Pith. sign in

REVIEW 4 major objections 4 minor 45 references

COF claims that one ECG cardiac cycle plus a single MRI volume can synthesize a patient-specific 4D cardiac cine.

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 · deepseek-v4-flash

2026-08-02 20:33 UTC pith:MI7HGRLQ

load-bearing objection The pipeline is plausible and the numbers are strong, but the headline claim of reconstructing 4D hearts directly from ECG is contradicted by the method's dependence on a static anatomical anchor and by the paper's own limitation statement. the 4 major comments →

arxiv 2602.22919 v2 pith:MI7HGRLQ submitted 2026-02-26 cs.CV

Chain of Flow: ECG-Conditioned 4D Cardiac Cine Generation from Patient-Specific Anatomical Anchor

Haofan Wu , Nay Aung , Theodoros N. Arvanitis , Joao A. C. Lima , Steffen E. Petersen , Le Zhang This is my paper
classification cs.CV
keywords cardiac digital twinECG-conditioned generation4D cardiac cineflow matchingdeformable registrationcardiac motion modelinggenerative modelcine CMR synthesis
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 sets out to show that a patient-specific, temporally coherent 4D cardiac cine — a moving 3D heart — can be generated from a single cardiac cycle of 12-lead ECG, provided one static anatomical MRI volume is available as an anchor. To do this, it learns a continuous velocity field conditioned on both the ECG and the anchor anatomy, then warps the anchor volume along the integrated field to produce every cardiac phase. The authors position this as a step from task-specific cardiac predictors toward fully generative, patient-specific virtual hearts that support volumetry, regional function analysis, and virtual cine synthesis. On a large population cohort, they report high image fidelity and stable downstream functional indices across disease categories, while explicitly conceding that the reconstruction depends fundamentally on the static anchor frame.

Core claim

On its own terms, Chain of Flow (COF) is a two-stage generative framework. First, a topology-preserving volumetric registration module estimates dense deformation fields between cardiac phases of cine MRI and condenses them into a continuous spatiotemporal velocity field via time integration. Second, an ECG-conditioned flow-matching model learns to predict that velocity field from a single R-R interval of 12-lead ECG together with a reference anatomy volume. At inference the predicted velocity field is integrated with an ODE solver to produce a deformation field, which is applied as a spatial warp to the reference volume to synthesize every phase of the cardiac cycle. The paper claims this p

What carries the argument

The central object is the ECG-and-anatomy-conditioned velocity field vθ(x, t, c), defined on space x, pseudo-time t, and conditioning c that jointly encodes the cropped 12-lead ECG segment and the reference anatomy volume. The framework is carried by a chain: registration-derived deformation samples (from the TOPPR module) are used to supervise a flow-matching objective that predicts this velocity field; at inference the field is integrated via an ODE solver to obtain a warp û(x, t); and the generated 4D volume is the reference volume warped along that field, Ĵ(x, t) = Vs(x + û(x, t)). This warp-based formulation is what separates static anatomy from temporal dynamics and preserves topology.

Load-bearing premise

The load-bearing premise is that a single cropped 12-lead ECG beat plus one static MRI volume is sufficient to determine the subject-specific deformation field that the registration module extracts from the target cine frames — a premise the paper's own ECG ablation (motion correlation 0.512 without ECG vs 0.474 with) leaves unsupported.

What would settle it

Hold one reference volume fixed and feed the model ECGs from different patients (matched for heart rate); if the generated cines are statistically indistinguishable across ECGs, then ECG is not carrying patient-specific dynamic information. Alternatively, run the published pipeline with the ECG conditioning disabled on a held-out test set and compare motion correlation and volume-time curve fidelity to the full model; if the anchor-only variant matches or exceeds COF, the 'ECG-to-4D' framing is falsified.

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

If this is right

  • If correct, a single static MRI frame plus one ECG beat could generate a complete cine loop, potentially reducing the need for full cine acquisition in serial assessment.
  • Generated volumes remain segmentable and support standard pipelines: LV/RV/myocardium Dice of 0.87/0.74/0.85 on the reported cohort, and stable EDV, ESV, EF, SV, and CO correlations across disease categories.
  • Stability across slice positions and acquisition resolutions suggests the method can generalize to heterogeneous clinical MRI protocols.
  • By disentangling anatomy from motion, the model provides a manipulable 4D heart suitable for virtual cine simulation and regional function analysis.

