Pith. sign in

REVIEW 3 major objections 4 minor 22 references

This paper argues that an apparent ejection-fraction calibration gain in cardiac digital twins is a measurement-convention artifact, not a genuine model improvement.

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

2026-08-05 00:17 UTC pith:MAYRO7VT

load-bearing objection The CAMUS phase-conditioning EF gain is almost certainly a plane-convention artefact, and the paper proves it convincingly on CAMUS itself; the EchoNet replication and one 'indistinguishable' claim need tightening, but this deserves a serious referee. the 3 major comments →

arxiv 2608.01602 v1 pith:MAYRO7VT submitted 2026-08-03 cs.CV

When Measurement Conventions Masquerade as Calibration Gains in Cardiac Digital Twins

Dang P. M. Cao , Hieu Pham This is my paper
classification cs.CV
keywords cardiac digital twinejection fraction calibrationobservation operatormeasurement conventionphase conditioningCAMUSEchoNet-Dynamicconformal prediction
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.

This paper argues that a reported calibration improvement for cardiac digital twins is an artifact of how ejection fraction is measured, not a real gain in the image-to-measurement mapping. On the CAMUS benchmark, phase-conditioned segmentation models appear to remove baseline EF bias, but the apparent gain disappears when EF is computed from the same single-plane ground-truth masks used by the extractor: matched-reference errors are statistically flat across all four models. The paper traces the CAMUS pattern to a measurement-convention offset: single-plane ground-truth EF exceeds CAMUS biplane clinical EF by +6.30 EF points, which accounts for nearly all of the baseline bias. A prespecified replication on EchoNet-Dynamic, where labels and extractor share the four-chamber plane, fails to reproduce the phase-conditioning advantage and reverses the model ranking. The paper concludes that EF calibration should be audited for the full model–extractor–reference pair and proposes a five-step Convention-Aware EF Audit protocol.

Core claim

The central discovery is that the phase-conditioning EF calibration gain observed on CAMUS is not a genuine observation-operator improvement. When EF is computed from ground-truth masks using the same single-plane method-of-disks extractor applied to predictions, the four front-ends (B1, B2, P1, P2) are statistically indistinguishable on extractor-matched error (MAE_gt), and the baseline models are not miscalibrated once the measured convention offset is removed. The apparent gain is instead explained by a +6.30 EF-point systematic difference between single-plane ground-truth EF and CAMUS's biplane clinical EF reference. A plane-aligned replication on EchoNet-Dynamic confirms the explanation

What carries the argument

The measurement-convention offset: the difference between single-plane ground-truth EF computed by the same method-of-disks extractor and the dataset's clinical biplane EF reference. This offset (+6.30 EF points on CAMUS) is the mechanism that makes baseline models look miscalibrated and phase-conditioned models look calibrated. The matched-reference diagnostic (EF MAE_gt, computed from ground-truth masks with the same extractor) is what exposes the illusion, by showing flat segmentation-error across front-ends while clinical-reference error varies.

Load-bearing premise

The EchoNet-Dynamic replication is a valid plane-aligned test: its EF labels, though aligned to the four-chamber plane, may still differ from the extractor in ED/ES frame selection, contour style, preprocessing, and calculation, and those differences could hide a genuine phase-conditioning effect.

What would settle it

A process-matched replication: construct a dataset where four-chamber EF labels are computed with the same ED/ES selection, contour style, and method-of-disks calculation as the extractor, then test whether phase-conditioned front-ends improve extractor-matched EF MAE_gt. If P1/P2 beat B1/B2 on MAE_gt under those conditions, the convention-offset explanation is incomplete or wrong.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • EF calibration studies must report extractor-matched error (MAE_gt) alongside clinical-reference error; a flat MAE_gt with varying MAE_clin is a red flag for a convention confound.
  • CAMUS baseline EF bias is consistent with zero after subtracting the convention offset, so the U-Net front-ends were never miscalibrated in the single-plane sense.
  • Plane-aligned evaluation on EchoNet-Dynamic shows B2 (adjacent-frame-smoothed) is the strongest deployable front-end, with streaming inference, no invalid EF estimates, and the best clinical EF MAE.
  • Downstream twin-state errors (stroke volume, cardiac output, contractility) are much smaller when the convention gap is removed; under plane-aligned conditions, haemodynamic errors fall to roughly 3–4%.
  • The paper's Convention-Aware EF Audit protocol—compute MAE_gt, characterize the convention offset, check seed consistency, add a safety gate, and replicate under an aligned reference—provides a template for auditing future EF observation operators.

