Why arterial tissue has to be tested in two directions at once

A uniaxial tensile test gives a clean curve, a reproducible number and a good fit. Applied to a blood vessel, it can also give a material model that is confidently wrong. The reason has less to do with the quality of the measurement than with what a single loading direction is able to reveal.

An artery is never loaded in one direction

In a pressurised vessel the wall carries load along two axes at once. For a thin walled cylinder the circumferential stress is roughly twice the axial stress, and both are present in every cardiac cycle. The vessel is also stretched axially in situ, so even at rest the tissue is not in the unloaded state it occupies on a bench.

The wall is not a homogeneous rubber either. It is a layered composite in which collagen fibres run in preferred orientations, embedded in elastin, smooth muscle and ground substance. Stiffness therefore depends on the direction of loading, and the dependence is strong. A uniaxial test samples one path through that behaviour: the curve is true for that strip, in that direction, but it cannot show how the tissue responds when both directions are loaded together, because the transverse direction stays free to contract throughout.

The difficulty is identifiability, not accuracy

Constitutive models for arterial tissue typically carry parameters for matrix stiffness, fibre stiffness, fibre non-linearity, mean fibre orientation and fibre dispersion. Several different combinations of those parameters reproduce a single uniaxial curve to within experimental scatter. The fit looks excellent. The parameters are nonetheless not uniquely determined.

That becomes visible as soon as the model is asked to do anything beyond reproducing the test it was fitted to. Parameter sets that agree on the uniaxial response can disagree substantially once the same tissue is pressurised or loaded at a different ratio between the axes. The model has not failed. It was never constrained enough to be identified.

Biaxial testing constrains it. Loading two perpendicular axes at several different ratios requires a single parameter set to reproduce several independent responses at once. Fibre orientation in particular becomes identifiable, because orientation determines how load is shared between the axes, and that sharing changes measurably when the ratio changes. A model fitted this way is not automatically correct, but it has been asked a harder question.

What the test actually is

The device loads a sample along two perpendicular axes by imposing force or displacement in each. Four motors and four load cells are controlled independently, so the axes can be driven at different ratios and each can run in force or displacement control. Samples are square or cruciform, mounted using rakes, sutures or clamps depending on the tissue.

The hard part is holding the tissue

Loading a soft tissue sample is straightforward. Holding it without changing what is being measured is not.

A clamp grips the full width of the sample edge, so when the perpendicular axis is loaded that edge cannot contract laterally. The boundary condition, rather than the material, then governs much of the strain field, and the centre of the sample deforms partly in response to how it was held. Rakes avoid most of this. Samples are mounted on sets of parallel needles roughly a millimetre apart, and because the needles are discrete rather than continuous, the tissue between them keeps its freedom to move laterally and deforms more nearly as the wall does in situ.

This reduces the problem without removing it, since tissue immediately around each needle is still locally distorted. Which is why strain is not taken from the actuators at all.

Measuring the specimen, not the machine

A speckle pattern is applied to the sample surface and tracked optically through the test. The displacement field is reconstructed from how that pattern moves, and only the central portion of the area enclosed by the rakes is used for analysis. Everything near the needles is discarded.

Two further artefacts are worth naming, because both are easy to miss and hard to detect afterwards. A sample that bows slightly out of plane produces apparent in plane strains that were never there. And soft tissue is not stress free when it is mounted, so treating the mounted state as the zero strain state introduces a systematic offset. Both are addressable, but only if they are measured rather than assumed.

The conditions around the test

Samples are held in phosphate buffered saline at 37 degrees Celsius, because the response of soft tissue depends on temperature and hydration. Preconditioning cycles precede the recorded cycle, because the first cycles of a biological tissue are not repeatable. None of this is glamorous, and all of it determines whether the parameters mean anything outside the laboratory that produced them.

From curves to a model that can be trusted

Fitting a constitutive model to biaxial data is the beginning rather than the end. The stronger step is to require the fitted model to reproduce a loading state that was not used in the fit: inflation of the intact vessel, or the in vivo pressure diameter behaviour measured by imaging in the same subject. The model is then required to describe the tissue as it behaves in the body, not only as it behaves clamped in a rig.

This is the direction regulatory expectations are moving as well. Credibility frameworks for computational modelling in medical devices, including ASME V&V 40, place the weight of the argument on validation evidence, and validation evidence is an experiment.

Questions worth settling before tissue is cut

How will the sample be held, and what does that boundary condition do to the strain field? How will strain be measured, and over what region? How many load ratios will be run, and is that enough to identify the parameters of the model you intend to fit? And against what independent loading state will the fitted model be validated?

A laboratory that has thought about these will answer quickly and is usually pleased to be asked, because they are the questions that separate a characterisation study from a set of curves.