Pith. sign in

REVIEW 2 major objections 5 minor 58 references

Physics-Informed Neural Networks for the High-Resolution Reconstruction of Flow Measurement Indicators in Fluid Dynamics

T0 review · 2 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Pith's one-line read Physics-informed neural networks turn sparse, noisy blood-flow velocity measurements into high-resolution velocity, pressure and wall-shear-stress fields that match ground truth better than pure CFD or pure data fitting.

desk verdict Solid engineering of a PINN pipeline for sparse flow data; the quasi-steady residual is a real but already-flagged soft spot that does not erase the FDA gains or the transparent ablations. read the letter →

arxiv 2607.11576 v1 pith:XXAW7GBZ submitted 2026-07-13 math.NA cs.NA

classification math.NAcs.NA MSC 65M3276Z0568T07
keywords PINNs4DflowMRIComputationalhemodynamicsScientificmachinelearningWallshearstressNavier-Stokesreconstruction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Clinical imaging methods such as 4D flow MRI and particle-image velocimetry give incomplete or noisy pictures of blood velocity, especially near vessel walls. Those gaps make derived quantities such as wall shear stress unreliable for diagnosis. This paper shows that a neural network trained simultaneously on the measured velocities and on the residual of the incompressible Navier–Stokes equations can fill in the missing information. On both a standard nozzle benchmark and a patient-derived aneurysm model the physics-regularized reconstructions recover pressure (never measured) and produce smoother, more accurate wall-shear-stress maps than either a pure CFD simulation or a network trained only on the data. The result is a mesh-free post-processing tool that upgrades under-resolved clinical flow measurements into spatially resolved hemodynamic indicators.

What carries the argument

The composite PINN loss: mean-squared and mean-squared-logarithmic residuals of mass and momentum conservation, boundary-condition residuals, plus data fidelity terms on velocity magnitude and direction; pressure is recovered a posteriori from the momentum residual without ever being supplied as training data.

What would settle it

Train the same PINN architecture on a fully time-resolved, high-resolution ground-truth CFD solution of the aneurysm (including the unsteady term) and check whether the reconstructed wall-shear-stress and enstrophy time series still match the ground truth within the error levels reported for the quasi-steady runs.

Watch

Extended reading notes

Core claim

Embedding the steady incompressible Navier–Stokes residuals directly into the training loss of a neural network allows sparse experimental velocity samples (PIV or 4D flow MRI) to be reconstructed into high-resolution velocity, pressure and wall-shear-stress fields that agree more closely with ground-truth observations than either classical CFD or pure data-driven fitting.

Load-bearing premise

Treating the Navier–Stokes equations as quasi-steady (no time derivative) still yields accurate reconstructions for pulsatile aneurysm flow when only a few discrete time snapshots are given to the network.

Editorial extensions

If this is right

  • Sparse 4D flow MRI acquisitions can be upgraded to clinically usable wall-shear-stress and pressure maps without additional imaging time.
  • Pressure, never measured by phase-contrast MRI, becomes available as a free by-product of the physics residual.
  • Mesh generation and geometry segmentation steps required by classical CFD can be skipped for post-processing of clinical velocity data.
  • The same loss construction can be reused on other under-resolved experimental modalities (PIV, Doppler ultrasound) once fluid properties and approximate boundary locations are known.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the quasi-steady approximation holds for moderate pulsatility, the method could be applied frame-by-frame to full cardiac-cycle 4D flow MRI without enlarging the network architecture.
  • Automatic balancing of the multi-term loss (rather than manual weight selection) would be the next practical step before routine clinical use.
  • Extending the residual to include a simple turbulence model or the unsteady term would test whether the same framework remains competitive in transitional or highly unsteady regimes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript proposes a PINN framework that embeds the incompressible Navier–Stokes residuals (Eqs. 1–3, 6–7) together with sparse velocity measurements into a composite loss (Eq. 5) to reconstruct high-resolution velocity, pressure and wall-shear-stress fields from under-resolved experimental data. Two benchmarks are examined: the laminar FDA nozzle (Re = 500), trained and tested against both FEM solutions and PIV measurements (FDA-CFD 1–7, FDA-EXP 1–3), and a patient-derived aneurysm model (Re ≈ 1500) trained on FEM snapshots and on in-vitro 4D-flow MRI (AN-CFD, AN-4DMRI weight sweeps). Systematic ablations of loss weights, MSLE terms, unit-vector losses and collocation density are reported, together with quantitative error metrics (Ez, Eu, kinetic energy, enstrophy) and visual comparisons of velocity, pressure, vorticity and WSS. The central claim is that the data–physics synergy yields reconstructions closer to ground truth than pure CFD or pure data-driven baselines.