Where Pith is reading between the lines

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

  • The proportional convention gap (slope 0.613) implies that a constant +6.30-point correction leaves structured residual error, especially at high EF; a view-aware or linear correction, validated prospectively, would likely be needed for patient-level EF adjustment.
  • The EchoNet replication is strong evidence against phase conditioning, but it remains process-mismatched; a fully process-matched test with identical ED/ES selection, contour style, and calculation on four-chamber labels would directly isolate how much of the reversal is due to the plane alone.
  • The same convention-mismatch mechanism may affect other benchmarks that mix single-plane and biplane references, and other physiological ratios computed from segmentation (e.g., stroke volume, regurgitant fractions), where small coherent contour shifts amplify into apparent model differences.
  • If the audit protocol were widely adopted, some previously published calibration gains on CAMUS-derived tasks would likely require re-evaluation with matched-reference EF.

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 / 4 minor

Summary. This paper audits the calibration of four U-Net-based ejection-fraction (EF) front-ends (frame-wise B1, smoothed B2, phase-conditioned P1, and P2 with additional consistency) on two echocardiography datasets. On CAMUS, phase-conditioned models appear to reduce bias relative to the dataset's biplane clinical EF. The authors argue this is a measurement-convention artefact: when EF is computed from ground-truth masks with the same single-plane extractor, MAE_gt is claimed to be statistically indistinguishable across front-ends, and a measured +6.30 EF-point offset between single-plane ground-truth EF and CAMUS biplane clinical EF accounts for the baseline clinical bias. A prespecified EchoNet-Dynamic replication, in which the clinical labels are four-chamber (plane-aligned) but not otherwise process-matched, removes the positive baseline bias and reverses the CAMUS ranking. The paper contributes a Convention-Aware EF Audit protocol and reports downstream twin-state, conformal, and EF-stratum analyses.

Significance. If the conclusions hold, the paper makes a useful methodological contribution: calibration gains for image-to-measurement operators must be evaluated under a matched reference convention, and otherwise benchmark artefacts can masquerade as model improvements. The strengths are the clean decomposition of baseline bias into convention and segmentation components, the use of a prespecified external replication, and the release of code/data. The main limitations are the small CAMUS sample (n=49-50) and the acknowledged process mismatch in the EchoNet replication, which leaves the external evidence partly equivocal. The downstream analyses are clearly labeled as simplified.