Where Pith is reading between the lines

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

  • The paper's own ablation undercuts the strongest reading of its claim: removing the ECG temporal embedding leaves motion correlation at 0.512 versus 0.474 for the full model, so the ECG conditioning is not demonstrated to carry the dynamic signal on this benchmark. A head-to-head anchor-only baseline is needed to establish what ECG actually contributes.
  • A sharper causal test — holding the anchor volume fixed and swapping ECGs from different subjects, then measuring whether the synthesized motion varies accordingly — would settle whether the ECG is a generative driver or a weak regularizer.
  • Because the framework requires a static anatomical anchor, the practical promise is better framed as 'cine from one MRI frame plus ECG' rather than 'cine from ECG alone'; the bottleneck is obtaining and aligning that anchor in settings where cine MRI is unavailable.
  • The paper identifies beat-to-beat variability and continuous ECG streams as future work; extending the single R-R input to multi-cycle ECG would be needed for arrhythmic patients.

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

4 major / 4 minor

Summary. The paper proposes Chain of Flow (COF), a two-stage framework for ECG-conditioned 4D cardiac cine generation. In the first stage, a topology-preserving registration module estimates deformation fields between pairs of cardiac-phase volumes from the same subject. In the second stage, a flow-matching network is trained to predict a continuous velocity field conditioned on a 12-lead ECG segment and a reference-anatomy embedding, supervised by the registration-derived deformations. At inference, the velocity field is integrated with an ODE solver and the resulting displacement field is applied to a static reference volume (Eq. 11) to synthesize a temporally coherent 4D cine sequence. Experiments on the UK Biobank report state-of-the-art image metrics (SSIM 0.984, PSNR 28.46, FID 6.39, FVD 17.60, M-Corr. 0.474, M-SSIM 0.894), strong segmentation Dice/IoU on generated volumes, and stable functional correlations across ECG-derived disease categories, with additional slice-wise, resolution-wise, and qualitative case analyses.

Significance. If the central claim were fully supported, this would be a notable advance: a framework that turns a static anatomical volume plus ECG into a full 4D cardiac twin with clinically meaningful downstream measurements. The paper has clear strengths: the explicit separation of static anatomy from temporal dynamics, the use of registration-derived motion priors with a segmentation-guided loss, careful multi-slice/multi-resolution evaluations, and bootstrap-based functional consistency checks. The code-release promise and the breadth of the evaluation are positive. However, the strongest advertised claim—'full 4D organ reconstruction directly from ECG'—is not supported by the method as described. The method is anchor-conditioned, and the Discussion explicitly concedes that the model cannot infer subject-specific geometry without a static anatomical anchor. This is a load-bearing mismatch that must be resolved before the paper can be accepted.

