The Illusion of Fit: Spatially Resolved Assessment of Constitutive Model Validity in Elastography and Physics-Based Inverse Problems

P.S. Koutsourelakis; Vincent C. Scholz

arxiv: 2502.07415 · v2 · submitted 2025-02-11 · 📊 stat.ML · cs.LG

The Illusion of Fit: Spatially Resolved Assessment of Constitutive Model Validity in Elastography and Physics-Based Inverse Problems

Vincent C. Scholz , P.S. Koutsourelakis This is my paper

Pith reviewed 2026-05-23 04:08 UTC · model grok-4.3

classification 📊 stat.ML cs.LG

keywords elastographyconstitutive model validityinverse problemsprobabilistic inferencestress fieldmodel mismatchvariational inferencesoft tissue mechanics

0 comments

The pith

A probabilistic framework infers a spatially resolved constitutive precision field that reveals where an assumed material law fails to match observed deformations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Standard elastography inversions can return plausible tissue-property maps even when the chosen constitutive model does not hold locally, creating an undetected illusion of fit. The paper replaces the usual derivation of stress from the constitutive law with an independent latent stress field, allowing direct pointwise comparison against the stress demanded by mechanical equilibrium. Separate precision hyperparameters are attached to the two virtual observables; the equilibrium precision is fixed high while the constitutive precision is learned under a sparsity-promoting prior. Stochastic variational inference then produces a constitutive precision field that flags regions of model invalidity without repeated forward solves. This turns an implicit modeling assumption into an explicit, spatially resolved diagnostic that can be obtained from noisy or sparse displacement data.

Core claim

By treating the stress field as an independent latent variable rather than deriving it from the constitutive law, the framework enables a pointwise comparison between stresses required by mechanical equilibrium and those predicted by the model. Both enter as virtual observables with separate precision hyperparameters; the conservation-law precision is fixed high while the constitutive precision is inferred sparsely. The resulting constitutive precision field maps local model validity and is obtained via stochastic variational inference without repeated forward solves.

What carries the argument

The constitutive precision field: a spatially varying inferred parameter that quantifies pointwise agreement between equilibrium-required stress and the stress predicted by the assumed constitutive model.

If this is right

The inferred constitutive precision field identifies an anisotropic inclusion with a five-order-of-magnitude precision contrast against the valid domain.
The contrast remains robust across 25-35 dB noise levels and four-fold sparser observations.
A phantom experiment on linear elastic material produces no false-positive violations and recovers the true stiffness contrast.
The same separation of precision hyperparameters applies to any physics-based inverse problem in which one equation is known to be exact and the other is model-dependent.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The precision field could serve as a local weight or mask when combining multiple constitutive models during a single inversion.
Similar latent-variable treatments of stress or flux might be used in other inverse problems where one governing equation is trusted more than another.
Clinical elastography pipelines could output both the property map and its validity map so that low-precision regions receive reduced diagnostic weight.

Load-bearing premise

Mechanical equilibrium is treated as exactly true with fixed high precision while the constitutive relation's precision is allowed to vary and is learned from data.

What would settle it

Applying the method to a homogeneous specimen known to obey the assumed constitutive law everywhere and verifying that the resulting precision field remains uniformly high with no spurious low-precision patches.

Figures

Figures reproduced from arXiv: 2502.07415 by P.S. Koutsourelakis, Vincent C. Scholz.

**Figure 2.** Figure 2: The synthetic data (ground truth) was generated with the finite element software Fenics [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 2.** Figure 2: Problem setup. • The Youngs modulus field m = E as based on an element-wise constant triangular 32 × 32 grid, i.e., dim(x) = 1922, • The stress field σ has three fields (σ11, σ12, and σ22), each of which is modeled by an element-wise constant triangular 32 × 32 grid, i.e., dim(χ) = 5766, • The weight functions w are two linear triangular elements on a 32×32 grid, i.e., dim(w) = 2048. • The displacement fie… view at source ↗

**Figure 3.** Figure 3: Expected weighted squared residual r (c) over the iterations (1 unit = 10, 000 iterations) [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

**Figure 4.** Figure 4: Displacement fields u1 (left) and u2 (right). On both subfigures: The observations uˆ are in the top left, the inferred posterior mean of the displacement fields µui are in the top right, the absolute error of these two are in the bottom left, and the absolute errors normed by the observation noise τ −1 are the bottom right subplot. 3.2 Experiment I) In this experiment, the ground truth material field cons… view at source ↗