major comments (3)
  1. [§3.2, Table 1] The claim that 'EF MAE_gt is statistically indistinguishable across front-ends' is load-bearing (used in the abstract and Section 3.2) but no statistical test is reported. Section 3.1 states that paired Wilcoxon tests, bootstrap CIs, Cohen's d_z, and Holm correction are used, yet none of the corresponding p-values or intervals appear in the text or tables. Table 1 reports mean±std across three seeds, not patient-level dispersion; with P1 MAE_gt=10.62 vs B2=8.08 and Bias_gt=-7.56 vs -1.37, the claim requires the missing patient-level test results. If P1 is significantly worse, the paper should say that; the main conclusion (no genuine improvement) still stands, but 'statistically indistinguishable' would be incorrect and should be revised. Moreover, failure to reject a null hypothesis is not evidence of equivalence; if the tests are underpowered, the phrase should be 'no significant diffe
  2. [§3.3, Table 2] The statement that the +6.30 EF-point convention offset 'explains nearly all baseline bias' is based on point estimates with wide intervals. The convention offset CI is [+3.69,+8.91], and the residual segmentation component CI is [-4.08,+1.14]. The point estimate of the residual is -1.47, which is about 30% of the B1 clinical bias (+4.83); the CI lower bound allows a residual of -4.08. Given the small sample (n=49), the authors should discuss the uncertainty more explicitly and avoid the phrase 'nearly all' without qualification. In addition, the measured offset is not purely a plane effect: it also includes differences in contour style, ED/ES selection, and calculation between the single-plane extractor and the CAMUS biplane clinical reference. The paper should either call it a 'reference-convention offset' or identify which components are separable.
  3. [§3.4, Tables 3-4] The EchoNet replication is presented as the decisive external confirmation ('reverses the CAMUS ranking'; 'expected when the dominant plane mismatch is removed'), but the paper concedes that EchoNet is 'plane-aligned, but not process-matched' (Section 3.4 and Limitations: ED/ES selection, contour style, preprocessing, and calculation may all differ). The prespecified criteria in Table 4 assume that a genuine CAMUS effect would require positive baseline bias and phase-model improvement on EchoNet; this only follows if the plane mismatch is the only systematic difference. Without a quantitative bound on the process-mismatch offset, the observed negative baseline bias on EchoNet could be a separate convention offset rather than evidence that the plane mismatch was removed. The authors should either quantify this offset (e.g., apply the extractor to EchoNet expert tracings and compare with E
minor comments (4)
  1. [Eq. (1)] The volume formula writes `cEF = [EDV− [ESV / [EDV ×100`; the bracket notation is confusing. Please use \widehat{EDV} or \hat{EDV} and add parentheses around (EDV-ESV)/EDV.
  2. [Fig. 1(c)] The caption reports 'Mean = -6.30 EF pts' for 'biplane clinical - GT single', while Table 2 reports +6.30 for 'single-plane GT - biplane clinical'. Please make the sign convention consistent.
  3. [§3.2, Table 1] The claim 'Baselines are seed-consistent, whereas phase-conditioned models vary more' is not uniformly supported by the seed std columns. B2 (baseline) has Bias_clin std 1.61 and Bias_gt std 0.93; P2 has Bias_clin std 1.31 and MAE_gt std 1.26; P1 has MAE_gt std 0.48, lower than both baselines. Please qualify this statement.
  4. [Table 3] The caption uses MAE_gt, MAE_clin, and Biasclin without defining them; refer readers to Section 2.2.

Circularity Check

0 steps flagged

No significant circularity: the convention-offset decomposition and EchoNet replication are independent evidence.

full rationale

The paper's central claim is that the apparent CAMUS phase-conditioning EF calibration gain is a measurement-convention artefact, not a genuine model improvement. This claim is supported by two independent lines of evidence. First, the CAMUS convention-offset analysis (Section 3.3, Table 2) measures the single-plane-GT-to-biplane-clinical EF gap (+6.30 EF points) directly from ground-truth masks using the same extractor, independent of any front-end model output. The residual segmentation component is then the algebraic difference between the observed B1 clinical bias and this measured offset; it is not a fitted value used to derive the main result. The post-hoc OLS slope and intercept in Table 2 are explicitly labeled 'Post-hoc OLS; not a validated correction' and are not load-bearing for the conclusion. Second, the EchoNet replication (Section 3.4, Table 3) is an external, prespecified test on a different dataset with plane-aligned labels. The paper explicitly acknowledges that EchoNet is 'plane-aligned, but not process-matched' (Introduction and Section 3.4), which is a stated limitation about residual confounds, not a circular reduction. No load-bearing step reduces to its own input by construction, and no self-citation is used to justify the core premise. The derivation chain is self-contained: the convention offset is measured, not fitted, and the replication is a genuine external check. The skeptical concern about EchoNet process mismatch is a validity limitation, not circularity, and is already disclosed in the paper.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The audit relies on measured rather than fitted quantities for its central claim. The only fitted numbers are the post-hoc OLS slope/intercept, explicitly labeled non-validated. Main axioms are the validity of the single-plane method-of-disks extractor, the plane-mismatch interpretation of CAMUS vs EchoNet, and the simplification of haemodynamic propagation. No invented entities.

