Pith. sign in

REVIEW 3 major objections 6 minor 44 references

As gliomas grow under MRI watch, models that couple growth to tissue mechanics stay more plausible than pure reaction–diffusion, and linear versus hyperelastic laws should be re-ranked for each animal as new scans arrive.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 17:28 UTC pith:GZHIDJYR

load-bearing objection Solid sequential Bayesian model-selection paper: mechanical coupling is favored over pure RD on four rats; linear-vs-hyperelastic ranking is real but only weakly identified from ϕ alone. the 3 major comments →

arxiv 2607.04551 v1 pith:GZHIDJYR submitted 2026-07-05 cs.CE

Dynamic Image-Informed Selection of Biomechanical Tumor Growth Models

Abdullah Al Noman , Pratyush Kumar Singh , David A Hormuth II , Danial Faghihi This is my paper
classification cs.CE
keywords biomechanical tumor growthfinite deformation elasticitydynamic model selectionimage-informed modelingBayesian inferenceglioblastomaMRIreaction-diffusion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper asks whether subject-specific glioma forecasts from longitudinal MRI need mechanical coupling, and which constitutive description of brain tissue is supported as tumor burden rises. It assimilates murine MRI into three competing reaction–diffusion models—one without mechanics, one with linear elasticity, one with hyperelastic finite-strain mechanics—updating spatially varying diffusivity, proliferation, and stiffness fields by sequential Bayesian filtering. Posterior model plausibility is recomputed after each scan so the preferred model can change for the next one-scan-ahead prediction. Across the four animals studied, mechanically coupled models consistently outrank the uncoupled model; hyperelasticity often gains support later, when mass effect and stress matter more. Morphology metrics alone cannot separate the two mechanical models, yet their stress, displacement, and inferred stiffness fields diverge, so the paper argues that constitutive choice must be data-driven and sequential rather than fixed once for all subjects.

Core claim

Within this longitudinal murine glioma MRI dataset, reaction–diffusion models that include mass effect and stress-mediated feedback are more plausible than pure reaction–diffusion for subject-specific one-scan-ahead prediction, and the relative support for linear elastic versus hyperelastic coupling evolves with each animal and imaging time, often favoring the hyperelastic form later when deformation is larger.

What carries the argument

Sequential Bayesian model selection via posterior model plausibility: at each assimilation time the model evidence of each candidate biomechanical formulation is approximated from a Laplace posterior on high-dimensional spatial parameter fields, then used to re-rank models for the next prediction.

Load-bearing premise

Model ranking rests almost entirely on how well each model matches MRI-derived tumor volume fraction; tissue displacement and stress are never measured directly, so mechanical constitutive laws are judged only through their indirect effect on that single image quantity under fixed coupling parameters and a two-dimensional slice approximation.

What would settle it

In a larger cohort, or with direct deformation measurements, if posterior plausibility no longer favors mechanically coupled models over pure reaction–diffusion, or if linear elasticity remains preferred even at late high-burden times when independent strain data show finite deformation, the central ranking claim would fail.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The manuscript develops a sequential Bayesian inference and dynamic model-selection framework that assimilates longitudinal murine MRI-derived tumor volume fraction to calibrate spatially varying diffusivity, proliferation, and stiffness in three competing glioma growth models: uncoupled reaction–diffusion (RD), RD coupled to linear elasticity, and RD coupled to compressible Neo-Hookean hyperelasticity. Posterior model plausibility is recomputed at each assimilation time for one-scan-ahead prediction. Across four rats, mechanically coupled models are consistently more plausible than pure RD, while relative support for linear versus hyperelastic coupling evolves with subject and time, often favoring hyperelasticity later. The authors further show that the two mechanical models can produce similar tumor morphology while yielding distinct displacement, stress, and inferred stiffness fields, and they validate the Laplace approximation against gpCN sampling for selected marginals, boundaries, and two-model relative plausibility.

