Digital Object Simulations

Duration: 2023
Technologies: C, Shell Scripting, Python
Collaborators: Martin Schilling, Martin Heide, Martin Juschitz, Martin Uecker

Simulated transient-state signal of an analytical brain geometry during an inversion-recovery FLASH acquisition.

The digital objects project developed a simple and accessible tool to generate digital objects and simulations for medical imaging research. It allows the user to systematically increase the complexity of datasets to enhance the development of imaging and reconstruction techniques and provide a reproducible tool to compare different methods. The tool was developed at the Graz University of Technology in collaboration with the University Medical Center in Göttingen.

The tool has been implemented in the Berkeley Advanced Reconstruction Toolbox (BART) and extended a previous interface for digital objects to:
  1. Arbitrary geometries of digital objects that can be passed as Bézier curves to the command line interface of BART.
  2. Include a Python 3 based tool for extracting Bézier curves from an SVG file to the BART compatible multi-cfl format.
  3. A realistic brain geometry segmented into four different regions.
  4. Realistic 64 channel 3D coil sensitivity profiles measured and extracted from a head coil at the university medical center in Göttingen.
  5. A comprehensive framework to simulate various MRI sequences with an adaptive step-size Runge-Kutta solver, state-transition matrices, or rotational matrices. The framework supports physical models of differing complexity, e.g. the Bloch equations.

Schematic concept of the developed component-based simulation of realistic signal evolutions and complex geometries. For simplicity the simulation in image domain is illustrated, but a comparison between image and native frequency is provided.

By selecting Bézier curves to define the contours of the digital objects it is ensured that their Fourier representation can be calculated analytically. This enables realistic simulations of samples in frequency domain, which is especially important in magnetic resonance research.
The created tool also provides an analytical frequency representation of realistic 2D and 3D coil sensitivity profiles. They were extracted from acquisitions at the university medical center in Göttingen and incorporate their lowest 5x5 frequency coefficients. They are stored and used in convolutions with the simulated frequency representations of the object geometries that entirely avoids the image domain for realistic spatial frequency simulations.

The extensions allow everyone to simulate realistic, synthetic datasets of simple up to complicated geometries with realistic temporal signal evolutions and coil sensitivity profiles even for non-Cartesian sampling trajectories.
The developed extensions are accessible from the command line with a few lines of code. Please check out the interactive tutorial for further details. The individual tools are designed to be modular. Therefore, exchanging parts of the simulation pipeline with other tools should be convenient.

Resources

Resources Location
Software Tools DOI, Berkeley Advanced Reconstruction Toolbox
Tutorial Github

References

  • Digital Reference Objects with BART
    Nick Scholand*, Martin Schilling*, Martin Heide, Martin Uecker
    In Proc. Intl. Soc. Mag. Reson. Med. 31 (2023); 3118
  • Quantitative MRI by nonlinear inversion of the Bloch equations
    Nick Scholand, Xiaoqing Wang, Volkert Roeloffs, Sebastian Rosenzweig, Martin Uecker
    Magn Reson Med. 2023; 90: 520-538.
    doi: 10.1002/mrm.29664