omids -r -a PATIENT_NAME /path/to/input/folder /path/to/output/folder 

Note: Output folder must exist.

Additional step for enhanced dicom

If you are working with the old dicom format, you can ignore this step.

As of now, for enhanced Dixon dicoms, we need a series_config.json file for the converter to work properly. For this, each magnitude and phase 4D image needs to be in its own folder (see example data folder structure). Copy the series_config.json file from the example data and edit it to contain the according series number of your data. Afterwards, the conversion can be executed like above.

T2 Mapping From Multi-Echo Spin Echo

The main function is used to obtain water T2 and fat fraction from multi-echo spin echo images as described in Marty et al.

This implementation uses the GPU for the generation of a dictionary of signals through Extended Phase Graph simulation.

Ideally, you should work with the ormir-mids format. See Section Prepare DICOM data above for instructions. However, as of now, it is also possible to use the code with dicom images.

In case of dicom data, the dataset must be a directory containing 2D DICOM images from a multiecho spin echo acquisition, ordered as (slice1echo1 - slice2echo1 - ... - slice1echo2 - slice2echo2 - ...).

Usage

usage: waterT2 [-h] [--fit-type T] [--bids] [--path-is-nifti] [--fat-t2 T2] [--noise-level N] [--nthreads T] [--plot-level L] [--t2-limits min max] [--b1-limits min max] [--use-gpu] [--ff-map dir] [--register-ff] [--etl-limit N] [--out-suffix ext] [--slice-range start end] [--refocusing-width factor] [--exc-profile path] [--ref-profile path] path

Fit a multiecho dataset

positional arguments:
  path                  path to the dataset

optional arguments:
  --bids                use muscle-BIDS format for input/output
  --path-is-nifti       set if path is pointing directly to a nifti file
  -h, --help            show this help message and exit
  --fit-type T, -y T    type of fitting: T=0: EPG, T=1: Single exponential, 
                        T=2: Double exponential (default: 0)
  --fat-t2 T2, -f T2    fat T2 (default: 151)
  --noise-level N, -n N
                        noise level for thresholding (default: 300)
  --nthreads T, -t T    number of threads to be used for fitting (default: 12)
  --plot-level L, -p L  do a live plot of the fitting (L=0: no plot, L=1: show 
                        the images, L=2: show images and signals)
  --t2-limits min max   set the limits for t2 calculation (default: 20-80)
  --b1-limits min max   set the limits for b1 calculation (default: 0.5-1.2)
  --use-gpu, -g         use GPU for fitting
  --ff-map dir, -m dir  load a fat fraction map
  --register-ff, -r     register the fat fraction dataset
  --etl-limit N, -e N   reduce the echo train length
  --out-suffix ext, -s ext
                        add a suffix to the output map directories
  --slice-range start end, -l start end
                        Restrict the fitting to a subset of slices
  --refocusing-width factor, -w factor
                        Slice width of the refocusing pulse with respect to the 
                        excitation (default 1.2) (Siemens standard)
  --exc-profile path    Path to the excitation slice profile file
  --ref-profile path    Path to the refocusing slice profile file

Note for Windows users: Currently, the (default) multithreading does not work on Windows.
Simply run the code with the option -t 1 and it should work fine.

Example

waterT2 --bids --path-is-nifti -n100 PATNAME_mese.nii.gz

Slice profile

An accurate slice profile is crucial to obtain accurate results. By default, a hanning-windowed sinc pulse is used, and the refocusing pulse has a 1.2x the slice width of the excitation pulses. This reflects the parameters of the standard Siemens Spin Echo sequence.

External slice profile files can be provided. They are text files with angle values (in degrees) across half the profile (i.e. starting with ~90 or ~180 and decreasing), with one value per line.

Either none or both slice profiles must be given, and both slice arrays must contain the same number of samples.

Example:

90
89.8014
89.2106
88.2423
86.9198
85.2734
83.3385
81.1537
78.7585
76.1921
...

Fat Water Separation

The 2 echo fat water separation uses an algorithm by the BMRR group in Munich, which utilizes hirarchical multi-resolution graph-cuts (for more information see https://github.com/BMRRgroup/2echo-WaterFat-hmrGC).

Download and Install Package

Download the code via

git clone https://github.com/BMRRgroup/2echo-WaterFat-hmrGC

or manually via the webpage.
Install it with

pip install -e /path/to/folder/2echo-WaterFat-hmrGC

Usage

usage: fatwater [-h] [-p PATH] [-e ECHONUMS ECHONUMS] [-c FATSHIFT [FATSHIFT ...]] [-a RELAMPS [RELAMPS ...]] [-ph]

Compute fat/water images from dual echo data

options:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  path to the folder where the data is located, 
                        default: current working directory
  -e ECHONUMS ECHONUMS, --echonums ECHONUMS ECHONUMS
                        if data contains more than two echoes, provide which
                        echoes to use, default: first two echoes
  -c FATSHIFT [FATSHIFT ...], --fatshift FATSHIFT [FATSHIFT ...]
                        Chemical shift of fat peak(s) in ppm, default: 3.5
  -a RELAMPS [RELAMPS ...], --relamps RELAMPS [RELAMPS ...]
                        Relative amplitude of fat peaks, default: 1
  -ph                   if given, the code additionally outputs the phase of
                        the computed fat and water images

Example

fatwater -p path/to/folder

Git repositories of the constituents

Pulseq/PyPulseq

https://pulseq.github.io/index.html

Ormir-mids

https://github.com/ormir-mids

MyoQMRI

https://github.com/bammri/MyoQMRI

Fat water separation

https://github.com/BMRRgroup/2echo-WaterFat-hmrGC

Citing

PyPulseq

Ravi KS, Geethanath S, Vaughan JT. PyPulseq: A python package for MRI pulse sequence design. Journal of Open Source Software, 2019;4:1725.

Pulseq mese sequence

Schäper J, Santini F, Weidensteiner C. A standardized open-source MESE sequence implemented in PyPulseq for reproducible water T2 quantification in skeletal muscle, Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) 2025

Water T2 quantification

Santini F, Deligianni X, Paoletti M, et al., Fast open-source toolkit for water T2 mapping in the presence of fat from multi-echo spin-echo acquisitions for muscle MRI, Frontiers in Neurology 2021, https://doi.org/10.3389/fneur.2021.630387.

Fat water separation

Stelter J et al. Hierarchical multi-resolution graph-cuts for water-fat-silicone separation in breast MRI, IEEE Transactions on Medical Imaging, 2022, DOI: 10.1109/TMI.2022.3180302, https://ieeexplore.ieee.org/document/9788478

Eggers H, Brendel B, Duijndam A, Herigault G. Dual-echo Dixon imaging with flexible choice of echo times, Magnetic Resonance in Medicine, 2010, DOI: 10.1002/mrm.22578, https://doi.org/10.1002/mrm.22578

Authors and Acknowledgment

Main contributors

• Jessica Schäper
• Francesco Santini
• Claudia Weidensteiner

EPG code was made possible by Matthias Weigel.

This project was supported by

• FSRMM
• SNF (grant number 320030_172876)