Significance. If the results hold under the stated limitations, the paper makes a useful contribution to image-informed biomechanical tumor modeling: it treats constitutive choice as a sequential, subject-specific inference problem rather than a fixed a priori assumption, and it couples high-dimensional Matérn GRF priors with scalable Laplace low-rank posteriors and model-evidence evaluation. Strengths include careful sequential validation (one-scan-ahead, not in-sample tautology), explicit comparison of LA versus gpCN, Dice/NTA threshold sensitivity, and public code. The practical message—that mechanical coupling is favored in this murine dataset and that linear versus hyperelastic support can change over time—is of interest for digital-twin style glioma modeling, even though the constitutive ranking rests on indirect observation of mechanics through ϕ alone.

major comments (3)
  1. §2.1–2.3 and Eq. (8): the likelihood compares only predicted tumor volume fraction ϕ to ADC-derived volume fraction; displacement and stress are never observed. The central claim that constitutive assumptions should be chosen sequentially (linear vs hyperelastic) therefore rests on an indirect identification path under fixed C, H, and ν (Table 1; §2.2) and a 2D plane-strain approximation. Figures 11–13 show that Dice/NTA do not separate the two mechanical models, while Fig. 14 shows that linear elasticity compensates by broadly softening E. The manuscript needs a stronger identification argument—e.g., a synthetic-data recovery test with known constitutive law, or explicit discussion of how much of the late-time hyperelastic preference could be compensatory flexibility rather than support for finite-strain kinematics—before the sequential constitutive recommendation can be treated as load
  2. §3.3 and Fig. 11: the claim that hyperelastic coupling “often” receives higher posterior plausibility at later times is supported by four animals and a 10–21 day window. With n=4 and subject-to-subject switching between linear and hyperelastic models, the population-level recommendation that constitutive assumptions “should be evaluated sequentially for each subject rather than fixed a priori” is only weakly powered. The paper should either (i) report formal Bayes factors / log-evidence differences with uncertainty, not only normalized plausibilities, or (ii) clearly reframe the constitutive conclusion as a subject-level observation within this limited cohort rather than a general modeling prescription.
  3. §2.5, Eq. (14) and §3.2: model ranking uses Laplace-approximated evidence with a fixed low-rank truncation (r=50). The LA–gpCN check is valuable but is reported for relative two-model plausibility on one animal/time (≈49% vs 51%) and selected marginals; it does not fully validate the three-model sequential ranking across all rats and times. Because the linear–hyperelastic evidence gap appears small, a modest bias in the Occam penalty or rank truncation could flip the ranking. A rank-sensitivity or evidence-approximation check that covers the full three-model comparison at multiple assimilation times would substantially strengthen the dynamic selection results.
minor comments (6)
  1. Abstract and §1: “stress-mediated feedback of tumor growth” is slightly awkward; “stress-mediated feedback on tumor growth” would be clearer.
  2. §2.2, Eq. (1)–(2): the coupling parameter C multiplies ∇0ϕ as a body force; a short sentence on units and phenomenological status relative to multiphase growth models would help readers outside continuum mechanics.
  3. Table 1 and §3.1: hyper-parameters (ρ_E, ρ_D, ρ_G, σ²_noise) are tuned on Rat III then frozen; state more explicitly whether any re-tuning was tried for other rats and whether model ranking is sensitive to that choice.
  4. Figures 2–5: Dice values are listed in captions but not tabulated; a compact table of Dice/NTA by animal, time, and model would aid comparison.
  5. §4: the digital-twin discussion is appropriate but long relative to the empirical scope; consider tightening so the limitations on 2D, ϕ-only data, and small n remain the focus.
  6. Minor typographical issues: “reaction-–diffusion” double dash in the abstract; inconsistent “RD+Linearelastic” vs “RD+linear elastic” labeling across text and figures.

