JuliaHealthZoo

Appearance

KomaMRI.jl ​

KomaMRI.jl is the simulation pillar of this workflow. KomaMRI is a fast and extensible MRI simulation framework in Julia, with Bloch-equation simulation and CPU/GPU execution.

Core characteristics used in this workflow:

  • Pulseq in, ISMRMRD out.

  • Fast CPU and GPU simulation.

  • Open-source MRI simulation framework.

Why KomaMRI in this workflow ​

KomaMRI lets us create controlled synthetic acquisitions before running reconstruction algorithms. This is important for:

  • testing algorithm behavior under known conditions

  • comparing reconstruction backends with the same input data

In user terms, this means you can tune and validate a reconstruction workflow before depending on scanner-acquired datasets.

KomaMRI also emphasizes practical interoperability, including Pulseq and ISMRMRD-oriented workflows. That is useful in research settings where simulated and measured data pipelines need to align as closely as possible.

Simulation parameters (sim_params) ​

KomaMRI exposes key simulation controls through sim_params. Key parameters include:

  • Delta t ("Δt"): gradient raster time

  • Delta t rf ("Δt_rf"): RF raster time

  • gpu: run on GPU (if available) or CPU

  • precision: "f32" or "f64"

  • Nblocks: split simulation time into blocks to reduce memory usage

  • Nthreads: parallel split over phantom/spins

The same parameter dictionary also includes sim_method, return_type, and other extensibility controls.

Example configuration pattern ​

julia
using KomaMRICore

sim_params = KomaMRICore.default_sim_params()
sim_params["gpu"] = false
sim_params["precision"] = "f64"
sim_params["Nblocks"] = 40
sim_params["Δt"] = 1e-6
sim_params["Δt_rf"] = 1e-6

# raw = simulate(obj, seq, sys; sim_params)

Synthetic-data generation pattern:

julia
# raw synthetic acquisition for downstream tasks
raw = simulate(obj, seq, sys; sim_params)

# Example uses:
# 1) training data generation
# 2) reconstruction benchmarking in MRIReco/BART

Customizable components ​

KomaMRI simulation is driven by:

  • phantom (obj)

  • sequence (seq)

  • scanner/system (sys)

This matches the requirement for customizable sequence, phantom, and scanner setups.

Synthetic data use in downstream steps ​

The raw simulated signal can be passed to:

  • MRIReco.jl (Julia-native reconstruction)

  • BART via BartIO.jl (cross-tool comparison)

Official sources ​