major comments (4)
  1. [Abstract / III-C / V] The full-text abstract states that COF 'reconstructs full 4D cardiac structure and motion from a single cardiac cycle' and enables 'full 4D organ reconstruction directly from ECG.' The method in Eq. (11) generates J(x,t) = W(Vs, û(x,t)), i.e., it warps a static reference volume Vs, and the conditioning vector c in Eq. (8) includes a reference-anatomy embedding c_rea. The Discussion (Section V) explicitly states: 'the reconstruction process fundamentally depends on a static anatomical anchor frame. Without this anchor, the model cannot reliably infer subject-specific geometry.' This is a direct contradiction of the ECG-only framing, and it undermines the stated clinical motivation of reconstructing hearts where CMR is unavailable. The arXiv metadata abstract is more measured ('patient-specific MRI and current ECG'), so the internal inconsistency is avoidable. Please either re-frame the co
  2. [IV-A / Tables I-II] No train/validation/test split is specified. The paper reports evaluation on 'approximately 10,000 subjects' with a single set of point estimates in Tables I and II and no variances. This is especially important because the supervision signal v* in Eq. (9) is derived from registering the same target volumes that are later used as ground truth for evaluation. If test subjects overlap with training subjects, the reported SSIM/Dice numbers could include memorization effects. Please specify the exact data split, report metrics with standard deviations or confidence intervals (e.g., across subjects or folds), and clarify whether the registration pseudo-labels used for training come from the same subjects used for evaluation.
  3. [Table III / IV-C] The ablation evidence for the contribution of temporal ECG is weaker than claimed. The row 'w/o Temp. ECG' is in the upper block where 'COF (no seg loss)' is the baseline, not the final λ_seg = 1.0 model. Under that correct comparison, removing the temporal ECG embedding changes M-Corr. from 0.548 to 0.512 and M-SSIM from 0.905 to 0.896—a small difference with no significance test. This does not convincingly demonstrate that the ECG carries unique dynamic information beyond the anchor. Conversely, removing the reference anatomy ('w/o REA') lowers M-Corr. to 0.226, highlighting that the anchor is the dominant conditioning signal. Please add a controlled experiment that isolates the ECG contribution (e.g., permuted or mismatched ECG, or a different-subject anchor) and report significance statistics.
  4. [III-A / IV-C] Two central components are used but never defined: the 'TOPPR' module (mentioned in 'Implementation details' and Table III) and the 'maximum-field constraint' (mentioned in Section IV-C and Table III). The maximum-field constraint is ablated and its removal causes a marked degradation, yet no equation or formal description of it appears in the methodology. Similarly, TOPPR is never expanded. Without precise definitions, the ablation results are not reproducible and the technical contribution cannot be fully assessed. Please define these terms formally in Section III or remove/replace them with explicit mathematical descriptions.
minor comments (4)
  1. [Abstract (metadata vs full text)] The arXiv metadata abstract mentions 'controlled phase-robustness analyses' and a 'cross-visit route' with 'serial evidence,' but these experiments do not appear in the full text. The two abstracts also differ in the framing of ECG-only versus anchor-conditioned generation. Please align the metadata abstract with the actual content and findings of the manuscript.
  2. [IV-A] The ECG preprocessing is described only briefly: R-peaks detected with NeuroKit2 and a representative R-R interval selected. It is not stated how leads are combined, whether the ECG is normalized, how the 500 Hz signal is downsampled or embedded, or how the R-R interval is aligned to the 50 cine frames. More detail is needed for reproducibility.
  3. [Section III-B] The conditioning embedding c is described as jointly encoding c_ecg and c_rea, but the exact architecture and fusion point are not shown. A figure or pseudo-code of the conditioning mechanism would help.
  4. [Fig. 5/8] Slice-wise heatmaps and resolution-wise scatter plots report means but no error bars or confidence intervals. Given the apparent stability claims, showing variability would be more convincing.

Circularity Check

2 steps flagged

The 'directly from ECG' reconstruction is, by Eq. 11, a warp of an input CMR anchor, and the segmentation/function metrics are produced by the same nnU-Net teacher that supplies the training labels.

specific steps
  1. other [Eq. (4); Sec. IV-A.1; Sec. IV-B; Sec. IV-E]
    "To obtain anatomical supervision at the organ level, 800 short-axis images were manually annotated ... and used to train a nnU-Net model. The trained network achieved a Dice score of 0.99 ... and was subsequently applied to infer voxel-wise labels for all frames in each cine sequence, producing complete 4D segmentations. ... An nnU-Net model is applied to the generated sequences, and its predictions are compared against ground-truth 4D segmentation volumes."

    The frozen nnU-Net S_phi supplies the hard labels Y_teach in Eq. (4) and also produces the 'ground-truth' segmentations used for evaluation. The model is explicitly trained to make S_phi(warped volume) agree with S_phi(target volume), so the reported Dice/IoU measure agreement with the same oracle on both sides, not independent anatomical truth. Functional indices in Sec. IV-E are also computed by applying 'the same nnU-Net' to real and generated volumes, making those correlations partly self-consistency with the training teacher.