Significance. If the claimed superiority holds, the work supplies a practical, mesh-free post-processing pipeline that can recover pressure and wall shear stress from sparse 4D-flow MRI or PIV without full CFD personalization. The multi-configuration ablations (Tables 2–5), explicit residual definitions, and dual validation against both FEM and experimental data constitute a reproducible demonstration of PINN utility for hemodynamic indicators. The FDA-nozzle results in particular are carefully controlled and already of interest to the validation community. The aneurysm application, while more ambitious, remains limited by the quasi-steady residual assumption that the authors themselves flag.

major comments (2)
  1. [§3.2.1, §3.2.2, §4.1] §3.2.1 and §3.2.2 (and Limitations §4.1): for the aneurysm the physics residual L_PDE is the steady form of the Navier–Stokes equations (Eqs. 1a–b, 2a–b) evaluated at isolated fixed times (t = 2.024 s for FEM; discrete snapshots for 4D-flow). The unsteady term ∂u/∂t is therefore omitted even though the inflow is pulsatile (Fig. 13b) and Re ≈ 1500. No residual-magnitude comparison or full-unsteady PINN control is supplied. Because the abstract and §4 assert superiority over CFD and pure data-driven methods for this more complex case, the quasi-steady assumption is load-bearing; either a quantitative check of the neglected acceleration term or a clear restriction of the superiority claim to the steady FDA nozzle is required.
  2. [Abstract, §4, Table 5] Abstract and §4 claim that PINN results “align more closely with ground truth … than standard CFD or pure data-driven approaches.” For the aneurysm the only pure-data control is AN-4DMRI-4 (Table 5, PDE weights set to zero). No corresponding pure-data ablation is reported for the FDA-EXP series, and the CFD baselines are themselves subject to modeling choices (VMS-LES, mesh, boundary conditions). A single, consistently defined pure-data baseline for both benchmarks would make the superiority statement falsifiable.
minor comments (5)
  1. [Table 1] Table 1 lists three different characteristic diameters and viscosities; a short sentence clarifying which length scale is used for the aneurysm Re would avoid ambiguity.
  2. [Eq. (11), Figs. 8d, 11c, 23] Eq. (11) defines WSS; the subsequent wall-averaged plots (Figs. 8d, 11c, 23) would be clearer if the averaging surface and the precise post-processing library (nisaba vs. Paraview) were stated once in the methods.
  3. [Figs. 7, 22] Several figures (e.g., Fig. 7 residual maps, Fig. 22 vorticity) use log-scale color bars without explicit units or reference values; adding a common color-bar range across panels would aid comparison.
  4. [Eqs. (6), (9)] The MSLE terms in Eqs. (6) and (9) are introduced without a reference or a short derivation of the +1 offset; a one-sentence justification would help readers unfamiliar with the device.
  5. [Graphical abstract, figure captions] Typographical inconsistencies appear in the graphical abstract (“Weusephysics-informed…”) and in a few figure captions; a final proof-reading pass is needed.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circularity; PINN reconstructions are validated against held-out sections and independent external PIV/4D-flow-MRI data, with only a minor non-forcing self-citation of the authors' CFD solver for in-silico verification data.

