Direct stress imaging from shear wave propagation

Technical Summary: Direct Stress Imaging from Shear Wave Propagation

Problem Statement
Quantitative imaging of stress fields in heterogeneous solids remains a significant challenge in mechanics and biomechanics. Unlike displacement, which is directly observable, stress is a tensorial state variable that cannot be measured directly. Conventional approaches typically infer stress indirectly from measured deformations using constitutive models and assumptions regarding material properties and reference configurations. These methods often require extensive material calibration and struggle to apply to soft, heterogeneous, or biologically relevant materials. Furthermore, existing non-destructive techniques based on the acoustoelastic effect (where pre-existing stress alters wave propagation) generally rely on local wave-speed measurements, calibrated higher-order elastic constants, or simplified assumptions about geometry and homogeneity. Consequently, high-resolution mapping of internal stress fields, particularly in complex materials, has remained largely inaccessible.

Methodology: Acoustoelastic Imaging (AEI)
The authors introduce Acoustoelastic Imaging (AEI), a non-destructive framework designed to reconstruct stress fields directly from shear wave propagation without explicit constitutive model specification or material-parameter calibration.

• Theoretical Basis: The method leverages the acoustoelastic effect, where pre-existing stress modifies the incremental elastic response of a material. The governing wave equation for infinitesimal displacements in a pre-stressed configuration contains elasticity coefficients directly linked to the stress state. Specifically, in the principal coordinate system, the difference between principal stresses ($\sigma_1 - \sigma_2$) is encoded in the difference between specific components of the Eulerian elasticity tensor ($A^0_{1'2'1'2'} - A^0_{2'1'2'1'}$).
• Inverse Problem Formulation: AEI treats stress recovery as an inverse problem. Instead of estimating local wave speeds, the framework seeks to reconstruct the spatially varying coefficients of the governing wave equations from full-field wave measurements.
• Physics-Informed Deep Learning: To solve this inverse problem efficiently, the authors employ Full Shear Waveform Inversion (FSWI) enabled by Physics-Informed Neural Networks (PINNs).
  • The architecture utilizes six fully connected neural networks to represent unknown fields as continuous functions of space. One network predicts the displacement field (parameterized via a stream function to satisfy incompressibility) and the Lagrange multiplier, while the remaining networks infer the principal stress directions and the wave-equation coefficients.
  • The loss function combines a data-driven term (minimizing the mismatch between predicted and measured displacement fields) and a physics-informed term (enforcing the governing acoustoelastic wave equations).
  • The inversion iteratively updates the stress difference ($\sigma_1 - \sigma_2$) and the principal direction ($\theta$) to minimize residuals.

Key Contributions and Results
The paper validates AEI through numerical simulations and experimental ultrasound shear wave elastography (SWE) measurements.

1. Reconstruction of Non-Uniform and Residual Stresses:

   • In numerical simulations involving a material with spatially non-uniform residual stresses (generated by constrained thermal expansion), AEI successfully reconstructed both the stress magnitude and principal directions.
   • The method demonstrated the ability to separate stress from material stiffness. In heterogeneous materials containing a stiff inclusion within a softer matrix, AEI accurately recovered the stress field despite elastic heterogeneity. In contrast, conventional SWE interpreted stress-induced wave changes as stiffness variations, leading to smoothed modulus maps.
   • The reconstruction of individual elasticity coefficients ($\alpha$ and $\gamma$) was found to be non-unique (differing by a uniform offset), but their difference, which determines the stress, was uniquely recoverable.

2. Sub-Wavelength Resolution:

   • A notable finding is the achievement of sub-wavelength spatial resolution (< 0.28$\lambda$). By utilizing the full wave field, including near-field components and wavefront distortions, AEI resolves stress features significantly smaller than the shear wavelength.
   • This resolution was shown to be largely independent of excitation frequency (comparing 6 kHz and 12 kHz), contrasting with conventional far-field imaging limited by diffraction. However, the authors note that using only partial wave-field data (e.g., a single displacement component) degrades resolution, though higher frequencies can partially compensate.

3. Experimental Validation:

   • The framework was implemented using a standard ultrasound SWE platform on a PVA hydrogel sample. Shear waves were generated by a PZT actuator and recorded via ultrafast ultrasound imaging.
   • Despite weak apparent distortions in the wave field (average velocity increase from 4.3 m/s to 4.5 m/s), AEI reconstructed stress fields and principal directions that agreed quantitatively with finite element reference solutions. The method proved robust to experimental noise, maintaining accuracy for signal-to-noise ratios exceeding approximately 30 dB.

Significance and Claims
The authors position AEI as a general framework for high-resolution stress imaging that shifts the paradigm from kinematic imaging (measuring displacement) to tensorial state-variable imaging (measuring stress).

• Direct Measurement: The work establishes that stress can be treated as a measurable mechanical state variable encoded directly in wave propagation, independent of specific constitutive laws.
• Complementarity: AEI and conventional elastography are presented as complementary modalities; while elastography characterizes material stiffness, AEI characterizes the mechanical state (stress) of the material.
• Applicability: The method provides a route toward non-invasive mapping of internal mechanical states in complex materials and biological tissues.
• Limitations: The authors modestly acknowledge current limitations, including the assumption of relatively smooth stress variations and the use of simplified wave-field representations. They note that experimental validation is currently limited to controlled environments and that extending the framework to compressible media and more complex nonlinearities is a necessary future step.

In summary, the paper demonstrates that by combining full-waveform inversion with physics-informed learning, it is possible to reconstruct internal stress fields with high spatial resolution and without prior material calibration, offering a new tool for understanding the mechanical behavior of soft materials and biological tissues.