Pith. sign in

REVIEW 2 major objections 1 minor 34 references

A simulation strategy that adds speckle decorrelation from real videos and iterative refinement produces training data enabling deep learning to outperform speckle tracking echocardiography on myocardial strain.

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-4.3

2026-06-29 09:27 UTC pith:5YDEZJ6N

load-bearing objection The new simulation pipeline mixing real speckle decorrelation with physics models is the actual contribution, but the claim that it produces training data realistic enough to beat clinical strain estimates rests on unshown evidence. the 2 major comments →

arxiv 2605.28697 v1 pith:5YDEZJ6N submitted 2026-05-27 eess.IV cs.AIcs.CV

Deep Learning Strain Estimation: Is Physics-Based Simulation the Solution?

Thierry Judge , Nicolas Duchateau , Andreas {\O}stvik , Khuram Faraz , Anders Austlid Task\'en , Sigve Karlsen , Thor Edvardsen , Harald Brunvand
show 8 more authors
Md Abulkalam Azad Havard Dalen Bj{\o}rnar Grenne Gabriel Kiss Pierre-Yves Courand Lasse Lovstakken Pierre-Marc Jodoin Olivier Bernard
This is my paper
classification eess.IV cs.AIcs.CV
keywords deep learningstrain estimationspeckle tracking echocardiographyphysics-based simulationmyocardial strainechocardiographymotion estimationsynthetic dataset
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 seeks to solve the shortage of trustworthy motion labels that limits deep learning for echocardiographic strain by creating more realistic physics-based simulations. It measures speckle decorrelation patterns directly from clinical videos and applies iterative refinement to make the synthetic motion sequences closer to real tissue behavior. These sequences form an open-source dataset of 1,478 videos that trains a motion estimation network. When tested, the network reports lower error on both global and regional strain than the clinical standard, with global longitudinal strain variability at 1.42 percent versus 1.78 percent for the reference method. Readers care because improved regional strain could support earlier detection of subtle heart abnormalities that global measures miss.

Core claim

The central claim is that incorporating speckle decorrelation measures from real videos together with an iterative refinement process yields physics-based simulations realistic enough to serve as motion references, allowing a deep learning model trained on the resulting 1,478-video dataset to estimate global and regional myocardial strain more accurately than speckle tracking echocardiography, including a global longitudinal strain inter-expert variability of 1.42 percent compared with 1.78 percent for the clinical reference.

What carries the argument

The novel simulation strategy that incorporates speckle decorrelation measures from real videos and uses an iterative refinement process to improve motion realism.

Load-bearing premise

The motion references created by the enhanced simulations are realistic enough that a model trained on them generalizes to real clinical data and beats the existing clinical method.

What would settle it

A model trained on the new dataset and evaluated on an independent clinical echocardiogram collection that shows higher strain error than speckle tracking echocardiography would falsify the central claim.

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

If this is right

  • The deep learning model trained on the dataset reaches unmatched performance on global and regional strain.
  • The open-source dataset of 1,478 videos supplies reference motion for training additional algorithms.
  • Global longitudinal strain variability falls to 1.42 percent in an inter-expert setting versus 1.78 percent for the clinical reference.
  • Regional strain accuracy improves, supporting earlier diagnosis of subtle cardiac abnormalities.

Where Pith is reading between the lines

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

  • Similar hybrid simulation techniques could address ground-truth shortages in other ultrasound or medical imaging tasks.
  • The approach may allow training larger models without increasing the need for expert manual annotations.
  • If the realism transfers, regional strain improvements could enable more localized diagnostic biomarkers than global measures alone provide.

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

2 major / 1 minor

Summary. The manuscript proposes a novel simulation strategy for echocardiographic videos that incorporates speckle decorrelation measures extracted from real clinical videos together with an iterative refinement process to increase motion realism. This pipeline is used to generate an open-source dataset of 1,478 videos containing reference motion fields; the dataset trains a deep-learning motion-estimation network whose performance on global and regional strain is reported to exceed that of clinical speckle-tracking echocardiography (STE), most notably achieving a GLS inter-expert variability of 1.42 % versus 1.78 % for the clinical reference.