**Figure 5.** Figure 5: First line shows the stress means µσ and second line shows the 95% credibility intervals. In the columns (f.l.t.r.) are shown the components σ11, σ12 and σ22 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: Shows Logarithmic Young’s modulus field ln E on the left and the precision λ (c) on the right. In the left subplot, the top left plot shows the true field (but transverse isotropic cannot be depicted), the top right shows the approximate posterior mean, the bottom left shows the 95% credibility intervals, and the bottom right shows where approximation envelopes the ground truth. 11 [PITH_FULL_IMAGE:figure… view at source ↗

read the original abstract

Inferring the mechanical properties of soft tissues from measured deformations is a fundamental challenge in elastography. A rarely examined assumption underlying existing approaches is that the assumed constitutive law correctly describes the imaged material. When it fails, inversion still yields plausible-looking estimates - an illusion of fit with no indication of local model invalidity, which can mislead clinical interpretation. We propose a probabilistic framework that transforms constitutive model validity from an implicit assumption into an explicit, spatially resolved inference target. The key is to treat the stress field as an independent latent variable rather than deriving it from the constitutive law. This enables a pointwise comparison between the stress required by mechanical equilibrium and the stress predicted by the assumed constitutive model. Both governing equations enter the probabilistic learning objective as virtual observables with separate precision hyperparameters: the conservation law precision is set a priori to a small value reflecting its undisputed validity, while the constitutive precision is inferred under a sparsity-promoting prior. The resulting constitutive precision field provides an uncertainty-aware map of where the assumed model is supported by the data and where it is not. Inference is carried out via stochastic variational inference and is forward-model-free. We validate the framework on synthetic harmonic elastography experiments on a brain-slice geometry with an anisotropic inclusion. The inferred precision field identifies the inclusion with a five-order-of-magnitude precision contrast against the valid domain, robustly across 25-35 dB noise and four-fold sparser observations. A phantom experiment with ultrasound measurements on a linear elastic material yields no false-positive violations and recovers the true stiffness contrast.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The latent-stress setup with separate precisions is a genuine new target for inference, but the five-order contrast likely depends on fixing conservation precision low and using a sparsity prior rather than emerging from data mismatch alone.

read the letter

The one thing to know is that this paper turns constitutive model validity into an explicit, spatially resolved output by treating stress as an independent latent variable and giving the constitutive equation its own learnable precision field under a sparsity prior, while fixing the conservation-law precision a priori to a small value. That construction does not appear in the referenced prior work and directly targets an assumption that most elastography inversions leave implicit. The synthetic brain-slice experiment with an anisotropic inclusion and the ultrasound phantom test both demonstrate the method producing a clear map, with the reported contrast holding across noise levels and sparser data, and no false positives on the valid phantom material. That is useful practical evidence for the core idea. The main soft spot is the one flagged in the stress test. The abstract sets conservation precision low because its validity is undisputed and places a sparsity-promoting prior on the constitutive precision, yet reports no sensitivity checks on either choice. If relaxing the fixed conservation precision by even one or two orders of magnitude or removing the sparsity prior collapses the contrast, the headline result is an artifact of those modeling decisions rather than a direct readout of pointwise mismatch. The phantom result is reassuring on false positives but does not address this. This work is aimed at researchers in physics-based inverse problems and elastography who want an explicit check on local model validity instead of assuming it everywhere. A reader focused on reliable mechanical property maps in imaging would get value from the framework and the validation experiments shown. I would send it to peer review because the problem it addresses is load-bearing and the proposed inference target is new enough to merit referee scrutiny on the hyperparameter dependence and implementation details.

Referee Report

2 major / 2 minor

Summary. The paper proposes a probabilistic framework for spatially resolved assessment of constitutive model validity in elastography and physics-based inverse problems. It treats the stress field as an independent latent variable to enable pointwise comparison between equilibrium-required stress and constitutive-predicted stress, entering both as virtual observables with separate precision hyperparameters: conservation-law precision fixed a priori to a small value, and constitutive precision inferred under a sparsity-promoting prior. Inference uses stochastic variational inference (forward-model-free). Validation on synthetic harmonic elastography data (brain-slice geometry with anisotropic inclusion) reports a five-order-of-magnitude precision contrast identifying the inclusion, robust across 25-35 dB noise and four-fold sparser observations; a phantom ultrasound experiment on linear elastic material reports no false-positive violations and recovers true stiffness contrast.

