dmipy

Diffusion Microstructure Imaging in Python — one physical substrate, two engines:

• dmipy-sim — a JAX Monte-Carlo simulator: spins random-walk through geometric substrates under arbitrary free gradient waveforms G(t), with surface relaxivity and membrane permeability baked into the walk. The forward-truth oracle.
• dmipy-fit — analytical multi-compartment signal models + JAX GPU fitting, with T2 and surface relaxivity as composable occupancy-gated factors, CSD, and a standard NNLS myelin-water-fraction estimator. The analytical inverse.

They share one free-waveform sequence/substrate interface; the dependency is one-directional, fit → sim. Build a fit and a simulation from the same parameters and they describe the same tissue.

Physics is the specification

Physical laws, invariants, and known analytical results are the correctness criteria — not "the code runs". Every analytical model is validated effect-by-effect against Monte Carlo. (More on the philosophy at dmrai-lab.org.)

from dmipy_fit.signal_models.gaussian_models import G1Ball
from dmipy_fit.signal_models.cylinder_models import C1Stick
from dmipy_fit.core.modeling_framework import MultiCompartmentModel

ball_stick = MultiCompartmentModel([G1Ball(), C1Stick()])
fit = ball_stick.fit(scheme, data, solver="jax")     # whole slice on GPU

Start with Install, then the forward (dmipy-sim) or inverse (dmipy-fit) quickstart, or the worked surface-relaxivity / MWF walkthrough.