Significance. If the simulation-to-clinical transfer is shown to be reliable, the work would supply both a publicly available training resource and evidence that deep-learning strain estimators can surpass the consistency of current clinical STE. The open-source dataset constitutes a concrete, reusable contribution that could accelerate further DL development in the field.

major comments (2)
  1. [Abstract] Abstract: the headline claim that the trained network reaches a GLS variability of 1.42 % (versus 1.78 % for the clinical reference) is presented without any quantitative verification that the simulated motion fields match real myocardial deformation statistics (e.g., endpoint error on held-out clinical sequences, strain bias, or distribution matching of decorrelation measures). This verification is load-bearing for the central generalization claim.
  2. [Abstract] Abstract: no description is given of the protocol used for the inter-expert comparison, the statistical test applied to the 1.42 % versus 1.78 % difference, or the exclusion criteria applied to the clinical data. These omissions directly affect the interpretability of the reported superiority.
minor comments (1)
  1. [Abstract] Abstract: the sentence beginning 'its accuracy for regional strain remains limited...' ends with the fragment '. from clinical data.' which appears to be a typographical or copy-paste error.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the abstract to improve clarity and self-containment while preserving the manuscript's core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claim that the trained network reaches a GLS variability of 1.42 % (versus 1.78 % for the clinical reference) is presented without any quantitative verification that the simulated motion fields match real myocardial deformation statistics (e.g., endpoint error on held-out clinical sequences, strain bias, or distribution matching of decorrelation measures). This verification is load-bearing for the central generalization claim.

    Authors: The full manuscript reports quantitative validation of the simulation pipeline, including distribution matching of speckle decorrelation measures extracted from real videos and iterative refinement to align strain statistics with clinical observations. We acknowledge that the abstract does not explicitly reference these metrics. We will revise the abstract to include a concise statement noting the use of real decorrelation measures and iterative refinement for realism, along with a reference to the validation results in the main text. revision: yes

  2. Referee: [Abstract] Abstract: no description is given of the protocol used for the inter-expert comparison, the statistical test applied to the 1.42 % versus 1.78 % difference, or the exclusion criteria applied to the clinical data. These omissions directly affect the interpretability of the reported superiority.

    Authors: The inter-expert comparison protocol, statistical testing (including the test used for the reported difference), and exclusion criteria are described in the Methods section. We agree that the abstract would be strengthened by a brief mention of the evaluation protocol. We will revise the abstract to include a short description of the inter-expert variability assessment. revision: yes

Circularity Check

0 steps flagged

No significant circularity; performance claims rest on external clinical benchmarks

full rationale

The paper derives its motion estimation network by training exclusively on a synthetic dataset whose motion fields are generated via a simulation pipeline that ingests speckle decorrelation statistics from real videos plus iterative refinement; the resulting model is then evaluated on held-out clinical sequences using an external inter-expert variability reference (1.42 % GLS vs 1.78 % clinical). No equation, fitted parameter, or performance metric is shown to equal its own training inputs by construction, nor does any load-bearing premise collapse to a self-citation chain. The comparison remains to an independent clinical standard, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, invented entities, or detailed axioms are stated. The approach implicitly assumes that physics-based models can be made sufficiently realistic by injecting real speckle statistics.

axioms (1)
  • domain assumption Physics-based cardiac motion models can be refined with speckle decorrelation statistics extracted from real echocardiographic videos to produce usable ground-truth motion for training.
    This premise underpins the entire simulation pipeline described in the abstract.

pith-pipeline@v0.9.1-grok · 5793 in / 1366 out tokens · 32848 ms · 2026-06-29T09:27:18.065810+00:00 · methodology