free parameters (3)
  • Single-plane to biplane convention offset (used as correction) = +6.30 EF points (95% CI 3.69-8.91)
    Measured mean difference between extractor EF on GT masks and CAMUS biplane clinical EF (n=49). It is an empirical estimate, not a fitted parameter for prediction, but it is the key quantity subtracted from baseline bias.
  • Post-hoc OLS slope = 0.613 (95% CI 0.47-0.76)
    Fitted slope of biplane clinical EF vs single-plane GT EF; labeled post-hoc and not a validated correction. Used only to show the gap is proportional, not constant.
  • Post-hoc OLS intercept = +13.34 EF points
    Intercept of the same post-hoc OLS; not used for correction.
axioms (5)
  • domain assumption Single-plane method-of-disks EF (Eq 1) is a valid observation operator for EF from four-chamber clips.
    Adopted from standard chamber quantification [8]; used for all models and for ground-truth masks. If the extractor itself biases EF, that bias is common to all comparisons.
  • domain assumption CAMUS biplane clinical EF is the reference convention that causes the apparent gain.
    CAMUS reports biplane EF; the extractor uses one four-chamber plane, so a systematic offset is expected. This premise underpins the whole analysis.
  • domain assumption EchoNet-Dynamic clinical EF labels are plane-aligned to the four-chamber extractor (though not process-matched).
    Stated in Section 2.1 and used to interpret the EchoNet replication as removing the dominant plane mismatch. The paper acknowledges residual process differences.
  • domain assumption Prespecified checks and exclusions were actually fixed before EchoNet evaluation.
    The paper asserts prespecification (Section 2.2, Section 3.1) without registration or timestamp; if false, the replication loses evidential value.
  • domain assumption Simplified lumped haemodynamic propagation (SV, CO, Ees) captures order-of-magnitude twin-state errors.
    Section 3.5 explicitly labels this as order-of-magnitude, not a patient-specific closed-loop simulation.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of When Measurement Conventions Masquerade as Calibration Gains in Cardiac Digital Twins." pith.science (2026). https://pith.science/paper/MAYRO7VT

@misc{pith2026260801602,
  author       = {Pith},
  title        = {Pith review of: When Measurement Conventions Masquerade as Calibration Gains in Cardiac Digital Twins},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MAYRO7VT}},
  note         = {Machine review of arXiv:2608.01602}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Cardiac digital twins convert clinical images into physiological measurements through observation operators, yet calibration studies often assume a fixed reference convention. Across four shared-backbone echocardiographic EF front-ends, phase conditioning appears to remove CAMUS baseline bias. Matched-reference analysis rejects this gain: singleplane ground-truth EF error is statistically indistinguishable across models, while single-plane ground-truth EF exceeds CAMUS biplane clinical EF by +6.30 points, explaining nearly all baseline bias. A prespecified EchoNet-Dynamic replication, with released data and our extractor aligned to the apical four-chamber plane, removes baseline overestimation and reverses the CAMUS ranking. We also quantify haemodynamic effects, conformal residual-width budgets, and EF-stratum changes, yielding a Convention-Aware EF Audit protocol that separates genuine observation operator calibration from measurement artefacts. GitHub: EjectionFraction-Bias-in-Cardiac-Digital-Twin.git

Figures

Figures reproduced from arXiv: 2608.01602 by Dang P. M. Cao, Hieu Pham.

Figure 1
Figure 1. Figure 1: Cross-dataset EF calibration audit. (a) CAMUS baselines overestimate clinical EF while phase-conditioned models appear near zero; plane-aligned EchoNet evaluation removes this pattern. Shaded band: ±2 EF pts. (b) B2 Bland–Altman on CAMUS [2]. (c) CAMUS convention offset (n = 49): single-plane GT EF exceeds biplane clinical EF by +6.30 points. (d) OLS mapping; slope < 1 indicates a proportional gap. and ven… view at source ↗
Figure 2
Figure 2. Figure 2: Downstream twin-state errors under CAMUS biplane-reference conditions (mean±95% CI), using Section 3.5. Under plane-aligned EchoNet conditions, SV and CO errors fall to about 3–4%, so much of the penalty is recoverable after removing the convention gap. tion assumes fixed volumes and heart rate rather than a patient-specific closed￾loop twin. Acknowledgments. The authors thank Mr. Ha-Hieu Pham for his valu… 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

22 extracted references · 12 canonical work pages

  1. [1]

    Foun- dations and Trends in Machine Learning16(4), 494–591 (2023)

    Angelopoulos, A.N., Bates, S.: Conformal prediction: A gentle introduction. Foun- dations and Trends in Machine Learning16(4), 494–591 (2023). https://doi.org/ 10.1561/2200000101

    doi:10.1561/2200000101 2023

  2. [2]

    The Lancet327(8476), 307–310 (1986)

    Bland, J.M., Altman, D.G.: Statistical methods for assessing agreement between two methods of clinical measurement. The Lancet327(8476), 307–310 (1986). https://doi.org/10.1016/S0140-6736(86)90837-8

    doi:10.1016/s0140-6736(86)90837-8 1986

  3. [3]