  1. self citation load bearing [§3.1.1 Setup and §3.2.1 Setup (lifeX citations [41,42])]
    "We perform the simulations using lifeX [41, 42], a high-performance solver of multiphysics and multiscale differential models developed at the MOX laboratory of Politecnico di Milano. ... We carry out the FEM simulations using lifeX [41]."

    lifeX is the authors' own code; it supplies the in-silico velocity/pressure fields used for the FDA-CFD and ANE-CFD verification cases. This is ordinary self-citation for data generation and is not load-bearing: the paper's central claims rest on the experimental PIV and 4D-flow-MRI validations that are independent of lifeX.

full rationale

The paper's derivation chain is the standard PINN construction: residual of the (quasi-steady) incompressible Navier-Stokes equations (Eqs. 1-3, 6-7) plus data-fidelity terms (Eqs. 8-9) minimized over network parameters. The residuals are the classical continuum equations, not defined in terms of the reconstructed fields or fitted weights. Training uses sparse velocity samples on selected cross-sections; testing and comparison use held-out sections plus independent FEM solutions and external experimental PIV (FDA) and 4D-flow-MRI (aneurysm) acquisitions that were never used to define the architecture or loss. Pressure and WSS are recovered a posteriori from the trained velocity via the momentum residual and Eq. 11; they are not fitted inputs re-labeled as predictions. The sole self-citation of note is lifeX (authors' own FEM library) used solely to generate the in-silico verification datasets of Sections 3.1.1 and 3.2.1; the central empirical claim of superiority over pure CFD or pure data-driven baselines is supported by the external experimental comparisons (Figs. 10-12, 20-24, Tables 3-6) and does not reduce to that citation. Hyper-parameter weights are chosen by trial-and-error and reported; they do not force the reported accuracy gains by construction. The quasi-steady modeling choice is a limitation (explicitly listed in §4.1), not a circularity. Hence the score is 1 (minor non-load-bearing self-citation) rather than 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the classical incompressible Newtonian Navier-Stokes model, a quasi-steady residual evaluation, and a large set of hand-chosen loss weights and network sizes. No new physical entities are postulated; the free parameters are purely algorithmic.

free parameters (3)
  • loss weights λ_u, λ_PDE,j, λ_BC,j, λ_û, λ_p̄, λ_MSLE
    Dozens of scalar multipliers (Tables 2, 4, 5) are chosen by trial-and-error to balance data fidelity against physics residuals; different choices reverse the ranking of networks.
  • number of neurons per layer (16 vs 32) and PDE collocation density
    Architecture size and local refinement of collocation points (N_PDE = 3093 vs 4341) are selected manually and affect pressure and velocity accuracy (Tests 4–5).
  • MSLE inclusion flag and unit-vector loss weight
    Whether to add logarithmic residual terms and the velocity-direction loss is decided empirically; both alter low-magnitude reconstruction quality.
assumptions (3)
  • domain assumption Blood obeys the steady incompressible Newtonian Navier-Stokes equations with the given Re, density and viscosity (Eqs. 1–2, Table 1).
    Standard continuum model for large-vessel hemodynamics; invoked throughout the residual definitions.
  • ad hoc to paper Time derivatives may be omitted from the momentum residual when training on discrete 4D-flow snapshots (quasi-steady assumption, §3.2.1).
    Reduces input dimension; validity for systolic peaks is unproven and listed as a limitation.
  • domain assumption No-slip wall residual plus zero-mean pressure penalty suffice to recover unique pressure when Neumann data are unavailable.
    Used for the aneurysm case (Eq. 14); classical for closed domains but sensitive to noisy wall data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Informed Neural Networks for the High-Resolution Reconstruction of Flow Measurement Indicators in Fluid Dynamics." pith.science (2026). https://pith.science/paper/XXAW7GBZ

@misc{pith2026260711576,
  author       = {Pith},
  title        = {Pith review of: Physics-Informed Neural Networks for the High-Resolution Reconstruction of Flow Measurement Indicators in Fluid Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XXAW7GBZ}},
  note         = {Machine review of arXiv:2607.11576}
}
read the original abstract