0 comments
read the original abstract

Speckle tracking echocardiography (STE) is the clinical standard for myocardial strain estimation. Despite good performance on global strain (GLS), its accuracy for regional strain remains limited, even though this biomarker is highly relevant for early diagnosis and the characterization of subtle abnormalities. from clinical data. Deep learning is a promising alternative, but its development is constrained by the lack of reliable motion references. Existing solutions rely either on STE-derived labels or on simulations generated by physics-based models, but these synthetic sequences still have limited realism compared with clinical data.In this paper, we propose a novel simulation strategy that incorporates speckle decorrelation measures from real videos and uses an iterative refinement process to improve the motion realism in the simulations. We created an open-source photorealistic dataset of 1,478 videos with reference motion, which was used to train an echocardiographic motion estimation algorithm. The proposed method achieves unmatched performance on global and regional strain, notably reaching a GLS variability of 1.42% in an inter-expert setting compared to 1.78% for the clinical reference.

Figures

Figures reproduced from arXiv: 2605.28697 by Anders Austlid Task\'en, Andreas {\O}stvik, Bj{\o}rnar Grenne, Gabriel Kiss, Harald Brunvand, Havard Dalen, Khuram Faraz, Lasse Lovstakken, Md Abulkalam Azad, Nicolas Duchateau, Olivier Bernard, Pierre-Marc Jodoin, Pierre-Yves Courand, Sigve Karlsen, Thierry Judge, Thor Edvardsen.

Figure 1
Figure 1. Figure 1: Overview of the proposed simulation pipeline. Starting from an echocardiographic video and mesh approximating the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sample frames from a simulated sequence and its corresponding real template sequence are shown. The last row displays [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Tracking results on respective worst-case sequences [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Bland–Altman analysis of TAS-Net (T) for GLS and [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Regional strain curves from the best-performing deep [PITH_FULL_IMAGE:figures/full_fig_p009_5.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

34 extracted references · 1 canonical work pages

  1. [1]

    Recommendations for cardiac chamber quantification by echocardiography in adults,

    R. M. Langet al., “Recommendations for cardiac chamber quantification by echocardiography in adults,”Eur Heart J Cardiovasc Imaging, vol. 16, pp. 233–71, 2015

    2015

  2. [2]

    Beyond ejection fraction: An integrative approach for assessment of cardiac structure and function in heart failure,

    M. Cikes and S. D. Solomon, “Beyond ejection fraction: An integrative approach for assessment of cardiac structure and function in heart failure,”Eur Heart J, vol. 37, pp. 1642–50, 2016

    2016

  3. [3]

    Automatic measurements of left ventricular volumes and ejection fraction by artificial intelligence: clinical validation in real time and large databases,

    S. Olaisenet al., “Automatic measurements of left ventricular volumes and ejection fraction by artificial intelligence: clinical validation in real time and large databases,”Eur Heart J Cardiovasc Imaging, vol. 25, pp. 383–95, 2023

    2023

  4. [4]

    Myocardial Strain Imaging,

    O. A. Smiseth, O. Rider, M. Cvijic, L. Valkovi ˇc, E. W. Remme, and J.-U. V oigt, “Myocardial Strain Imaging,”JACC Cardiovasc Imaging, vol. 18, pp. 340–81, 2025

    2025

  5. [5]

    Speckle-Tracking Strain Echocardiography for the Assessment of Left Ventricular Structure and Function,

    C. G. Mihoset al., “Speckle-Tracking Strain Echocardiography for the Assessment of Left Ventricular Structure and Function,”Circulation, vol. 152, 2025

    2025

  6. [6]

    EchoTracker: Advancing Myocardial Point Tracking in Echocardiography,

    M. A. Azadet al., “EchoTracker: Advancing Myocardial Point Tracking in Echocardiography,” inMICCAI 2024, 2024, vol. 15004, pp. 645–55

    2024

  7. [7]

    Low Complexity Point Tracking of the My- ocardium in 2D echocardiography,

    A. Chernyshovet al., “Low Complexity Point Tracking of the My- ocardium in 2D echocardiography,”arXiv:2503.10431, 2025

    work page arXiv 2025

  8. [8]

    Motion estimation by deep learning in 2D echocar- diography: Synthetic dataset and validation,

    E. Evainet al., “Motion estimation by deep learning in 2D echocar- diography: Synthetic dataset and validation,”IEEE TMI, vol. 41, pp. 1911–24, 2022

