KomaMRI.j: Device-Agnostic, Highly Efficient, Pulseq-Compatible MRI Simulations

Open-source Software Description: KomaMRI.jl¹ is an MRI simulation package written in the Julia programming language.² Julia offers an unparalleled ability to write general code by using multi-argument dispatch, which facilitates composability between different packages in the ecosystem, while at the same time obtaining high performance. Julia has one of the best performances per lines of code, which has been utilized by the rapidly growing MRI ecosystem with packages for image reconstruction and analysis (MagneticResonanceImaging organization) and also health, medicine, and biomedical research (JuliaHealth organization).

KomaMRI.jl is part of JuliaHealth and takes advantage of all of benefits to perform device-agnostic, highly efficient MRI simulations, covering relaxation effects, off-resonance, and motion. This is done by simulating the Bloch equations using the operator splitting method, which separates relaxation effects and field-induced magnetization rotation (excitation and precession). Currently, KomaMRI.jl has more than 850 downloads, 112 GitHub stars, and 14 collaborators from all around the world, and its speed and accuracy are regularly tested by the use of Continuous Integration (CI) on GitHub for CPU-related tasks, and Buildkite through JuliaGPU for GPU-related tasks.

To bring state-of-the-art simulation methods to as many people as possible, KomaMRI.jl has been made open-source, well-documented with reproducible examples (running on Binder), and provides an easy-to-use Graphical User Interface (GUI), so no Julia programming knowledge is required. The main inputs of our simulator are files that can be generated externally, such as vendor-agnostic pulse sequences with Pulseq³ as the Sequence definition and an HDF5-based Phantom format, which contains all the information about the magnetic properties and motion, facilitating reproducibility by sharing sequences and phantoms. As output, the ISMRMRD raw data format is used, which can be reconstructed in any ISMRMRD-compatible software, like BART⁴ or MRIReco.jl.⁵ Moreover, we recently made the simulations device-agnostic, as not everyone has an NVIDIA GPU, making Koma compatible with multiple devices, such as CPU multi-threading, NVIDIA GPUs (CUDA.jl), AMD GPUs (AMDGPU.jl), Apple Silicon GPUs (Metal.jl), and integrated and dedicated Intel GPUs (oneAPI.jl).

Recently, flexible motion models were implemented, enabling many cardiac MRI applications. Regular rigid transforms like translation and rotation, non-rigid motion, and flow are possible. Moreover, these motions can be mixed and matched by putting them inside a MotionList. Phantoms with complex motion can be shared in an HDF5-based file format to enable reproducibility.

Open-source software access instructions: GitHub: https://juliahealth.org/KomaMRI.jl

Cardiac MRI application showcase (Figure 2)

Simple tutorial examples (Figure 3)

We offer numerous tutorials at https://juliahealth.org/KomaMRI.jl/dev/. To run the examples shown in this section, the tutorials can be downloaded on the following pages:

• Low-Field CMRA Optimization: Example 1


• Patient's Motion During Acquisition: Example 2


• Diffusion-induced Signal Attenuation: Example 3

We hope that this document was a good introduction to Koma's capabilities and potential applications in cardiac MRI. We welcome new contributors. If you have any questions, you can talk to Carlos Castillo on the Julia Slack or post them as a GitHub discussion or issue, which we will try to respond to as soon as possible.