Accurate, spatially resolved flow field measurements are essential for the reliable assessment of hemodynamic quantities in cardiovascular research and clinical practice. Experimental techniques, such as 4D flow MRI, PIV, or Doppler ultrasound, often yield data that are sparse, noisy, or under-resolved, particularly near vessel walls and in regions of complex flow. This limits the fidelity of distributed or derived hemodynamic indicators such as the wall shear stress and the clinical utility of such measurements. To address these challenges, we propose a physics-informed neural network (PINN) framework that integrates the incompressible Navier-Stokes equations with velocity measurements coming from experimental flow field data. By embedding physical laws into data, PINN enhances the reconstruction of velocity fields, enables the estimation of unmeasured quantities such as pressure and wall shear stress, and improves the spatial resolution of hemodynamic indicators. We show the effectiveness of our approach using both in silico and experimental data. First, we apply our method to the FDA nozzle benchmark, leveraging both control particle image velocimetry (PIV) measurements and computational fluid dynamics (CFD) simulations. Next, we apply our method to the more complex case of blood flow in an aneurysm model, exploiting in vitro 4D flow MRI data. In both cases, the synergy between data-driven learning and physics-based regularization yields results that align more closely with ground truth observations than standard CFD or pure data-driven approaches. Our findings highlight the potential of PINNs to improve the fidelity of under-resolved flow field measurements and yield spatially resolved hemodynamic indicators.

Figures

Figures reproduced from arXiv: 2607.11576 by the authors.

Figure 1
Figure 1. Graphical abstract. We use physics-informed neural networks (PINNs) to integrate physical knowledge through the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. First, in Section 3.1.1, we verify the proposed PINN method by performing a CFD simulation [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Collocation points for the FDA nozzle benchmark. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (21 more)
Figure 4
Figure 4. Figure 4: Test 1: Comparison of approximation properties of FDA-CFD-1 and FDA-CFD-2 for low magnitude quantities and [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Test 2: Comparison between FDA-CFD-2 and FDA-CFD-3 setups. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Tests 3 and 4: Comparison of computed quantities for the FEM simulation (blue) and the PINNs. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Test 4: Pointwise comparison between FDA-CFD-4 and FDA-CFD-5. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Test 5: Comparison of computed quantities for the FEM simulation (blue) and the PINNs. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: PIV measurements from the 999 dataset [32]. (a) Axial velocity (u)z profile along the y-axis at different z values. (b) Radial velocity (u)x and (u)y along the y-axis at different z values. (c) Training and testing data sections partition. Test sections are: 1 at z1 = …
Figure 10
Figure 10. Figure 10: Test 6: Comparison of the PINN results with the [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Test 6: Comparison of errors computed from the PINNs with the ones computed using the experimental data on [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Test 5: Comparison of errors with respect to the experimental data computed from the PINNs with the ones [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Aneurysm geometry and simulation features. [PITH_FULL_IMAGE:figures/full_fig_p017_13.png]
Figure 14
Figure 14. Figure 14: Collocation points. (a) Training and testing dataset partition. (b) Collocation points for the PDE regularization. where Lp¯ represents the zero mean constraint penalty on the pressure: Lp¯(θ˜) = λp¯ NPDE [PITH_FULL_IMAGE:figures/full_fig_p018_14.png]
Figure 15
Figure 15. Figure 15: Computed magnitude of the velocity field. [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Reconstructed pressure field. (a) FEM. (b) AN-CFD-1. (c) AN-CFD-2. (d) AN-CFD-3. (a) (b) (c) [PITH_FULL_IMAGE:figures/full_fig_p020_16.png]
Figure 17
Figure 17. Figure 17: Normalized error Eu computed on different sections. (a) AN-CFD-1. (b) AN-CFD-2. (c) AN-CFD-3. to the aneurysm domain. The 4D flow MRI dataset is very noisy and dispersed at the boundary. The measurement points cross the lateral boundaries, while at the bottom of the a…
Figure 18
Figure 18. Figure 18: Collocation points and velocity distribution for the [PITH_FULL_IMAGE:figures/full_fig_p021_18.png]
Figure 19
Figure 19. Figure 19: Collocation points. (a) Collocation points at wall. (b) PDE collocation points. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_19.png]
Figure 20
Figure 20. Figure 20: Velocity field evaluated at the collocation points of the [PITH_FULL_IMAGE:figures/full_fig_p022_20.png]
Figure 21
Figure 21. Figure 21: Pressure field reconstruction. (a) AN-4DMRI-1. (b) AN-4DMRI-2. (c) AN-4DMRI-3. (a) (b) (c) (d) (e) (f) [PITH_FULL_IMAGE:figures/full_fig_p022_21.png]
Figure 22
Figure 22. Figure 22: Computed vorticity field for the 4D flow MRI data and for the PINN outputs. Recovered vorticity paths are [PITH_FULL_IMAGE:figures/full_fig_p022_22.png]
Figure 23
Figure 23. Figure 23: Computed wall shear stress magnitude in log scale. [PITH_FULL_IMAGE:figures/full_fig_p024_23.png]
Figure 24
Figure 24. Figure 24: Integral quantities during a single heartbeat. [PITH_FULL_IMAGE:figures/full_fig_p024_24.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 9 canonical work pages