  2. self definitional [Abstract; Eq. (11); Sec. V]
    "COF reconstructs full 4D cardiac structure and motion from a single cardiac cycle. ... enabling full 4D organ reconstruction directly from ECG. ... J_hat(x,t) = W(V_s, u_hat(x,t)) = V_s(x+u_hat(x,t)). ... the reconstruction process fundamentally depends on a static anatomical anchor frame. Without this anchor, the model cannot reliably infer subject-specific geometry."

    The output 4D volume is defined as a warp of the input anatomical anchor V_s: J_hat = W(V_s, u_hat). Therefore the patient-specific structure in the reconstruction is supplied by the CMR anchor by construction, and the ECG only conditions the deformation field. The abstract's claim of 'full 4D organ reconstruction directly from ECG' is thus not derivable from the paper's own inference equation; the Discussion's limitation statement makes this explicit.

full rationale

COF's core image-generation pipeline is a two-stage supervised learning system: registration estimates deformation fields, flow matching learns to predict those fields from ECG plus a reference volume, and ODE integration produces the warped output. This is not, by itself, circular if a proper train/test split is used. However, two load-bearing parts of the paper reduce to self-referential inputs. First, the segmentation teacher nnU-Net is used both to create the hard training labels in Eq. (4) and to generate the 'ground-truth' segmentations for Tables II and the functional analyses, so the reported anatomical and function metrics partly measure agreement with the model's own teacher rather than independent ground truth. Second, the claimed ECG-only reconstruction is contradicted by Eq. (11), where the generated volume is explicitly the input anchor V_s warped by a learned deformation; the Discussion concedes that without this anchor subject-specific geometry cannot be inferred. The SSIM/PSNR/FID/FVD image metrics remain externally computable and give some independent support, but the central 'directly from ECG' anatomy claim and the segmentation-based validation are partially circular by construction. Thus a score of 5, rather than 0, is appropriate.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

No new physical entities are introduced; COF, TOPPR, and REA are software components, not ontological additions. The central claim rests on fitted network weights, a few hand-chosen loss weights, and strong domain assumptions about ECG informativeness, registration validity, and the nnU-Net teacher's correctness.

free parameters (5)
  • λ_seg (segmentation loss weight) = 1.0 (main); 0.5/3.0 in ablation
    Weight in Eq. (5), chosen by hand; affects anatomical fidelity metrics and is central to the segmentation-guided training.
  • λ_rec (reconstruction loss weight)
    Appears in Eq. (5) but its value is not reported, so the optimization balance is underspecified.
  • Slice exclusion threshold = 25% of peak myocardium area
    Post-hoc data-driven cutoff in slice-wise analysis (Sec. IV-D1); affects which slices contribute to the reported metrics.
  • Resolution bin count N
    Used in resolution-wise partition (Sec. IV-D2); N is not specified, so the binned resolution analysis is not reproducible.
  • Model parameters of registration and flow matching networks = Not enumerated
    All weights are fit to UK Biobank data via Adam; without code and data split these are not independently reproducible.
axioms (6)
  • domain assumption The R-R cropped, resampled 12-lead ECG segment contains sufficient subject-specific information to predict cardiac deformation fields (v*)
    Used to train vθ via Eq. (9); if ECG is uninformative, the model reduces to anatomy-driven interpolation; the w/o Temp. ECG ablation (Table III) does not clearly show a motion-correlation benefit.
  • domain assumption nnU-Net segmentation teacher Sφ, trained on 800 manual images, produces correct LV/RV/Myo hard labels for all frames and is a valid oracle for both training and evaluation
    Reported Dice 0.99 on held-out test, but no independent anatomical validation; same teacher used as training loss (Eq. 4) and evaluation metric (Table II) creates overlap.
  • domain assumption TOPPR registration deformation fields are topology-preserving, physiologically valid motion priors
    No direct validation against true motion or independent landmarks; these pseudo-labels are the only motion supervision for the flow model.
  • standard math Flow matching and ODE integration reconstruct continuous cardiac motion from the learned velocity field
    Equations (9)-(11) rely on standard flow matching and ODE theory.
  • ad hoc to paper The 'maximum-field constraint' is a valid regulariser
    Mentioned only in Sec. IV-C ablation; never defined in the methodology, so its role is unspecified.
  • domain assumption Canonical cardiac cycle alignment, using a single representative R-R interval, represents the full cardiac cycle
    The paper aligns ECG and cine to a canonical cycle; beat-to-beat variability is explicitly not modeled, per the Discussion.