Significance. If the central claim holds, the work addresses a significant gap in elastography by converting an implicit modeling assumption into an explicit, uncertainty-aware map of local model validity, potentially reducing misleading property estimates in clinical applications. Strengths include the forward-model-free SVI implementation and dual validation on synthetic and experimental data.

major comments (2)

[Abstract] Abstract: The reported five-order-of-magnitude constitutive precision contrast is presented as arising from pointwise mismatch between equilibrium and constitutive stresses, yet the framework fixes conservation-law precision a priori to a small value (justified only by 'undisputed validity') while applying a sparsity-promoting prior to the constitutive precision hyperparameter. No sensitivity study on the numerical value of the fixed precision or the prior strength is described; if the contrast collapses under modest relaxation of these choices, the headline result would be an artifact of the modeling decisions rather than a robust detection of invalidity.
[Abstract] Abstract (phantom experiment paragraph): The claim of 'no false-positive violations' and recovery of 'true stiffness contrast' is stated without quantitative metrics, error propagation details, or explicit comparison to ground-truth stiffness values; this makes it difficult to evaluate whether the precision field behaves as expected on a known-valid material.

minor comments (2)

The term 'virtual observables' is used without an explicit definition or reference in the abstract; a short clarifying sentence would improve accessibility.
Ensure all hyperparameter values (including the exact a priori conservation precision and sparsity prior parameters) are tabulated or stated with numerical values in the methods for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments identify opportunities to strengthen the presentation of robustness and quantitative support for the claims. We respond to each major comment below and will incorporate revisions accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: The reported five-order-of-magnitude constitutive precision contrast is presented as arising from pointwise mismatch between equilibrium and constitutive stresses, yet the framework fixes conservation-law precision a priori to a small value (justified only by 'undisputed validity') while applying a sparsity-promoting prior to the constitutive precision hyperparameter. No sensitivity study on the numerical value of the fixed precision or the prior strength is described; if the contrast collapses under modest relaxation of these choices, the headline result would be an artifact of the modeling decisions rather than a robust detection of invalidity.

Authors: We agree that the absence of a sensitivity study on the fixed conservation-law precision value and the sparsity-promoting prior hyperparameters leaves open the possibility that the reported contrast depends on these modeling choices. The manuscript selects a small fixed precision to reflect the physical certainty of equilibrium and a sparsity prior to localize potential invalidity, but does not vary these parameters. In revision we will add a dedicated sensitivity analysis that perturbs the fixed precision over multiple orders of magnitude and varies the prior strength, showing that the five-order contrast is preserved under these changes. This will confirm that the result reflects data-driven detection rather than hyperparameter artifacts. revision: yes
Referee: [Abstract] Abstract (phantom experiment paragraph): The claim of 'no false-positive violations' and recovery of 'true stiffness contrast' is stated without quantitative metrics, error propagation details, or explicit comparison to ground-truth stiffness values; this makes it difficult to evaluate whether the precision field behaves as expected on a known-valid material.

Authors: We acknowledge that the abstract statement on the phantom experiment is brief and omits quantitative metrics. The full manuscript describes the linear-elastic phantom setup and the recovered stiffness map, but the abstract does not include explicit comparisons to ground-truth values or error measures. In the revised abstract we will add concise quantitative support, such as the recovered stiffness contrast magnitude and confirmation that the precision field remains uniformly high (no localized low-precision regions) consistent with a valid constitutive model. revision: yes

Circularity Check

0 steps flagged

No significant circularity; inference output independent of inputs

full rationale

The framework infers the constitutive precision field as a latent variable from pointwise mismatch between equilibrium-required stress and constitutive-predicted stress, using stochastic variational inference. The conservation precision is fixed a priori (as stated in the abstract), but the constitutive precision is not defined in terms of itself, nor is the reported five-order contrast a direct renaming or statistical forcing of the sparsity prior or fixed value. No self-citations, uniqueness theorems, or ansatzes are invoked to justify the core result. The derivation chain produces an output quantity that is not equivalent to its modeling choices by construction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on treating stress as an independent latent variable and on the a-priori assignment of conservation-law precision; these are introduced as modeling choices rather than derived from data.

free parameters (2)

constitutive precision hyperparameter
Learned under sparsity-promoting prior to produce the validity map
conservation law precision
Fixed a priori to a small value

axioms (2)

domain assumption Mechanical equilibrium (conservation of momentum) holds with undisputed validity
Used to fix its precision a priori
domain assumption Stress field can be treated as an independent latent variable decoupled from the constitutive law
Core modeling choice enabling the pointwise comparison

pith-pipeline@v0.9.0 · 5823 in / 1323 out tokens · 43958 ms · 2026-05-23T04:08:14.496116+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