Circularity Check

1 steps flagged

Central sequential model-selection claim is not circular by construction; only mild hyperparameter tuning on one animal and non-load-bearing literature priors from overlapping authors.

specific steps
  1. fitted input called prediction [§3.1 Calibration parameters and hyper-parameter specification; Table 1]
    "MRI data from Rat III over days 10–16 are used in a grid search, with Pareto-optimal hyper-parameters selected by maximizing Dice agreement and minimizing NTA prediction error at the final prediction time at day 19. These hyper-parameters are then fixed and applied to the remaining subjects."

    For Rat III, prior correlation lengths and observation-noise variance are chosen by optimizing the same morphology metrics later reported as predictive performance at day 19. That is a mild train-on-target hyperparameter fit for one animal’s Dice/NTA numbers; it does not redefine model evidence or force the RD vs mechanical ranking, and hyperparameters are held fixed for the other three subjects.

full rationale

The paper’s load-bearing chain is standard sequential Bayesian filtering and Bayesian model comparison, not a tautology. At each assimilation time, parameters θ are updated from cumulative MRI-derived tumor volume fraction d^(1:k) (Eqs. 5–8), one-scan-ahead predictions for t_{k+1} use only data up to t_k (Eq. 6; Figs. 2–5), and posterior model plausibility is the normalized model evidence (marginal likelihood) under each a priori specified candidate (Eqs. 9–10, 14–15; Fig. 11). That ranking is not forced by defining the models in terms of the ranking, nor by renaming a fitted quantity as a prediction. Discrimination between RD, RD+linear, and RD+hyperelastic is an empirical comparison of how well each PDE hypothesis explains the same ϕ observations under shared priors and fixed C, H, ν—not Eq. X ≡ Eq. Y by construction. Self-citations to prior Hormuth/Faghihi imaging and inference work supply data provenance, prior means, and the scalable LA/gpCN machinery; they do not supply a uniqueness theorem that forbids alternatives or force the mechanical-coupling conclusion. The only mild circularity-adjacent practice is hyperparameter selection (ρ, σ_noise) on Rat III by maximizing Dice / minimizing NTA at day 19, which can slightly favor that animal’s reported morphology metrics but does not define model evidence or force the linear-vs-hyperelastic ranking across subjects. Identifiability concerns (ϕ-only likelihood, unobserved stress/displacement, fixed coupling constants, 2D plane strain) are correctness/identification risks, not circular reductions. Overall circularity is low.

Axiom & Free-Parameter Ledger

7 free parameters · 7 axioms · 0 invented entities

The central claim rests on standard continuum tumor-growth and soft-tissue mechanics plus Bayesian inverse-problem machinery, with several free hyperparameters and fixed biophysical constants that are not jointly inferred from the MRI volume-fraction data. No new physical entities are postulated; the ‘invention’ is the sequential selection workflow and its application to three pre-specified models.