pith-pipeline@v1.3.0-alltime-deepseek · 14324 in / 17452 out tokens · 168700 ms · 2026-08-02T20:33:21.635621+00:00 · methodology

0 comments
read the original abstract

Cardiac cine magnetic resonance imaging (MRI) is central to functional cardiac assessment, yet a full current cine sequence may not always be directly available at the point of analysis. We introduce Chain of Flow (COF), an electrocardiography (ECG)-conditioned framework that combines patient-specific MRI and current ECG for subject-specific 4D cardiac cine generation. On the UK Biobank dataset, COF achieves strong image-level fidelity and downstream function-oriented performance on a shared same-visit evaluable benchmark. Multi-slice and multi-resolution analyses indicate stable structural generation quality across the short-axis stack and heterogeneous acquisition resolutions. Controlled phase-robustness analyses across resampled input MRI phases further provide same-visit proxy support for patient-specific MRI plus current ECG when a target MRI phase is not directly observed. A cross-visit route provides exploratory serial evidence, with the clearest gains in current-facing region-of-interest readout. Disease-category functional audits, case-level volume-trajectory evidence review further delineate where the current patient-specific MRI plus ECG formulation remains stable for anatomy-aware downstream cardiac analysis. Code is available at https://anonymous.4open.science/r/COF-paper-release-C88B.

Figures

Figures reproduced from arXiv: 2602.22919 by Haofan Wu, Joao A. C. Lima, Le Zhang, Nay Aung, Steffen E. Petersen, Theodoros N. Arvanitis.

Figure 1
Figure 1. Figure 1: Overview of the COF for ECG-driven cardiac digital twin generation. The framework integrates CMR volumes and 12-lead ECG during training to learn a unified representation of cardiac anatomy, electrophysiology, and motion dynamics. Cine-CMR provides anatomical geometry and motion supervision, while ECG supplies subject-specific electrophysiological dynamics over a single cardiac cycle. These multimodal sign… view at source ↗
Figure 2
Figure 2. Figure 2: Algorithmic pipeline of physiological motion modelling and ECG-conditioned digital heart construction. The pipeline starts with a topology￾preserving volumetric registration module that estimates 3D displacement fields between source and target CMR volumes at different cardiac phases. Discrete deformation samples obtained from registration are parameterised as a continuous spatiotemporal velocity field. A … view at source ↗
Figure 4
Figure 4. Figure 4: (a) Radar-based comprehensive comparison across six image￾level metrics (SSIM, PSNR, FID, FVD, M-Corr., and M-SSIM), where all indicators are normalised so that larger area indicates better overall performance. (b) Segmentation-based comparison using per-frame Dice and IoU for the left ventricle, right ventricle, and myocardium. accurate anatomical boundaries and spatial relationships. C. Cross-Modal Physi… view at source ↗
Figure 5
Figure 5. Figure 5: Per-slice segmentation accuracy across cardiac phases. Per-slice Dice, HD95, and IoU are reported at the end-diastolic and end-systolic phases. Slices are ordered by slice rank from basal to apical (slice ranks 2–7). For each phase, metrics are computed separately for the left ventricle, right ventricle, and myocardium. Dice and IoU heatmaps (a, b, e, f) use a normalised scale from 0 to 1, while HD95 heatm… view at source ↗
Figure 6
Figure 6. Figure 6: Category-wise bootstrap distributions of functional agreement. Horizontal boxplots illustrate the distribution of Pearson correlation coefficients between generated and reference functional indices within each ECG diagnostic category. Panels report correlations for EDV, ESV, SV, EF, and CO estimated via bootstrap resampling (1,000 replicates per category), and the per-case Pearson correlation of LV volume-… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative functional consistency in representative disease cases. Representative cases from hypertrophy and ischaemia/infarction groups are shown. For each case, the top row displays mid-ventricular short-axis CMR slices at end-diastole and end-systole, overlaid with left-ventricular contours from the real annotations and the generated results. The bottom row shows the corresponding LV volume–time curves… view at source ↗
Figure 8
Figure 8. Figure 8: Resolution-wise robustness analysis. Dice (top), IoU (middle), and HD95 (bottom) are plotted against the in-plane pixel spacing sx (mm) for each case. Each point denotes one subject, with separate markers for LV, RV, and myocardium. Solid lines show linear trend fits over all cases for each structure. Large outlined markers indicate the mean metric value within each spacing bin. rhythm, conduction, and rep… 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

