KomaMRI.jl

Why it matters

I started KomaMRI because MRI simulations were slow and difficult to use, and support for simulating sequences defined in the open-source Pulseq format was limited. I wanted to make it practical to test an MRI experiment before taking it to the scanner.

Contribution

I designed the framework from the ground up for speed, extensibility, and simplicity: a graphical interface for students getting started, and a powerful API for researchers building their own methods. First presented as MRIsim.jl, it connects sequence definitions, digital phantoms, GPU-accelerated simulation, and image reconstruction.

Speed and accuracy

In the original validation study, KomaMRI produced mean absolute differences below 0.1% relative to JEMRIS in the tested simulations. In a separate student experiment, it ran eight times faster than JEMRIS on participants’ personal computers.

More recently, I developed the theory and high-performance GPU kernels behind our Magnus-based Bloch simulations. Higher-order methods improve accuracy at a given time step, or allow larger steps for comparable accuracy; the method documentation explains this trade-off.

I also applied these methods to inverse RF pulse design, demonstrating cases where numerical inaccuracies in existing simulation methods lead to incorrect pulse designs. The key lesson is that a fast forward simulation is not enough: its accuracy matters when an optimizer uses it to design an experiment.

Motion and flow

I was heavily involved in KomaMRI’s arbitrary-motion extension, which we continue to develop at Stanford. It makes motion part of the simulated experiment rather than treating the object as static.

Our ISMRM 2026 flow study provides an initial in-silico evaluation of joint velocity–acceleration encoding for 4D-Flow MRI (4D-FlowP), toward simultaneous flow assessment and more robust pressure-gradient estimation.

Open development

As of September 8, 2026, KomaMRI on GitHub has 220 stars and 26 contributors, excluding bot accounts. Community contributions expand the framework, including its GPU support and motion capabilities. The documentation brings together introductory tutorials, reproducible MRI examples, and an API reference for advanced users.