    Medical Image Analysis100, 103361 (2025)

    Camps, J., Wang, Z.J., Doste, R., Berg, L.A., Holmes, M., Lawson, B., Tomek, J., Burrage, K., Bueno-Orovio, A., Rodriguez, B.: Harnessing 12-lead ECG and MRI data to personalise repolarisation profiles in cardiac digital twin models for enhanced virtual drug testing. Medical Image Analysis100, 103361 (2025). https: //doi.org/10.1016/j.media.2024.103361

    arXiv 2025

  4. [4]

    European Heart Journal41(48), 4556– 4564 (2020)

    Corral-Acero, J., Margara, F., Marciniak, M., Rodero, C., Loncaric, F., Feng, Y., Gilbert, A., Fernandes, J.F., Bukhari, H.A., Wajdan, A., et al.: The ‘digital twin’ to enable the vision of precision cardiology. European Heart Journal41(48), 4556– 4564 (2020). https://doi.org/10.1093/eurheartj/ehaa159

    doi:10.1093/eurheartj/ehaa159 2020

  5. [5]

    LectureNotesinComputerScience,vol.10553,pp.100–108.Springer(2017).https: //doi.org/10.1007/978-3-319-67558-9_12

    Dezaki, F.T., Dhungel, N., Abdi, A.H., Luong, C., Tsang, T., Jue, J., Gin, K., Hawley, D., Rohling, R., Abolmaesumi, P.: Deep residual recurrent neural networks forcharacterisationofcardiaccyclephasefromechocardiograms.In:DeepLearning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. LectureNotesinComputerScience,vol.10553,p...

    work page doi:10.1007/978-3-319-67558-9_12 2017

  6. [6]

    Circulation145(18), e895–e1032 (2022)

    Heidenreich, P.A., Bozkurt, B., Aguilar, D., Allen, L.A., Byun, J.J., Colvin, M.M., Deswal, A., Drazner, M.H., Dunlay, S.M., Evers, L.R., Fang, J.C., Fed- son, S.E., Fonarow, G.C., Hayek, S.S., Hernandez, A.F., Khazanie, P., Kittle- son, M.M., Lee, C.S., Link, M.S., Milano, C.A., Nnacheta, L.C., Sandhu, A.T., Stevenson, L.W., Vardeny, O., Vest, A.R., Yanc...

    2022

  7. [7]

    Journal of Basic Engineering82(1), 35–45 (1960)

    Kalman, R.E.: A new approach to linear filtering and prediction problems. Journal of Basic Engineering82(1), 35–45 (1960). https://doi.org/10.1115/1.3662552

    doi:10.1115/1.3662552 1960

  8. [8]

    Journal of the American Society of Echocardio- graphy28(1), 1–39.e14 (2015)

    Lang, R.M., Badano, L.P., Mor-Avi, V., Afilalo, J., Armstrong, A., Ernande, L., Flachskampf, F.A., Foster, E., Goldstein, S.A., Kuznetsova, T., et al.: Recom- mendations for cardiac chamber quantification by echocardiography in adults: An update from the American Society of Echocardiography and the European Associ- ation of Cardiovascular Imaging. Journal...

    doi:10.1016/j.echo.2014.10.003 2015

  9. [9]

    IEEE Transactions on Medical Imaging 38(9), 2198–2210 (2019)

    Leclerc, S., Smistad, E., Pedrosa, J., Østvik, A., Cervenansky, F., Espinosa, F., Espeland, T., Berg, E.A.R., Jodoin, P.M., Grenier, T., Lartizien, C., D’hooge, J., Lovstakken, L., Bernard, O.: Deep learning for segmentation using an open large-scale dataset in 2D echocardiography. IEEE Transactions on Medical Imaging 38(9), 2198–2210 (2019). https://doi....

    work page arXiv 2019

  10. [10]

    In: International Conference on Learning Representations (2019), https://openreview.net/forum? id=Bkg6RiCqY7

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: International Conference on Learning Representations (2019), https://openreview.net/forum? id=Bkg6RiCqY7

    work page 2019

  11. [11]

    Nature ComputationalScience1(5), 313–320(2021)