    1911

  9. [9]

    Estimation of Segmental Longitudinal Strain in Transesophageal Echocardiography by Deep Learning,

    A. A. Task ´enet al., “Estimation of Segmental Longitudinal Strain in Transesophageal Echocardiography by Deep Learning,”IEEE JBHI, vol. 30, pp. 436–47, 2026

    2026

  10. [10]

    Large-scale simulation of realistic cardiac ultrasound data with clinical appearance: Methodology and open-access database,

    N. Burman, C. A. Manetti, S. V . Heymans, M. Ingram, J. Lumens, and J. D’hooge, “Large-scale simulation of realistic cardiac ultrasound data with clinical appearance: Methodology and open-access database,”IEEE Access, vol. 12, pp. 117 040–55, 2024

    2024

  11. [11]

    CoTracker: It Is Better to Track Together,

    N. Karaev, I. Rocco, B. Graham, N. Neverova, A. Vedaldi, and C. Rup- precht, “CoTracker: It Is Better to Track Together,” inECCV, 2025, vol. 15120, pp. 18–35

    2025

  12. [12]

    Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume,

    D. Sun, X. Yang, M.-Y . Liu, and J. Kautz, “Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume,” inCVPR, 2018, pp. 8934–43

    2018

  13. [13]

    RAFT: Recurrent All-Pairs Field Transforms for Optical Flow,

    Z. Teed and J. Deng, “RAFT: Recurrent All-Pairs Field Transforms for Optical Flow,” inECCV, 2020, pp. 402–19

    2020

  14. [14]

    Tap-vid: A benchmark for tracking any point in a video,

    C. Doerschet al., “Tap-vid: A benchmark for tracking any point in a video,”NeurIPS, vol. 35, pp. 13 610–26, 2022

    2022

  15. [15]

    Kubric: A scalable dataset generator,

    K. Greffet al., “Kubric: A scalable dataset generator,” inCVPR, 2022, pp. 3749–61

    2022

  16. [16]

    Pointodyssey: A large-scale synthetic dataset for long-term point track- ing,

    Y . Zheng, A. W. Harley, B. Shen, G. Wetzstein, and L. J. Guibas, “Pointodyssey: A large-scale synthetic dataset for long-term point track- ing,” inICCV, 2023, pp. 19 855–65

    2023

  17. [17]

    Particle Video Revisited: Tracking Through Occlusions Using Point Trajectories,

    A. W. Harley, Z. Fang, and K. Fragkiadaki, “Particle Video Revisited: Tracking Through Occlusions Using Point Trajectories,” inECCV, 2022, vol. 13682, pp. 59–75

    2022

  18. [18]

    Tapir: Tracking any point with per-frame initialization and temporal refinement,

    C. Doerschet al., “Tapir: Tracking any point with per-frame initialization and temporal refinement,” inICCV, 2023, pp. 10 061–72

    2023

  19. [19]

    Cotracker3: Simpler and better point tracking by pseudo- labelling real videos,

    N. Karaev, Y . Makarov, J. Wang, N. Neverova, A. Vedaldi, and C. Rupprecht, “Cotracker3: Simpler and better point tracking by pseudo- labelling real videos,” inICCV, 2025, pp. 6013–22

    2025

  20. [20]

    Introduction to speckle tracking in cardiac ultrasound imaging,

    D. Garcia, P. Lantelme, and E. Saloux, “Introduction to speckle tracking in cardiac ultrasound imaging,”Speckle filtering and tracking in cardio- vascular ultrasound imaging and video, pp. 571–98, 2018

    2018

  21. [21]

    Inter-vendor variability in strain measurements depends on software rather than image characteristics,

    S. ¨Unl¨uet al., “Inter-vendor variability in strain measurements depends on software rather than image characteristics,”Int J Cardiovasc Imaging, vol. 37, 2021

    2021

  22. [22]

    SIMUS: An open-source simulator for medical ultrasound imaging. Part I: Theory & examples,