  1. [1]

    Stankovic, B

    Z. Stankovic, B. D. Allen, J. Garcia, K. B. Jarvis, M. Markl, 4D flow imaging with MRI, Cardiovascular Diagnosis and Therapy 4 (2014) 173

  2. [2]

    Miyazaki, K

    S. Miyazaki, K. Itatani, T. Furusawa, T. Nishino, M. Sugiyama, Y. Takehara, S. Yasukochi, Validation of numerical simulation methods in aortic arch using 4D Flow MRI, Heart and Vessels 32 (2017) 1032–1044

  3. [3]

    Sughimoto, Y

    K. Sughimoto, Y. Shimamura, C. Tezuka, K. Tsubota, H. Liu, K. Okumura, Y. Masuda, H. Haneishi, Effects of arterial blood flow on walls of the abdominal aorta: distributions of wall shear stress and oscillatory shear index determined by phase-contrast magnetic resonance imaging, Heart and Vessels 31 (2016) 1168–1175

  4. [4]

    Markl, A

    M. Markl, A. Frydrychowicz, S. Kozerke, M. Hope, O. Wieben, 4D flow MRI, Journal of Magnetic Resonance Imaging 36 (2012) 1015–1036

  5. [5]

    Casas, J

    B. Casas, J. Lantz, P. Dyverfeldt, T. Ebbers, 4D flow MRI-based pressure loss estimation in stenotic flows: Evaluation using numerical simulations, Magnetic Resonance in medicine 75 (2016) 1808–1821

  6. [6]

    H. Baek, M. V. Jayaraman, G. E. Karniadakis, Wall shear stress and pressure distribution on aneurysms and infundibulae in the posterior communicating artery bifurcation, Annals of Biomedical Engineering 37 (2009) 2469–2487

  7. [7]

    C. G. Caro, Discovery of the role of wall shear in atherosclerosis, Arteriosclerosis, Thrombosis, and Vascular Biology 29 (2009) 158–161

  8. [8]

    Zingaro, L

    A. Zingaro, L. Dedè, F. Menghini, A. Quarteroni, Hemodynamics of the heart’s left atrium based on a Variational Multiscale-LES numerical method, European Journal of Mechanics-B/Fluids 89 (2021) 380–400

Show all 58 references
  1. [9]

    A. Harloff, et al., 3D blood flow characteristics in the carotid artery bifurcation assessed by flow- sensitive 4D MRI at 3T, Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine 61 (2009) 65–74

  2. [10]

    M. T. Ngo, C. I. Kim, J. Jung, G. H. Chung, D. H. Lee, H. S. Kwak, Four-dimensional flow magnetic resonance imaging for assessment of velocity magnitudes and flow patterns in the human carotid artery bifurcation: Comparison with computational fluid dynamics, Diagnostics 9 (2019) 223

  3. [11]

    M. S. Pravdivtseva, F. Gaidzik, P. Berg, P. Ulloa, N. Larsen, O. Jansen, J.-B. Hövener, M. Salehi Ravesh, Influence of spatial resolution and compressed sense acceleration factor on flow quantification with 4D Flow MRI at 3 Tesla, Tomography 8 (2022) 457–478

  4. [12]

    Zingaro, I

    A. Zingaro, I. Fumagalli, L. Dede’, M. Fedele, P. C. Africa, A. F. Corno, A. Quarteroni, A geometric multiscale model for the numerical simulation of blood flow in the human left heart, Discrete and Continuous Dynamical Systems - S 15 (2022) 2391–2427. doi:10.3934/dcdss.2022052

  5. [13]

    Futami, T

    K. Futami, T. Uno, K. Misaki, S. Tamai, I. Nambu, N. Uchiyama, M. Nakada, Identification of vortex cores in cerebral aneurysms on 4D flow MRI, American Journal of Neuroradiology 40 (2019) 2111–2116