    Niederer, S.A., Sacks, M.S., Girolami, M., Willcox, K.: Scaling digital twins from the artisanal tothe industrial. Nature ComputationalScience1(5), 313–320(2021). https://doi.org/10.1038/s43588-021-00072-5 10 D. P. M. Cao and H. Pham

    work page doi:10.1038/s43588-021-00072-5 2021

  12. [12]

    Nature580(7802), 252–256 (2020)

    Ouyang, D., He, B., Ghorbani, A., Yuan, N., Ebinger, J., Langlotz, C.P., Heidenre- ich, P.A., Harrington, R.A., Liang, D.H., Ashley, E.A., Zou, J.Y.: Video-based AI for beat-to-beat assessment of cardiac function. Nature580(7802), 252–256 (2020). https://doi.org/10.1038/s41586-020-2145-8

    doi:10.1038/s41586-020-2145-8 2020

  13. [13]

    Biomechanics and Modeling in Mechanobiology20(3), 803–831 (2021)

    Peirlinck, M., Sahli Costabal, F., Yao, J., Guccione, J.M., Tripathy, S., Wang, Y., Ozturk, D., Segars, P., Morrison, T.M., Levine, S., Kuhl, E.: Precision medicine in human heart modeling: Perspectives, challenges, and opportunities. Biomechanics and Modeling in Mechanobiology20(3), 803–831 (2021). https://doi.org/10.1007/ s10237-021-01421-z

    work page 2021

  14. [14]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Perez, E., Strub, F., de Vries, H., Dumoulin, V., Courville, A.: FiLM: Visual rea- soning with a general conditioning layer. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 32, pp. 3942–3951 (2018). https://doi.org/10.1609/ aaai.v32i1.11671

    work page 2018

  15. [15]

    Nature Cardiovascular Research4(5), 624–636 (2025)

    Qian, S., Ugurlu, D., Fairweather, E., Dal Toso, L., Deng, Y., Strocchi, M., Ci- cci, L., Jones, R.E., Zaidi, H., Prasad, S., Halliday, B.P., Hammersley, D., Liu, X., Plank, G., Vigmond, E., Razavi, R., Young, A., Lamata, P., Bishop, M., Niederer, S.: Developing cardiac digital twin populations powered by machine learning provides electrophysiological ins...

    work page 2025

  16. [16]

    In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

    Ronneberger, O., Fischer, P., Brox, T.: U-Net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. Lecture Notes in Computer Science, vol. 9351, pp. 234–241. Springer (2015). https://doi.org/10.1007/978-3-319-24574-4_28

    doi:10.1007/978-3-319-24574-4_28 2015

  17. [17]

    Circulation Research35(1), 117–126 (1974)

    Suga, H., Sagawa, K.: Instantaneous pressure-volume relationships and their ratio in the excised, supported canine left ventricle. Circulation Research35(1), 117–126 (1974). https://doi.org/10.1161/01.RES.35.1.117

    work page doi:10.1161/01.res.35.1.117 1974

  18. [18]

    In: Advances in Neural Information Processing Systems

    Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In: Advances in Neural Information Processing Systems. vol. 30, pp. 1195–1204 (2017)

    work page 2017

  19. [19]

    Springer (2005)

    Vovk, V., Gammerman, A., Shafer, G.: Algorithmic Learning in a Random World. Springer (2005). https://doi.org/10.1007/b106715

    doi:10.1007/b106715 2005

  20. [20]

    In: Medical Image Computing and Computer-Assisted Intervention – MICCAI

    Wei, H., Cao, H., Cao, Y., Zhou, Y., Xue, W., Ni, D., Li, S.: Temporal-consistent segmentation of echocardiography with co-learning from appearance and shape. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI

    work page

  21. [21]

    Echocardiography31(1), 87–100 (2014)

    Wood, P.W., Choy, J.B., Nanda, N.C., Becher, H.: Left ventricular ejection fraction and volumes: It depends on the imaging method. Echocardiography31(1), 87–100 (2014). https://doi.org/10.1111/echo.12331

    work page doi:10.1111/echo.12331 2014

  22. [2020]

    12264, pp

    Lecture Notes in Computer Science, vol. 12264, pp. 623–632. Springer (2020). https://doi.org/10.1007/978-3-030-59713-9_60

    work page doi:10.1007/978-3-030-59713-9_60 2020

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.