twixtools

Purpose

twixtools provide reading and limited writing capability of Siemens MRI raw data files (.dat). In addition, it also includes the compression utility twixzip (see further below for a description).

Requirements

The tool works under Python 3.7 with the following package installed:

• numpy ≥ 1.17.3

Installation

Navigate to the twixtools folder in an open terminal and install twixtools with pip:

pip install .

Installation through python setup.py install is currently not possible.

Demo code

A jupyter notebook that demonstrates the basic functionality of the read_twix, map_twix, and write_twix tools can be found in demo/recon_example.ipynb.

read_twix: "low-level" access to twix data

The raw data file can be parsed using the read_twix function:

import twixtools
multi_twix = twixtools.read_twix(filename)

The function returns a list of individual measurements (of length >=1). The last measurement usually corresponds to the imaging scan, earlier measurements often include calibration data. Each measurement contains a python dict() with the following entries:

• 'mdb': measurement data divided into blocks (return type: list)
• 'hdr': dict of parsed protocol header strings (each dict element contains another dict with protocol information)
• 'hdr_str': dict of original protocol header strings (divided into different protocol types)
  • note that this is the protocol information that is used for twix file writing (by write_twix), so make sure to make necessary adjustments here and not in ['hdr']
• ('raidfile_hdr': required for twix file writing, otherwise of little importance)

Each invididual 'mdb' in the list of mdbs consists of a data and a header (line counters and such) part, which can be accessed as follows:

mdb = multi_twix[-1]['mdb'][0] # first mdb element of last measurement
mdb.data # data of first mdb (may or may not be imaging data)
mdb.mdh # full miniheader information stored as a numpy dtype object

Different data types can be distinguished by returning a list of active flags, or by directly checking whether the data is assumed to be from an imaging scan (and not from a calibration scan such as a phase correction scan or a noise measurement):

mdb.get_active_flags() # get all active MDH flags
mdb.is_image_scan() # check if this an image scan (True or False)

Line Counters can be accessed as follows:

mdb.cLin   # returns line number
mdb.cPar   # returns partition number
mdb.c<tab> # with line completion enabled, this should give you a list of all counters

The full minidata header (mdh) information is stored in a mdb.mdh special numpy dtype object. You can print a list of its entry names by printing mdb.mdh.dtype.names.

Example code

import numpy as np
import twixtools

# read all image data from file
def read_image_data(filename):
    out = list()
    for mdb in twixtools.read_twix(filename)[-1]['mdb']:
        if mdb.is_image_scan():
            out.append(mdb.data)
    return np.asarray(out)  # 3D numpy array [acquisition_counter, n_channel, n_column]


# read image data from list of mdbs and sort into 3d k-space (+ coil dim.)
def import_kspace(mdb_list)
    image_mdbs = []
    for mdb in mdb_list:
        if mdb.is_image_scan():
            image_mdbs.append(mdb)

    n_line = 1 + max([mdb.cLin for mdb in image_mdbs])
    n_part = 1 + max([mdb.cPar for mdb in image_mdbs])
    n_channel, n_column = image_mdbs[0].data.shape

    out = np.zeros([n_part, n_line, n_channel, n_column], dtype=np.complex64)
    for mdb in image_mdbs:
        # '+=' takes care of averaging, but careful in case of other counters (e.g. echoes)
        out[mdb.cPar, mdb.cLin] += mdb.data

    return out  # 4D numpy array [n_part, n_line, n_channel, n_column]

map_twix: "high level" access to twix data

map_twix is a high-level function that takes the data obtained from read_twix (in the form of Mdb objects), and maps it to multi-dimensional "k-space" arrays. These twix_array objects are generated for different data types (image/noise adjust/phase-correction/... scan) and can be accessed with numpy.ndarray array-slicing syntax.

Optional flags control additional feature and also have an impact on size and shape of the multidimensional arrays. The following flags are currently available (stored in the flags dict within each twix_array object):

• average: dict of bools that determines which dimensions should be averaged.
• squeeze_ave_dims: bool that determines whether averaged dimensions should be removed/squeezed from the array's shape.
• remove_os: oversampling removal. Reduces the number of columns by a factor of two.
• regrid: bool that controls ramp-sampling regridding (if applicable)
• skip_empty_lead: skips to first line & partition that is found in mdb list (e.g. if first line counter is 10, the output array starts at line counter 10).
• zf_missing_lines: zero-fill k-space to include lines and partitions that are higher than the maximum counter found in the mdb list, but are still within the k-space matrix according to the twix header.

For example code, please look at the demo/recon_example.ipynb jupyter file.

twixzip Compression Utility

Purpose

twixzip is a Python based command line tool for Siemens MRI raw data compression. Following compression methods can be selected via the command line:

• Oversampling removal
• Lossy floating point compression using the zfp library
• Single coil compression (scc) based on singular value decomposition (SVD)
• Geometric coil compression (gcc) based on SVD
• Optionally FID navigators can be removed

Before applying the selected compression method(s), lossless compression (gzip) is applied to the header and meta data information which is then added to a hdf5 file. All additional meta information necessary for decompression (e.g. coil compression matrices) are also stored in the hdf5 file.

Additional Requirements

• pyzfp ≥ 0.3.1
• pytables ≥ 3.6.1

The pyzfp and pytables libraries can be installed via pip:

pip install pyzfp
pip install tables

Usage

Executing the command twixzip.py in an open terminal gives an overview of all possible arguments. Optional arguments are:

-h:  help  
-d:  decompress data

Input and output directories & filenames are required arguments that can be selected via:

-i infile:  input file  
-o outfile: output file

In the compression mode the input file should be an MRI raw data file, in the decompression mode (-d) it should be the hdf5 file containing the compressed data. The output file is then an hdf5 file (compression mode) or an MRI raw data file (decompression mode).

Compression methods can be selected via:

--remove_fidnav:            removes FID navigators  
--remove_os:                removes oversampling
--scc -n NCC:               single coil compression (SCC) - keep NCC virtual coils
--scc -t CC_TOL:            SCC - number of coils is calculated with a tolerance for the singular values
--scc_bart -n NCC:          SCC using BART  
--gcc -n NCC:               geometric coil compression (GCC) - keep NCC virtual coils
--gcc -t CC_TOL:            GCC - number of coils is calculated with a tolerance for the singular values
--gcc_bart -n NCC:          GCC using the Berkeley Advanved Reconstruction Toolbox (BART) [1]         
--zfp --zfp_tol ZFP_TOL:    floating point compression with ZFP_TOL tolerance
--zfp --zfp_prec ZFP_PREC:  floating point compression with ZFP_PREC precision (not recommended)

The optional argument --testmode can be used to automatically decompress the data after compression. The created decompressed MRI raw data filename contains the selected compression method. The option --profile can be used to profile the compression code.

[1] BART Toolbox for Computational Magnetic Resonance Imaging, DOI: 10.5281/zenodo.592960

Acknowledgements

The protocol header parsing code originates from William Clarke's excellent pymapvbvd project (https://github.com/wexeee/pymapvbvd).