  6. [14]

    T. B. Le, D. R. Troolin, D. Amatya, E. K. Longmire, F. Sotiropoulos, Vortex phenomena in sidewall aneurysm hemodynamics: experiment and numerical simulation, Annals of Biomedical Engineering 41 (2013) 2157–2170

  7. [15]

    Quarteroni, Numerical Models for Differential Problems, volume 16 ofModeling, Simulation and Applications, 3rd ed., Springer International Publishing, Cham, Switzerland, 2017

    A. Quarteroni, Numerical Models for Differential Problems, volume 16 ofModeling, Simulation and Applications, 3rd ed., Springer International Publishing, Cham, Switzerland, 2017. 27

  8. [16]

    Totorean, I.-C

    A.-F. Totorean, I.-C. Totorean, S. I. Bernad, T. Ciocan, D. C. Malita, D. Gaita, E. S. Bernad, Patient- Specific Image-Based Computational Fluid Dynamics Analysis of Abdominal Aorta and Branches, Jour- nal of Personalized Medicine 12 (2022) 1502. URL:https://www.mdpi.com/2075-...

  9. [17]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, MIT press, 2016

  10. [18]

    Tassi, A

    T. Tassi, A. Zingaro, L. Dede’, A machine learning approach to enhance the SUPG stabilization method for advection-dominated differential problems, Mathematics in Engineering 5 (2023) 1–26. doi:10.3934/mine.2023032

  11. [19]

    Quarteroni, P

    A. Quarteroni, P. Gervasio, F. Regazzoni, Combining physics-based and data-driven models: ad- vancing the frontiers of research with scientific machine learning, Mathematical Models and Methods in Applied Sciences 35 (2025) 905–1071. URL:https://www.worldscientific.com/doi/10....

  12. [20]

    J. D. Toscano, V. Oommen, A. J. Varghese, Z. Zou, N. Ahmadi Daryakenari, C. Wu, G. E. Karniadakis, From PINNs to PIKANs: recent advances in physics-informed machine learning, Machine Learning for Computational Science and Engineering 1 (2025) 15. URL:https://link.springer.com/...

  13. [21]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations, arXiv preprint arXiv:1711.10561 (2017)

  14. [22]

    N. K. Schiavone, P. J. Nair, C. J. Elkins, D. B. McElhinney, D. B. Ennis, J. K. Eaton, A. L. Marsden, Assessing the impact of cardiac output and valve orientation on bioprosthetic pulmonary valve hemodynamics using in vitro 4d-flow mri and high-speed imaging, Cardiovascular En...

  15. [23]

    Zhang, M

    J. Zhang, M. C. Brindise, S. Rothenberger, S. Schnell, M. Markl, D. Saloner, V. L. Rayz, P. P. Vlachos, 4D flow MRI pressure estimation using velocity measurement-error-based weighted least-squares, IEEE transactions on medical imaging 39 (2019) 1668–1680

  16. [24]

    Ferdian, D

    E. Ferdian, D. J. Dubowitz, C. A. Mauger, A. Wang, A. A. Young, WSSNet: aortic wall shear stress estimation using deep learning on 4D flow MRI, Frontiers in Cardiovascular Medicine (2022) 1969

  17. [25]

    Zhang, S

    J. Zhang, S. M. Rothenberger, M. C. Brindise, M. B. Scott, H. Berhane, J. J. Baraboo, M. Markl, V. L. Rayz, P. P. Vlachos, Divergence-Free Constrained Phase Unwrapping and Denoising for 4D Flow MRI Using Weighted Least-Squares, IEEE Transactions on Medical Imaging 40 (2021) 33...

  18. [26]

    S. Cai, Z. Mao, Z. Wang, M. Yin, G. E. Karniadakis, Physics-informed neural networks (PINNs) for fluid mechanics: A review, Acta Mechanica Sinica (2022) 1–12

  19. [27]

    Kissas, Y

    G. Kissas, Y. Yang, E. Hwuang, W. R. Witschey, J. A. Detre, P. Perdikaris, Machine learning in cardiovascular flows modeling: Predicting arterial blood pressure from non-invasive 4D flow MRI data using physics-informed neural networks, Computer Methods in Applied Mechanics and...