free parameters (7)
  • Prior means of log(E), log(D), log(G) = log(E)=-0.1562; log(D)=-0.3492; log(G)=-1.8965
    Spatially homogeneous prior means taken from prior Wistar glioma studies and held fixed for sequential inference (Table 1).
  • Prior variances of log(E), log(D), log(G) = 0.01, 0.09, 0.0064
    Control prior strength on stiffness, diffusivity, and proliferation fields (Table 1).
  • Prior correlation lengths ρ_E, ρ_D, ρ_G = 5.5 mm, 6.5 mm, 3.5 mm
    Selected via grid search/cross-validation on Rat III days 10–16 maximizing Dice and minimizing NTA at day 19, then applied to all subjects (§3.1).
  • Observation noise variance σ²_noise = 1.23e-3
    Aggregate model+measurement noise in the Gaussian likelihood; chosen with the same Rat III hyperparameter procedure (Table 1).
  • Fixed coupling and material constants H, C, ν = H=1.5 kPa^-1; C=0.65 kPa; ν=0.45
    Not inferred as fields; prescribed globally because of weak identifiability from volume fraction alone (§3.1; sensitivity ±25% for C,H in SI).
  • Tumor volume-fraction threshold for Dice/NTA = ϕ=0.25
    Defines tumor region for predictive metrics; modest sensitivity reported for Rat III day 19 (§3).
  • Low-rank Hessian truncation rank r = 50
    Number of retained data-misfit eigenmodes in Laplace posterior/evidence approximation; r=50 used for all animals (§3.2).
axioms (7)
  • domain assumption Tumor evolution is adequately described by logistic reaction–diffusion for a single volume-fraction field without explicit necrosis/apoptosis compartments.
    Stated as intentional parsimony constrained by MRI volume-fraction data (§2.2; §4).
  • domain assumption ADC maps convert to normalized tumor volume fraction via an inverse cellularity relationship within the segmented tumor.
    Primary data construction for likelihood (§2.1); literature-supported but indirect.
  • domain assumption Mass effect is a body force proportional to ∇ϕ with coefficient C; stress modulates diffusivity as exp(-H p_τ).
    Phenomenological coupling in Eqs. (1)–(2); form fixed a priori.
  • domain assumption Tissue mechanics are plane-strain compressible Neo-Hookean (or small-strain linear elasticity) with u=0 on the brain boundary.
    §2.2 constitutive law and BCs; 2D slice approximation of 3D brain.
  • standard math Spatially varying parameters admit Gaussian Matérn random-field priors via SPDE (7); likelihood is i.i.d. Gaussian on ϕ−d.
    Standard Bayesian inverse-problem assumptions (§2.3).
  • ad hoc to paper Laplace approximation with low-rank Hessian update yields reliable model evidence for ranking competing PDE models.
    Eq. (14); partially checked vs gpCN for two mechanical models on Rat III day 19 (§2.5; §3.2).
  • ad hoc to paper Equal prior model probabilities π_pr(M_i)=1/m.
    Used to reduce posterior model probability to normalized evidence (Eq. 15).

pith-pipeline@v1.1.0-grok45 · 24398 in / 3962 out tokens · 41393 ms · 2026-07-11T17:28:41.943103+00:00 · methodology

0 comments
read the original abstract

Glioblastoma progression is strongly influenced by evolving mechanical interactions between the tumor and surrounding brain tissue. However, the extent to which finite-deformation mechanics and constitutive assumptions improve subject-specific prediction as tumor burden evolves remains unclear. We introduce a sequential Bayesian inference and dynamic model selection framework that assimilates longitudinal murine magnetic resonance imaging (MRI) data to calibrate spatially varying tumor diffusivity, proliferation rate, and tissue stiffness in biomechanical tumor growth models. Competing formulations were compared at each imaging time, including reaction-diffusion without mechanics and reaction-diffusion coupled to linear elasticity or hyperelastic mechanics, using posterior model plausibility to adapt model choice for individualized one-scan-ahead prediction as new MRI scans are acquired. Across the studied animals, mechanically coupled models were consistently more plausible than the uncoupled reaction-diffusion model, and the evolution of model plausibility indicated an increasing role of mass effect and stress-mediated feedback of tumor growth during progression. While linear and hyperelastic coupled tumor growth models often produced similar tumor morphology, they yield distinct stress, deformation, and inferred stiffness fields, with the hyperelastic formulation often receiving higher posterior plausibility at later imaging times. These results indicate that, within the present longitudinal murine dataset, mechanical coupling is favored for image-informed glioma growth prediction and that constitutive assumptions should be evaluated sequentially for each subject rather than fixed a priori.

Figures

Figures reproduced from arXiv: 2607.04551 by Abdullah Al Noman, Danial Faghihi, David A Hormuth II, Pratyush Kumar Singh.