45 extracted references · 2 linked inside Pith

  1. [1]

    Global burden of cardiovascular diseases and risk factors, 1990–2019: update from the gbd 2019 study,

    G. A. Roth, G. A. Mensah, C. O. Johnson, G. Addolorato, E. Ammirati, L. M. Baddour, N. C. Barengo, A. Z. Beaton, E. J. Benjamin, C. P. Benzigeret al., “Global burden of cardiovascular diseases and risk factors, 1990–2019: update from the gbd 2019 study,”Journal of the American college of cardiology, vol. 76, no. 25, pp. 2982–3021, 2020

    1990

  2. [2]

    Standardized cardiovascular magnetic resonance imaging (cmr) proto- cols, society for cardiovascular magnetic resonance: board of trustees task force on standardized protocols,

    C. M. Kramer, J. Barkhausen, S. D. Flamm, R. J. Kim, and E. Nagel, “Standardized cardiovascular magnetic resonance imaging (cmr) proto- cols, society for cardiovascular magnetic resonance: board of trustees task force on standardized protocols,”Journal of Cardiovascular Mag- netic Resonance, vol. 10, no. 1, p. 35, 2008

    2008

  3. [3]

    Assessment of left ventricular function by echocardiography: the case for routinely adding global longitudinal strain to ejection fraction,

    E. Potter and T. H. Marwick, “Assessment of left ventricular function by echocardiography: the case for routinely adding global longitudinal strain to ejection fraction,”JACC: Cardiovascular Imaging, vol. 11, no. 2 Part 1, pp. 260–274, 2018

    2018

  4. [4]

    A. H. A. W. G. on Myocardial Segmentation, R. for Cardiac Imaging:, M. D. Cerqueira, N. J. Weissman, V . Dilsizian, A. K. Jacobs, S. Kaul, W. K. Laskey, D. J. Pennell, J. A. Rumberger, T. Ryanet al., “Stan- dardized myocardial segmentation and nomenclature for tomographic imaging of the heart: a statement for healthcare professionals from the cardiac imag...

    2002

  5. [5]

    The health digital twin to tackle cardiovascular disease—a review of an emerging interdisciplinary field,

    G. Coorey, G. A. Figtree, D. F. Fletcher, V . J. Snelson, S. T. Vernon, D. Winlaw, S. M. Grieve, A. McEwan, J. Y . H. Yang, P. Qianet al., “The health digital twin to tackle cardiovascular disease—a review of an emerging interdisciplinary field,”NPJ digital medicine, vol. 5, no. 1, p. 126, 2022

    2022

  6. [6]

    The ‘digital twin’to enable the vision of precision cardiology,

    J. Corral-Acero, F. Margara, M. Marciniak, C. Rodero, F. Loncaric, Y . Feng, A. Gilbert, J. F. Fernandes, H. A. Bukhari, A. Wajdan et al., “The ‘digital twin’to enable the vision of precision cardiology,” European heart journal, vol. 41, no. 48, pp. 4556–4564, 2020

    2020

  7. [7]

    Patient-specific cardiovascular compu- tational modeling: diversity of personalization and challenges,

    R. A. Gray and P. Pathmanathan, “Patient-specific cardiovascular compu- tational modeling: diversity of personalization and challenges,”Journal of cardiovascular translational research, vol. 11, no. 2, pp. 80–88, 2018. 10 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. XX, NO. XX, XXXX 2020

    2018

  8. [8]

    Ref- erence ranges for cardiac structure and function using cardiovascular magnetic resonance (cmr) in caucasians from the uk biobank population cohort,