  20. [28]

    Sarabian, H

    M. Sarabian, H. Babaee, K. Laksari, Physics-informed neural networks for brain hemodynamic predic- tions using medical imaging, IEEE Transactions on Medical Imaging (2022)

  21. [29]

    URL:https://www.cambridge.org/core/ product/identifier/S002211202100135X/type/journal_article

    S.Cai, Z.Wang, F.Fuest, Y.J.Jeon, C.Gray, G.E.Karniadakis, Flowoveranespressocup: inferring3- DvelocityandpressurefieldsfromtomographicbackgroundorientedSchlierenviaphysics-informedneu- ral networks, Journal of Fluid Mechanics 915 (2021) A102. URL:https://www.cambridge.org/cor...

  22. [30]

    Hub, Benchmark dataset for validating computational fluid dynamic (CFD) simulation of blood flow through FDA nozzle and FDA blood pump — NCI Hub, 2022

    N. Hub, Benchmark dataset for validating computational fluid dynamic (CFD) simulation of blood flow through FDA nozzle and FDA blood pump — NCI Hub, 2022. URL:https://ncihub.org/wiki/FDA_ CFD?version=30, [Online; accessed March 2022]

  23. [31]

    S. F. Stewart, E. G. Paterson, G. W. Burgreen, P. Hariharan, M. Giarra, V. Reddy, S. W. Day, K. B. Manning, S. Deutsch, M. R. Berman, et al., Assessment of CFD performance in simulations of an idealized medical device: results of FDA’s first computational interlaboratory study...

  24. [32]

    Hariharan, M

    P. Hariharan, M. Giarra, V. Reddy, S. W. Day, K. B. Manning, S. Deutsch, S. F. Stewart, M. R. Myers, M. R. Berman, G. W. Burgreen, et al., Multilaboratory particle image velocimetry analysis of the fda benchmark nozzle model to support validation of computational fluid dynamic...

  25. [33]

    Hariharan, R

    P. Hariharan, R. A. Malinauskas, Round robin 1 data sets, 2017. URL:https://ncihub.org/ publications/43/2. doi:doi:/10.17917/C78G69

  26. [34]

    Quarteroni, L

    A. Quarteroni, L. Dede’, A. Manzoni, C. Vergara, et al., Mathematical modelling of the human cardio- vascular system: data, numerical approximation, clinical applications, volume 33, Cambridge University Press, 2019

  27. [35]

    Bischof, M

    R. Bischof, M. A. Kraus, Multi-Objective Loss Balancing for Physics-Informed Deep Learning, Com- puter Methods in Applied Mechanics and Engineering 439 (2025) 117914. URL:https://linkinghub. elsevier.com/retrieve/pii/S0045782525001860. doi:10.1016/j.cma.2025.117914

  28. [36]

    Kissas, E

    G. Kissas, E. Hwuang, E. W. Thompson, N. Schwartz, J. A. Detre, W. R. Witschey, P. Perdikaris, Feasibility of vascular parameter estimation for assessing hypertensive pregnancy disorders, Journal of Biomechanical Engineering 144 (2022) 121011

  29. [37]

    Chollet, Deep Learning with Python, Simon and Schuster, 2021

    F. Chollet, Deep Learning with Python, Simon and Schuster, 2021

  30. [38]

    Regazzoni, S

    F. Regazzoni, S. Pagani, A. Quarteroni, Universal Solution Manifold Networks (USM-Nets): non- intrusive mesh-free surrogate models for problems in variable domains, 2022. URL:https://arxiv. org/abs/2204.07805. doi:10.48550/ARXIV.2204.07805, version Number: 1

  31. [39]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, CoRR abs/1412.6980 (2015)

  32. [40]

    C. Zhu, R. H. Byrd, P. Lu, J. Nocedal, Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization, ACM Transactions on mathematical software (TOMS) 23 (1997) 550– 560

  33. [41]

    P. C. Africa, lifex: a flexible, high performance library for the numerical solution of complex finite element problems, SoftwareX 20 (2022) 101252

  34. [42]

    P. C. Africa, I. Fumagalli, M. Bucelli, A. Zingaro, M. Fedele, L. Dede’, A. Quarteroni, lifex- cfd: An open-source computational fluid dynamics solver for cardiovascular applications, Computer Physics Communications 296 (2024) 109039. URL:https://linkinghub.elsevier.com/retrie...