Figure 1
Figure 1. Figure 1: Scalability of the LA solution of Bayesian inference for Rat III from MRI data of days 10-16. [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sequential one-scan-ahead prediction for Rat I at MAP parameters [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sequential one-scan-ahead prediction for Rat-II at MAP parameters [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sequential one-scan-ahead prediction for Rat-III at MAP parameters [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sequential one-scan-ahead prediction for Rat IV at MAP parameters using the hyperelastic mechanically coupled tumor growth model defined in (1) and (2). Top row: MRI-derived tumor volume fraction. Bottom rows: model predictions obtained by assimilating data up to tk and predicting the subsequent scan at tk+1 (boxed panels). Dice values between the predicted and MRI-derived tumor at Day 14, Day 16, Day 18 a… view at source ↗
Figure 6
Figure 6. Figure 6: Evolution of MAP estimates of spatially varying elastic modulus [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Spatial maps of the local posterior uncertainty reduction index Rθ for the inferred parameter fields for Rat III on Day 19 including the overlay of the model-predicted tumor boundary. Higher values correspond to regions where the longitudinal MRI data provide stronger local constraints on the parameter field, while lower values indicate regions that remain weakly informed by the data. Figures 8—10 compare … view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of LA and gpCN posterior estimates: point-wise kernel density estimates [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of tumor boundary predictions from LA and gpCN posterior samples (blue) [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of displacement and stress fields for Rat III: MAP estimates and posterior [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Temporal evolution of posterior model plausibility, one-scan-ahead predictive Dice score, [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of one-scan-ahead predictions for Rat-I under [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of RD+Hyperelastic and RD+Linearelastic predictions for Rat I in terms of normalized tumor area (NTA), domain-averaged displacement magnitude, and domain-averaged first Piola–Kirchhoff stress magnitude over time. RD+Linearelastic and RD+Hyperelastic, the stiffness field log(E) differs markedly ( [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Evolution of MAP estimates of spatially varying elastic modulus [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 1 canonical work pages

  1. [1]

    Lucci, A

    G. Lucci, A. Agosti, P. Ciarletta, C. Giverso, Coupling solid and fluid stresses with brain tumour growth and white matter tract deformations in a neuroimaging- informed model, Biomechanics and Modeling in Mechanobiology 21 (5) (2022) 1483–1509

    2022

  2. [2]

    Ghahramani, O

    M. Ghahramani, O. Bavi, Biomechanical modeling of glioblastoma progression: a comprehensive review from classic mathematical frameworks to data-driven strate- gies: M. ghahramani and o. bavi, Biomechanics and Modeling in Mechanobiology 25 (1) (2026) 1

    2026

  3. [3]

    Mpekris, S

    F. Mpekris, S. Angeli, A. P. Pirentis, T. Stylianopoulos, Stress-mediated progres- sion of solid tumors: effect of mechanical stress on tissue oxygenation, cancer cell proliferation, and drug delivery, Biomechanics and modeling in mechanobiology 14 (6) (2015) 1391–1402

    2015

  4. [4]

    Goriely, M

    A. Goriely, M. G. Geers, G. A. Holzapfel, J. Jayamohan, A. Jérusalem, S. Sivalo- ganathan, W. Squier, J. A. van Dommelen, S. Waters, E. Kuhl, Mechanics of the brain: perspectives, challenges, and opportunities, Biomechanics and modeling in mechanobiology 14 (5) (2015) 931–965

    2015

  5. [5]

    Helmlinger, P

    G. Helmlinger, P. A. Netti, H. C. Lichtenbeld, R. J. Melder, R. K. Jain, Solid stress inhibits the growth of multicellular tumor spheroids, Nature biotechnology 15 (8) (1997) 778–783

    1997

  6. [6]

    Stylianopoulos, J

    T. Stylianopoulos, J. D. Martin, M. Snuderl, F. Mpekris, S. R. Jain, R. K. Jain, Coevolution of solid stress and interstitial fluid pressure in tumors during progres- sion: implications for vascular collapse, Cancer research 73 (13) (2013) 3833–3841. 29