[1]

Ophir, I

J. Ophir, I. Cespedes, H. Ponnekanti, Y . Yazdi, X. Li, Elastography: a quantitative method for imaging the elasticity of biological tissues, Ultrasonic imaging 13 (2) (1991) 111–134

work page 1991
[2]

M. M. Doyley, Model-based elastography: a survey of approaches to the inverse elasticity problem, Physics in Medicine & Biology 57 (3) (2012) R35

work page 2012
[3]

Muthupillai, D

R. Muthupillai, D. Lomas, P. Rossman, J. F. Greenleaf, A. Manduca, R. L. Ehman, Magnetic resonance elastography by direct visualization of propagating acoustic strain waves, science 269 (5232) (1995) 1854–1857

work page 1995
[4]

A. S. Khalil, R. C. Chan, A. H. Chau, B. E. Bouma, M. R. K. Mofrad, Tissue elasticity estimation with optical coherence elastography: toward mechanical characterization of in vivo soft tissue, Annals of biomedical engineering 33 (2005) 1631–1639

work page 2005
[5]

Raissi, P

M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational physics 378 (2019) 686–707

work page 2019
[6]

C. R. V ogel, Computational methods for inverse problems, SIAM, 2002

work page 2002
[7]

P. J. Green, K. Łatuszy ´nski, M. Pereyra, C. P. Robert, Bayesian computation: a summary of the current state, and samples backwards and forwards, Statistics and Computing 25 (2015) 835–862

work page 2015
[8]

D. M. Blei, A. Kucukelbir, J. D. McAuliffe, Variational inference: A review for statisticians, Journal of the American statistical Association 112 (518) (2017) 859–877

work page 2017
[9]

Brynjarsdóttir, A

J. Brynjarsdóttir, A. O’Hagan, Learning about physical parameters: The importance of model discrepancy, Inverse problems 30 (11) (2014) 114007

work page 2014
[10]

Kaipio, E

J. Kaipio, E. Somersalo, Statistical and computational inverse problems, V ol. 160, Springer Science & Business Media, 2006

work page 2006
[11]

G. A. Holzapfel, Similarities between soft biological tissues and rubberlike materials, in: Constitutive models for rubber IV , Routledge, 2017, pp. 607–617. 12

work page 2017
[12]

M. C. Kennedy, A. O’Hagan, Bayesian calibration of computer models, Journal of the Royal Statistical Society: Series B (Statistical Methodology) 63 (3) (2001) 425–464

work page 2001
[13]

L. M. Berliner, K. Jezek, N. Cressie, Y . Kim, C. Q. Lam, C. J. Van Der Veen, Modeling dynamic controls on ice streams: a bayesian statistical approach, Journal of Glaciology 54 (187) (2008) 705–714

work page 2008
[14]

Andrés Arcones, M

D. Andrés Arcones, M. Weiser, P.-S. Koutsourelakis, J. F. Unger, Model bias identification for bayesian calibration of stochastic digital twins of bridges, Applied Stochastic Models in Business and Industry (2024)

work page 2024
[15]

Pernot, F

P. Pernot, F. Cailliez, A critical review of statistical calibration/prediction models handling data inconsistency and model inadequacy, AIChE Journal 63 (10) (2017) 4642–4665

work page 2017
[16]

Pernot, The parameter uncertainty inflation fallacy, The Journal of Chemical Physics 147 (10) (2017)

P. Pernot, The parameter uncertainty inflation fallacy, The Journal of Chemical Physics 147 (10) (2017)

work page 2017
[17]

Sargsyan, H

K. Sargsyan, H. N. Najm, R. Ghanem, On the statistical calibration of physical models, Interna- tional Journal of Chemical Kinetics 47 (4) (2015) 246–276

work page 2015
[18]

Sargsyan, X

K. Sargsyan, X. Huan, H. N. Najm, Embedded model error representation for bayesian model calibration, International Journal for Uncertainty Quantification 9 (4) (2019)

work page 2019
[19]

R. E. Kass, A. E. Raftery, Bayes factors, Journal of the american statistical association 90 (430) (1995) 773–795

work page 1995
[20]