  35. [43]

    Arndt, W

    D. Arndt, W. Bangerth, D. Davydov, T. Heister, L. Heltai, M. Kronbichler, M. Maier, J.-P. Pelteret, B. Turcksin, D. Wells, The deal.II finite element library: Design, features, and insights, Comput- ers & Mathematics with Applications 81 (2021) 407–422. URL:https://linkinghub....

  36. [44]

    P. C. Africa, D. Arndt, W. Bangerth, B. Blais, M. Fehling, R. Gassmöller, T. Heister, L. Heltai, S. Kinnewig, M. Kronbichler, M. Maier, P. Munch, M. Schreter-Fleischhacker, J. P. Thiele, B. Tur- cksin, D. Wells, V. Yushutin, The deal.II library, Version 9.6, Journal of Numeric...

  37. [45]

    Forti, L

    D. Forti, L. Dedè, Semi-implicit BDF time discretization of the Navier–Stokes equations with VMS-LES modeling in a high performance computing framework, Computers & Fluids 117 (2015) 168–182

  38. [46]

    URL:https://mox.polimi.it/research-areas/hpcmox/ hardware/, [Accessed 13-Aug-2022]

    Hardware — mox.polimi.it, 2022. URL:https://mox.polimi.it/research-areas/hpcmox/ hardware/, [Accessed 13-Aug-2022]

  39. [47]

    D. V. Dung, N. D. Song, P. S. Palar, L. R. Zuhal, On The Choice of Activation Functions in Physics- Informed Neural Network for Solving Incompressible Fluid Flows, in: AIAA SCITECH 2023 Forum, American Institute of Aeronautics and Astronautics, National Harbor, MD & Online, 20...

  40. [48]

    Géron, Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems, O’Reilly Media, Inc., 2019

    A. Géron, Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems, O’Reilly Media, Inc., 2019

  41. [49]

    Ioffe, C

    S. Ioffe, C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift, in: International conference on machine learning, PMLR, 2015, pp. 448–456

  42. [50]

    Glorot, Y

    X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, in: Proceedings of the thirteenth international conference on artificial intelligence and statistics, JMLR Workshop and Conference Proceedings, 2010, pp. 249–256

  43. [51]

    Raschka, Python Machine Learning, Packt publishing ltd, 2015

    S. Raschka, Python Machine Learning, Packt publishing ltd, 2015

  44. [52]

    Abadi, P

    M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat, G. Irving, M. Isard, et al., TensorFlow: a system for Large-Scale machine learning, in: 12th USENIX symposium on operating systems design and implementation (OSDI 16), 2016, pp. 265–283

  45. [53]

    Chollet, et al., Keras,https://keras.io, 2015

    F. Chollet, et al., Keras,https://keras.io, 2015

  46. [54]

    Ahrens, B

    J. Ahrens, B. Geveci, C. Law, Paraview: An end-user tool for large data visualization, The visualization handbook 717 (2005)

  47. [55]

    M. S. Pravdivtseva, E. Peschke, T. Lindner, F. Wodarg, J. Hensler, D. Gabbert, I. Voges, P. Berg, A. J. Barker, O. Jansen, et al., 3D-printed, patient-specific intracranial aneurysm models: From clinical data to flow experiments with endovascular devices, Medical Physics 48 (2...

  48. [56]

    Weiss, V

    F.Pedregosa, G.Varoquaux, A.Gramfort, V.Michel, B.Thirion, O.Grisel, M.Blondel, P.Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, Édouard Duchesnay, Scikit-learn: Machine learning in python, Journal of Machine Learning Resear...

  49. [57]

    A. D. Jagtap, K. Kawaguchi, G. E. Karniadakis, Adaptive activation functions accelerate convergence in deep and physics-informed neural networks, Journal of Computational Physics 404 (2020) 109136

  50. [58]

    Henry, J

    E. Henry, J. Hofrichter, [8] singular value decomposition: Application to analysis of experimental data, in: Methods in enzymology, volume 210, Elsevier, 1992, pp. 129–192. 30

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.