    2013

  7. [7]

    R. K. Jain, J. D. Martin, T. Stylianopoulos, The role of mechanical forces in tumor growth and therapy, Annual review of biomedical engineering 16 (1) (2014) 321–346

    2014

  8. [8]

    Mascheroni, C

    P. Mascheroni, C. Stigliano, M. Carfagna, D. P. Boso, L. Preziosi, P. Decuzzi, B. A. Schrefler, Predicting the growth of glioblastoma multiforme spheroids using a multiphase porous media model, Biomechanics and modeling in mechanobiology 15 (5) (2016) 1215–1228

    2016

  9. [9]

    K. R. Swanson, E. C. Alvord Jr, J. D. Murray, A quantitative model for differential motility of gliomas in grey and white matter, Cell proliferation 33 (5) (2000) 317– 329

    2000

  10. [10]

    B. Tunç, D. A. Hormuth II, G. Biros, T. E. Yankeelov, Modeling of glioma growth with mass effect by longitudinal magnetic resonance imaging, IEEE Transactions on Biomedical Engineering 68 (12) (2021) 3713–3724

    2021

  11. [12]

    Hogea, C

    C. Hogea, C. Davatzikos, G. Biros, Modeling glioma growth and mass effect in 3d mr images of the brain, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2007, pp. 642–650

    2007

  12. [13]

    E. Lima, J. Oden, D. Hormuth, T. Yankeelov, R. Almeida, Selection, calibration, and validation of models of tumor growth, Mathematical Models and Methods in Applied Sciences 26 (12) (2016) 2341–2368

    2016

  13. [14]

    Faghihi, X

    D. Faghihi, X. Feng, E. A. Lima, J. T. Oden, T. E. Yankeelov, A coupled mass transport and deformation theory of multi-constituent tumor growth, Journal of the Mechanics and Physics of Solids 139 (2020) 103936

    2020

  14. [15]

    K. C. Wong, R. Summers, E. Kebebew, J. Yao, Tumor growth prediction with hy- perelastic biomechanical model, physiological data fusion, and nonlinear optimiza- tion, in: International Conference on Medical Image Computing and Computer- Assisted Intervention, Springer, 2014, pp. 25–32

    2014

  15. [16]

    D.A.HormuthII,J.A.Weis, S.L.Barnes, M.I.Miga, E.C.Rericha, V.Quaranta, T. E. Yankeelov, Predicting in vivo glioma growth with the reaction diffusion equation constrained by quantitative magnetic resonance imaging data, Physical biology 12 (4) (2015) 046006. 30

    2015

  16. [17]

    D. A. Hormuth, J. A. Weis, S. L. Barnes, M. I. Miga, E. C. Rericha, V. Quaranta, T. E. Yankeelov, A mechanically coupled reaction–diffusion model that incorpo- rates intra-tumoural heterogeneity to predict in vivo glioma growth, Journal of The Royal Society Interface 14 (128) (2017) 20161010

    2017

  17. [18]

    Liang, J

    B. Liang, J. Tan, L. Lozenski, D. A. Hormuth, T. E. Yankeelov, U. Villa, D. Faghihi, Bayesian inference of tissue heterogeneity for individualized prediction ofgliomagrowth, IEEEtransactionsonmedicalimaging42(10)(2023)2865–2875

    2023

  18. [19]

    D. A. I. Hormuth, Predicting the spatio-temporal evolution of tumor growth and treatment response in a murine model of glioma, Ph.D. thesis, Vanderbilt Univer- sity (2016)

    2016

  19. [20]

    D. A. Hormuth II, J. T. Skinner, M. D. Does, T. E. Yankeelov, A comparison of individual and population-derived vascular input functions for quantitative DCE- MRI in rats, Magnetic resonance imaging 32 (4) (2014) 397–401

    2014

  20. [21]