Hoerig, J

C. Hoerig, J. Ghaboussi, Y . Wang, M. F. Insana, Machine learning in model-free mechanical property imaging: Novel integration of physics with the constrained optimization process, Frontiers in Physics 9 (2021) 600718

work page 2021
[21]

Newman, J

W. Newman, J. Ghaboussi, M. Insana, Improving image quality in a new method of data-driven elastography, in: Medical Imaging 2024: Ultrasonic Imaging and Tomography, V ol. 12932, SPIE, 2024, pp. 51–56

work page 2024
[22]

P.-S. Koutsourelakis, A novel bayesian strategy for the identification of spatially varying material properties and model validation: an application to static elastography, International Journal for Numerical Methods in Engineering 91 (3) (2012) 249–268

work page 2012
[23]

Bruder, P.-S

L. Bruder, P.-S. Koutsourelakis, Beyond black-boxes in bayesian inverse problems and model validation: applications in solid mechanics of elastography, International Journal for Uncertainty Quantification 8 (5) (2018)

work page 2018
[24]

Kaltenbach, P.-S

S. Kaltenbach, P.-S. Koutsourelakis, Incorporating physical constraints in a deep probabilistic machine learning framework for coarse-graining dynamical systems, Journal of Computational Physics 419 (2020) 109673

work page 2020
[25]

V . C. Scholz, Y . Zang, P.-S. Koutsourelakis, Weak neural variational inference for solving bayesian inverse problems without forward models: applications in elastography, Computer Methods in Applied Mechanics and Engineering 433 (2025) 117493

work page 2025
[26]

J. M. Bardsley, J. Kaipio, Gaussian markov random field priors for inverse problems., Inverse Problems & Imaging 7 (2) (2013). A Alternative Formulation of the Proposed Method This alternative formulation focuses on how the fields are constructed from random variables, with particular emphasis on how the latent variablez generates the displacement field u...

work page 2013
[27]

Updating the approximate posterior qξ(λ(c)) according to (36) given samples of {z, x, χ}, and

work page
[28]

Updating the approximate posterior q(z, x, χ) according to the SVI scheme introduced in section 2, where we can calculate D λ(c) i E = a λ(c) i b λ(c) i in closed form. C Transversaly isotropic material law A transversely isotropic material is one with physical properties that are symmetric about an axis a, while the properties in said direction may diffe...

work page

[1] [1]

Ophir, I

J. Ophir, I. Cespedes, H. Ponnekanti, Y . Yazdi, X. Li, Elastography: a quantitative method for imaging the elasticity of biological tissues, Ultrasonic imaging 13 (2) (1991) 111–134

work page 1991

[2] [2]

M. M. Doyley, Model-based elastography: a survey of approaches to the inverse elasticity problem, Physics in Medicine & Biology 57 (3) (2012) R35

work page 2012

[3] [3]

Muthupillai, D

R. Muthupillai, D. Lomas, P. Rossman, J. F. Greenleaf, A. Manduca, R. L. Ehman, Magnetic resonance elastography by direct visualization of propagating acoustic strain waves, science 269 (5232) (1995) 1854–1857

work page 1995

[4] [4]

A. S. Khalil, R. C. Chan, A. H. Chau, B. E. Bouma, M. R. K. Mofrad, Tissue elasticity estimation with optical coherence elastography: toward mechanical characterization of in vivo soft tissue, Annals of biomedical engineering 33 (2005) 1631–1639

work page 2005

[5] [5]

Raissi, P

M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational physics 378 (2019) 686–707

work page 2019

[6] [6]

C. R. V ogel, Computational methods for inverse problems, SIAM, 2002

work page 2002

[7] [7]

P. J. Green, K. Łatuszy ´nski, M. Pereyra, C. P. Robert, Bayesian computation: a summary of the current state, and samples backwards and forwards, Statistics and Computing 25 (2015) 835–862

work page 2015

[8] [8]

D. M. Blei, A. Kucukelbir, J. D. McAuliffe, Variational inference: A review for statisticians, Journal of the American statistical Association 112 (518) (2017) 859–877

work page 2017

[9] [9]

Brynjarsdóttir, A

J. Brynjarsdóttir, A. O’Hagan, Learning about physical parameters: The importance of model discrepancy, Inverse problems 30 (11) (2014) 114007

work page 2014

[10] [10]

Kaipio, E

J. Kaipio, E. Somersalo, Statistical and computational inverse problems, V ol. 160, Springer Science & Business Media, 2006

work page 2006

[11] [11]

