How to Choose a Material Model

Sentinel implements seven material models. This page helps you pick one. For each model's parameters, constructor, and defaults see Material Models in the API reference; for the constitutive equations see the Mathematical Reference.

Start with Model 1

If you are reconstructing brain, liver, or most other soft tissue, use IsotropicIncompressible. It is the default, it is the best-tested path, and it is the only model the batched GPU driver and the pressure-condensation fast path support.

model = IsotropicIncompressible()

Only move to another model if you have a specific reason below.

Pick by what you are imaging

Your situation	Model
Most soft tissue — brain, liver	IsotropicIncompressible
Compressibility genuinely matters, e.g. lung	IsotropicCompressible
Fluid-saturated tissue, e.g. cartilage	Poroelastic
Three distinct material axes	Orthotropic
Fibre-reinforced tissue — muscle, white matter	TransverseIsotropicV1
Fibre-reinforced, alternate parameterization	TransverseIsotropicV2
General anisotropy, Itzkov-Askel incompressibility	GeneralizedAnisotropic

Check you can meet the model's requirements

Anisotropic models need more than displacement data, and one model needs a different mesh. Confirm you can satisfy these before committing:

• Fibre directions. Orthotropic, TransverseIsotropicV1, TransverseIsotropicV2, and GeneralizedAnisotropic each need a fibre direction per element, from a co-registered DTI acquisition — Orthotropic uses it to define its principal axis. See How to Add DTI Fiber Directions. Check any model with uses_fiber_directions.

• Mesh element type. Every model uses Hex27 except Poroelastic, which uses Tet4. A Hex27 mesh will not drive a poroelastic reconstruction.

• More parameters means more data. Each additional independent parameter has to be constrained by the same displacement measurement. An anisotropic model fitted to data that does not resolve the anisotropy will produce confident, meaningless numbers.

Conventions that catch people out

Complex moduli use ${\mu }^{\ast }={\mu }^{\prime }+i{\mu }^{″}$ with ${\mu }^{″}<0$ for physical damping. If your storage and loss moduli come out with the loss term positive, check this sign convention before concluding the reconstruction failed.

Next

• Material Models — parameters, constructors, and the type hierarchy.

• Mathematical Reference — the constitutive equation for each model.

• How to Run Inversions in Parallel — which models the batched driver supports.