dmipy

Diffusion Microstructure Imaging in Python — the umbrella over the two engines, and the home of the documentation at dmipy.org.

• dmipy-sim — JAX Monte-Carlo forward simulator (arbitrary G(t), restricted diffusion, surface relaxivity, permeability, T2).
• dmipy-fit — analytical multi-compartment fitting + JAX GPU (T2 + surface-relaxivity factors, CSD, NNLS myelin-water fraction).

One free-waveform substrate interface, fit → sim.

pip install dmipy            # dmipy-sim + dmipy-fit
pip install "dmipy[cuda12]"  # + JAX GPU

import dmipy_sim, dmipy_fit   # no importable `dmipy`; import the engines directly

Upgrading from dmipy 1.x? dmipy 2.0 is a ground-up rewrite with a different architecture and API. The 1.x line (the original 2019 toolbox, Fick–Wassermann–Deriche) was a single importable dmipy package; 2.x is a meta-package with no importable dmipy module — you import dmipy_sim / dmipy_fit. Old import dmipy code will not run under 2.x. To stay on the original toolbox, pin pip install "dmipy<2" (it remains available on PyPI).

This repo

• pyproject.toml — the thin dmipy meta-package (installs the two engines).
• docs/ + mkdocs.yml — the dmipy.org site (MkDocs Material; built and deployed by .github/workflows/docs.yml).
• papers/ — reproducible manuscripts (source + PDF + rerunnable figures); currently papers/surface_relaxivity_bias/ — "Surface relaxivity biases diffusion and relaxometry microstructure estimates." See papers/README.md.

Build the docs locally:

pip install ".[docs]"
mkdocs serve

License

Dual-licensed: GNU AGPL-3.0 for open-source use, or a commercial license for proprietary/closed use — the same terms apply to both engines. See LICENSE and LICENSING.md (commercial: rutger.fick@dmrai-lab.org).

The lab and its philosophy: dmrai-lab.org.