G. A. Holzapfel, Similarities between soft biological tissues and rubberlike materials, in: Constitutive models for rubber IV , Routledge, 2017, pp. 607–617. 12

work page 2017

[12] [12]

M. C. Kennedy, A. O’Hagan, Bayesian calibration of computer models, Journal of the Royal Statistical Society: Series B (Statistical Methodology) 63 (3) (2001) 425–464

work page 2001

[13] [13]

L. M. Berliner, K. Jezek, N. Cressie, Y . Kim, C. Q. Lam, C. J. Van Der Veen, Modeling dynamic controls on ice streams: a bayesian statistical approach, Journal of Glaciology 54 (187) (2008) 705–714

work page 2008

[14] [14]

Andrés Arcones, M

D. Andrés Arcones, M. Weiser, P.-S. Koutsourelakis, J. F. Unger, Model bias identification for bayesian calibration of stochastic digital twins of bridges, Applied Stochastic Models in Business and Industry (2024)

work page 2024

[15] [15]

Pernot, F

P. Pernot, F. Cailliez, A critical review of statistical calibration/prediction models handling data inconsistency and model inadequacy, AIChE Journal 63 (10) (2017) 4642–4665

work page 2017

[16] [16]

Pernot, The parameter uncertainty inflation fallacy, The Journal of Chemical Physics 147 (10) (2017)

P. Pernot, The parameter uncertainty inflation fallacy, The Journal of Chemical Physics 147 (10) (2017)

work page 2017

[17] [17]

Sargsyan, H

K. Sargsyan, H. N. Najm, R. Ghanem, On the statistical calibration of physical models, Interna- tional Journal of Chemical Kinetics 47 (4) (2015) 246–276

work page 2015

[18] [18]

Sargsyan, X

K. Sargsyan, X. Huan, H. N. Najm, Embedded model error representation for bayesian model calibration, International Journal for Uncertainty Quantification 9 (4) (2019)

work page 2019

[19] [19]

R. E. Kass, A. E. Raftery, Bayes factors, Journal of the american statistical association 90 (430) (1995) 773–795

work page 1995

[20] [20]

Hoerig, J

C. Hoerig, J. Ghaboussi, Y . Wang, M. F. Insana, Machine learning in model-free mechanical property imaging: Novel integration of physics with the constrained optimization process, Frontiers in Physics 9 (2021) 600718

work page 2021

[21] [21]

Newman, J

W. Newman, J. Ghaboussi, M. Insana, Improving image quality in a new method of data-driven elastography, in: Medical Imaging 2024: Ultrasonic Imaging and Tomography, V ol. 12932, SPIE, 2024, pp. 51–56

work page 2024

[22] [22]

P.-S. Koutsourelakis, A novel bayesian strategy for the identification of spatially varying material properties and model validation: an application to static elastography, International Journal for Numerical Methods in Engineering 91 (3) (2012) 249–268

work page 2012

[23] [23]

Bruder, P.-S

L. Bruder, P.-S. Koutsourelakis, Beyond black-boxes in bayesian inverse problems and model validation: applications in solid mechanics of elastography, International Journal for Uncertainty Quantification 8 (5) (2018)

work page 2018

[24] [24]

Kaltenbach, P.-S

S. Kaltenbach, P.-S. Koutsourelakis, Incorporating physical constraints in a deep probabilistic machine learning framework for coarse-graining dynamical systems, Journal of Computational Physics 419 (2020) 109673

work page 2020

[25] [25]

V . C. Scholz, Y . Zang, P.-S. Koutsourelakis, Weak neural variational inference for solving bayesian inverse problems without forward models: applications in elastography, Computer Methods in Applied Mechanics and Engineering 433 (2025) 117493

work page 2025

[26] [26]

J. M. Bardsley, J. Kaipio, Gaussian markov random field priors for inverse problems., Inverse Problems & Imaging 7 (2) (2013). A Alternative Formulation of the Proposed Method This alternative formulation focuses on how the fields are constructed from random variables, with particular emphasis on how the latent variablez generates the displacement field u...

work page 2013

[27] [27]

Updating the approximate posterior qξ(λ(c)) according to (36) given samples of {z, x, χ}, and

work page

[28] [28]

Updating the approximate posterior q(z, x, χ) according to the SVI scheme introduced in section 2, where we can calculate D λ(c) i E = a λ(c) i b λ(c) i in closed form. C Transversaly isotropic material law A transversely isotropic material is one with physical properties that are symmetric about an axis a, while the properties in said direction may diffe...

work page