    D. Garcia, “SIMUS: An open-source simulator for medical ultrasound imaging. Part I: Theory & examples,”Comput Methods Programs Biomed, vol. 218, p. 106726, 2022

    2022

  23. [23]

    Adaptation to mechanical load determines shape and properties of heart and circulation: The CircAdapt model,

    T. Arts, T. Delhaas, P. Bovendeerd, X. Verbeek, and F. W. Prinzen, “Adaptation to mechanical load determines shape and properties of heart and circulation: The CircAdapt model,”Am J Physiol Heart Circ Physiol, vol. 288, pp. 1943–54, 2005

    1943

  24. [24]

    A fast convolution-based methodology to simulate 2- Dd/3-D cardiac ultrasound images,

    H. Gaoet al., “A fast convolution-based methodology to simulate 2- Dd/3-D cardiac ultrasound images,”IEEE TUFFC (56), pp. 404–9, 2009

    2009

  25. [25]

    Myocardial strain: a clinical review

    B. Brady, G. King, R. Murphy, and D. Walsh, “Myocardial strain: a clinical review.”Inter J Med Sci, vol. 16, pp. 1–8, 2022

    2022

  26. [26]

    So you think you can das? a viewpoint on delay-and-sum beamforming,

    V . Perrot, M. Polichetti, F. Varray, and D. Garcia, “So you think you can das? a viewpoint on delay-and-sum beamforming,”Ultrasonics, vol. 111, p. 106309, 2021

    2021

  27. [27]

    Realistic vendor-specific synthetic ultrasound data for quality assurance of 2-d speckle tracking echocardiography: Simulation pipeline and open access database,

    M. Alessandriniet al., “Realistic vendor-specific synthetic ultrasound data for quality assurance of 2-d speckle tracking echocardiography: Simulation pipeline and open access database,”IEEE TUFFC, vol. 65, pp. 411–22, 2018

    2018

  28. [28]

    Segment anything,

    A. Kirillovet al., “Segment anything,” inICCV, 2023, pp. 4015–26

    2023

  29. [29]

    Domain Adaptation of Echocardiography Segmentation Via Reinforcement Learning ,

    A. Judgeet al., “ Domain Adaptation of Echocardiography Segmentation Via Reinforcement Learning ,” inMICCAI, 2024

    2024

  30. [30]

    Deep learning for segmentation using an open large- scale dataset in 2D echocardiography,

    S. Leclercet al., “Deep learning for segmentation using an open large- scale dataset in 2D echocardiography,”IEEE TMI, pp. 2198–210, 2019 (38)

    2019

  31. [31]

    Echocar- diography segmentation with enforced temporal consistency,

    N. Painchaud, N. Duchateau, O. Bernard, and P.-M. Jodoin, “Echocar- diography segmentation with enforced temporal consistency,”IEEE TMI, vol. 41, pp. 2867–78, 2022

    2022

  32. [32]

    Extraction of V olumetric Indices from Echocardiography: Which Deep Learning Solution for Clinical Use?

    H. J. Ling, N. Painchaud, P.-Y . Courand, P.-M. Jodoin, D. Garcia, and O. Bernard, “Extraction of V olumetric Indices from Echocardiography: Which Deep Learning Solution for Clinical Use?” inFIMH, 2023, vol. 13958, pp. 245–54

    2023

  33. [33]

    Deep learning for improved precision and repro- ducibility of left ventricular strain in echocardiography: A test-retest study,

    I. M. Salteet al., “Deep learning for improved precision and repro- ducibility of left ventricular strain in echocardiography: A test-retest study,”Journal of the American Society of Echocardiography, vol. 36, no. 7, pp. 788–799, 2023, 34th ASE Annual Scientific Sessions

    2023

  34. [34]

    Echocardiographic reference ranges of global longitu- dinal strain for all cardiac chambers using guideline-directed dedicated views,

    J. Nyberget al., “Echocardiographic reference ranges of global longitu- dinal strain for all cardiac chambers using guideline-directed dedicated views,”JACC: Cardiovascular Imaging, vol. 16, no. 12, pp. 1516–1531, 2023

    2023