    A. R. Padhani, G. Liu, D. Mu-Koh, T. L. Chenevert, H. C. Thoeny, T. Takahara, A.Dzik-Jurasz, B.D.Ross, M.VanCauteren, D.Collins, etal., Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommenda- tions, Neoplasia 11 (2) (2009) 102–125

    2009

  21. [22]

    Sugahara, Y

    T. Sugahara, Y. Korogi, M. Kochi, I. Ikushima, Y. Shigematu, T. Hirai, T. Okuda, L. Liang, Y. Ge, Y. Komohara, et al., Usefulness of diffusion-weighted mri with echo-planar technique in the evaluation of cellularity in gliomas, Journal of Mag- netic Resonance Imaging: An Official Journal of the International Society for Magnetic Resonance in Medicine 9 (1...

    1999

  22. [23]

    Anderson, J

    A. Anderson, J. Xie, J. Pizzonia, R. Bronen, D. Spencer, J. Gore, Effects of cell volume fraction changes on apparent diffusion in human cells, Magnetic resonance imaging 18 (6) (2000) 689–695

    2000

  23. [24]

    S. L. Barnes, A. G. Sorace, M. E. Loveless, J. G. Whisenant, T. E. Yankeelov, Correlation of tumor characteristics derived from DCE-MRI and DW-MRI with histology in murine models of breast cancer, NMR in Biomedicine 28 (10) (2015) 1345–1356.doi:10.1002/nbm.3377

    work page doi:10.1002/nbm.3377 2015

  24. [25]

    N. C. Atuegwu, L. R. Arlinghaus, X. Li, E. B. Welch, B. A. Chakravarthy, J. C. Gore, T. E. Yankeelov, Integration of diffusion-weighted mri data and a sim- ple mathematical model to predict breast tumor cellularity during neoadjuvant chemotherapy, Magnetic resonance in medicine 66 (6) (2011) 1689–1696

    2011

  25. [26]

    Roininen, J

    L. Roininen, J. M. Huttunen, S. Lasanen, Whittle-matérn priors for bayesian statistical inversion with applications in electrical impedance tomography., Inverse Problems & Imaging 8 (2) (2014). 31

    2014

  26. [27]

    J. Tan, D. Faghihi, A scalable framework for multi-objective pde-constrained de- sign of building insulation under uncertainty, Computer Methods in Applied Me- chanics and Engineering 419 (2024) 116628

    2024

  27. [28]

    P. K. Singh, D. Faghihi, Chance-constrained optimal design of porous thermal insulation systems under spatially correlated uncertainty, Structural and Multi- disciplinary Optimization 68 (9) (2025) 178

    2025

  28. [29]

    Y. Daon, G. Stadler, Mitigating the influence of the boundary on pde-based co- variance operators, arXiv preprint arXiv:1610.05280 (2016)

    Pith/arXiv arXiv 2016

  29. [30]

    Alghamdi, M

    A. Alghamdi, M. A. Hesse, J. Chen, U. Villa, O. Ghattas, Bayesian poroelastic aquifer characterization from insar surface deformation data. 2. quantifying the uncertainty, Water Resources Research 57 (11) (2021) e2021WR029775

    2021

  30. [31]

    Villa, N

    U. Villa, N. Petra, O. Ghattas, HIPPYlib: An Extensible Software Framework for Large-Scale Inverse Problems Governed by PDEs: Part I: Deterministic Inversion and Linearized Bayesian Inference, ACM Trans. Math. Softw. 47 (2) (Apr. 2021). doi:10.1145/3428447. URLhttps://doi.org/10.1145/3428447

    doi:10.1145/3428447 2021

  31. [32]

    Budday, P

    S. Budday, P. Steinmann, E. Kuhl, Physical biology of human brain development, Frontiers in cellular neuroscience 9 (2015) 257

    2015

  32. [33]