    S. E. Petersen, N. Aung, M. M. Sanghvi, F. Zemrak, K. Fung, J. M. Paiva, J. M. Francis, M. Y . Khanji, E. Lukaschuk, A. M. Leeet al., “Ref- erence ranges for cardiac structure and function using cardiovascular magnetic resonance (cmr) in caucasians from the uk biobank population cohort,”Journal of cardiovascular magnetic resonance, vol. 19, no. 1, p. 18, 2016

    2016

  9. [9]

    Barriers to cardiovascular magnetic resonance imaging scan performance and reporting by cardiol- ogists: a systematic literature review,

    T. Betemariam, A. Aleka, E. Ahmed, T. Worku, Y . Mebrahtu, E. An- droulakis, S. E. Petersen, and R. Friebel, “Barriers to cardiovascular magnetic resonance imaging scan performance and reporting by cardiol- ogists: a systematic literature review,”European Heart Journal-Imaging Methods and Practice, vol. 3, no. 1, p. qyaf010, 2025

    2025

  10. [10]

    An artificial intelligence-enabled ecg algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction,

    Z. I. Attia, P. A. Noseworthy, F. Lopez-Jimenez, S. J. Asirvatham, A. J. Deshmukh, B. J. Gersh, R. E. Carter, X. Yao, A. A. Rabinstein, B. J. Ericksonet al., “An artificial intelligence-enabled ecg algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction,”The Lancet, vol. 394...

    2019

  11. [11]

    Noninvasive electrocardiographic imaging,

    Y . Rudy and J. E. Burnes, “Noninvasive electrocardiographic imaging,” Annals of Noninvasive electrocardiology, vol. 4, no. 3, pp. 340–359, 1999

    1999

  12. [12]

    Noninvasive electrocardiographic imaging: reconstruction of epicardial potentials, electrograms, and isochrones and localization of single and multiple electrocardiac events,

    H. S. Oster, B. Taccardi, R. L. Lux, P. R. Ershler, and Y . Rudy, “Noninvasive electrocardiographic imaging: reconstruction of epicardial potentials, electrograms, and isochrones and localization of single and multiple electrocardiac events,”Circulation, vol. 96, no. 3, pp. 1012– 1024, 1997

    1997

  13. [13]

    N. G. Bellenger, M. I. Burgess, S. G. Ray, A. Lahiri, A. J. Coats, J. G. Cleland, and D. J. Pennell, “Comparison of left ventricular ejection fraction and volumes in heart failure by echocardiography, radionuclide ventriculography and cardiovascular magnetic resonance. are they interchangeable?”European heart journal, vol. 21, no. 16, pp. 1387–1396, 2000

    2000

  14. [14]

    R. M. Lang, L. P. Badano, V . Mor-Avi, J. Afilalo, A. Armstrong, L. Ernande, F. A. Flachskampf, E. Foster, S. A. Goldstein, T. Kuznetsova et al., “Recommendations for cardiac chamber quantification by echocar- diography in adults: an update from the american society of echocar- diography and the european association of cardiovascular imaging,” European He...

    2015

  15. [15]

    Computational cardiology: the heart of the matter,

    N. A. Trayanova, “Computational cardiology: the heart of the matter,” International Scholarly Research Notices, vol. 2012, no. 1, p. 269680, 2012

    2012

  16. [16]

    Deep learning and the electrocardiogram: review of the current state-of-the-art,

    S. Somani, A. J. Russak, F. Richter, S. Zhao, A. Vaid, F. Chaudhry, J. K. De Freitas, N. Naik, R. Miotto, G. N. Nadkarniet al., “Deep learning and the electrocardiogram: review of the current state-of-the-art,”EP Europace, vol. 23, no. 8, pp. 1179–1191, 2021

    2021

  17. [17]

    Computing ischemic regions in the heart with the bidomain model—first steps towards validation,

    B. F. Nielsen, M. Lysaker, and P. Grøttum, “Computing ischemic regions in the heart with the bidomain model—first steps towards validation,” IEEE Transactions on Medical Imaging, vol. 32, no. 6, pp. 1085–1096, 2013

    2013

  18. [18]