    L. A. Mihai, S. Budday, G. A. Holzapfel, E. Kuhl, A. Goriely, A family of hyper- elastic models for human brain tissue, Journal of the Mechanics and Physics of Solids 106 (2017) 60–79

    2017

  33. [34]

    Isaac, N

    T. Isaac, N. Petra, G. Stadler, O. Ghattas, Scalable and efficient algorithms for the propagation of uncertainty from data through inference to prediction for large- scale problems, with application to flow of the antarctic ice sheet, Journal of Computational Physics 296 (2015) 348–368

    2015

  34. [35]

    K.-T. Kim, U. Villa, M. Parno, Y. Marzouk, O. Ghattas, N. Petra, hippylib-muq: A bayesian inference software framework for integration of data with complex pre- dictive models under uncertainty, ACM Transactions on Mathematical Software 49 (2) (2023) 1–31

    2023

  35. [36]

    S. L. Cotter, G. O. Roberts, A. M. Stuart, D. White, Mcmc methods for functions: modifying old algorithms to make them faster, Statistical Science (2013) 424–446

    2013

  36. [37]

    F. J. Pinski, G. Simpson, A. M. Stuart, H. Weber, Algorithms for kullback–leibler approximation of probability measures in infinite dimensions, SIAM Journal on Scientific Computing 37 (6) (2015) A2733–A2757. 32

    2015

  37. [38]

    Alnæs, J

    M. Alnæs, J. Blechta, J. Hake, A. Johansson, B. Kehlet, A. Logg, C. Richardson, J. Ring, M. E. Rognes, G. N. Wells, The fenics project version 1.5, Archive of numerical software 3 (100) (2015)

    2015

  38. [39]

    Villa, N

    U. Villa, N. Petra, O. Ghattas, hIPPYlib: an Extensible Software Framework for Large-scale Deterministic and Bayesian Inverse Problems, Journal of Open Source Software 3 (30) (2018).doi:10.21105/joss.00940

    doi:10.21105/joss.00940 2018

  39. [40]

    Villa, N

    U. Villa, N. Petra, O. Ghattas, hIPPYlib: an Extensible Software Framework for Large-scale Deterministic and Bayesian Inverse Problems (2016).doi:10.5281/ zenodo.596931. URLhttp://hippylib.github.io

    2016

  40. [41]

    P. K. Singh, K. A. Farrell-Maupin, D. Faghihi, A framework for strategic discovery ofcredibleneuralnetworksurrogatemodelsunderuncertainty, ComputerMethods in Applied Mechanics and Engineering 427 (2024) 117061

    2024

  41. [42]

    M. A. Islam, D. Deighan, S. Bhattacharjee, D. Tantalo, P. K. Singh, D. Salac, D. Faghihi, Stochastic deep learning surrogate models for uncertainty propagation in microstructure–properties of ceramic aerogels, Computational Materials Science 258 (2025) 114035

    2025

  42. [43]

    Chaudhuri, G

    A. Chaudhuri, G. Pash, D. A. Hormuth, G. Lorenzo, M. Kapteyn, C. Wu, E. A. Lima, T. E. Yankeelov, K. Willcox, Predictive digital twin for optimizing patient- specific radiotherapy regimens under uncertainty in high-grade gliomas, Frontiers in Artificial Intelligence 6 (2023) 1222612

    2023

  43. [44]

    G. Pash, U. Villa, D. A. Hormuth II, T. E. Yankeelov, K. Willcox, Predictive digital twins with quantified uncertainty for patient-specific decision making in oncology, Journal of Computational Physics (2026) 114937

    2026

  44. [45]

    M. G. Kapteyn, A. Chaudhuri, E. A. Lima, G. Pash, R. Bravo, K. E. Willcox, T. E. Yankeelov, D. A. Hormuth II, Tumortwin: a python framework for patient- specific digital twins in oncology, BMC Medical Informatics and Decision Making (2026). 33

    2026