    Recent progress in inverse prob- lems in electrocardiology,

    R. S. MacLeod and D. H. Brooks, “Recent progress in inverse prob- lems in electrocardiology,”IEEE Engineering in Medicine and Biology Magazine, vol. 17, no. 1, pp. 73–83, 1998

    1998

  19. [19]

    Embedding high-dimensional bayesian optimization via generative modeling: parameter personaliza- tion of cardiac electrophysiological models,

    J. Dhamala, P. Bajracharya, H. J. Arevalo, J. L. Sapp, B. M. Hor ´acek, K. C. Wu, N. A. Trayanova, and L. Wang, “Embedding high-dimensional bayesian optimization via generative modeling: parameter personaliza- tion of cardiac electrophysiological models,”Medical image analysis, vol. 62, p. 101670, 2020

    2020

  20. [20]

    Toward enabling cardiac digital twins of myocardial infarction using deep computational models for inverse inference,

    L. Li, J. Camps, Z. J. Wang, M. Beetz, A. Banerjee, B. Rodriguez, and V . Grau, “Toward enabling cardiac digital twins of myocardial infarction using deep computational models for inverse inference,”IEEE transactions on medical imaging, vol. 43, no. 7, pp. 2466–2478, 2024

    2024

  21. [21]

    Solving the inverse problem of electrocardiography for cardiac digital twins: A survey,

    L. Li, J. Camps, B. Rodriguez, and V . Grau, “Solving the inverse problem of electrocardiography for cardiac digital twins: A survey,” IEEE Reviews in Biomedical Engineering, 2024

    2024

  22. [22]

    Physics-constrained deep learning for robust inverse ecg modeling,

    J. Xie and B. Yao, “Physics-constrained deep learning for robust inverse ecg modeling,”IEEE Transactions on Automation Science and Engineering, vol. 20, no. 1, pp. 151–166, 2022

    2022

  23. [23]

    Factorised spatial representation learning: Application in semi-supervised myocardial seg- mentation,

    A. Chartsias, T. Joyce, G. Papanastasiou, S. Semple, M. Williams, D. Newby, R. Dharmakumar, and S. A. Tsaftaris, “Factorised spatial representation learning: Application in semi-supervised myocardial seg- mentation,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2018, pp. 490–498

    2018

  24. [24]

    Generative adversarial network in medical imaging: A review,

    X. Yi, E. Walia, and P. Babyn, “Generative adversarial network in medical imaging: A review,”Medical image analysis, vol. 58, p. 101552, 2019

    2019

  25. [25]

    Deepmesh: Mesh- based cardiac motion tracking using deep learning,

    Q. Meng, W. Bai, D. P. O’Regan, and D. Rueckert, “Deepmesh: Mesh- based cardiac motion tracking using deep learning,”IEEE transactions on medical imaging, vol. 43, no. 4, pp. 1489–1500, 2023

    2023

  26. [26]

    Joint learning of motion estimation and segmentation for cardiac mr image sequences,

    C. Qin, W. Bai, J. Schlemper, S. E. Petersen, S. K. Piechnik, S. Neubauer, and D. Rueckert, “Joint learning of motion estimation and segmentation for cardiac mr image sequences,” inInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention. Springer, 2018, pp. 472–480

    2018

  27. [27]

    The perception-distortion tradeoff,

    Y . Blau and T. Michaeli, “The perception-distortion tradeoff,” inPro- ceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 6228–6237

    2018

  28. [28]

    Echopulse: Ecg controlled echocardio-grams video generation,

    Y . Li, S. Kim, Z. Wu, H. Jiang, Y . Pan, P. Jin, S. Song, Y . Shi, T. Liu, Q. Liet al., “Echopulse: Ecg controlled echocardio-grams video generation,”arXiv preprint arXiv:2410.03143, 2024

    Pith/arXiv arXiv 2024

  29. [29]

    Deep generative model for synthesising and analysing cardiac magnetic resonance images,

    N. Cheng, “Deep generative model for synthesising and analysing cardiac magnetic resonance images,” Ph.D. dissertation, University of Leeds, 2025

    2025

  30. [30]

    Are medical image generative models bi- ologically trustworthy?

    S. Ahn, W. Park, and J. Park, “Are medical image generative models bi- ologically trustworthy?” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 633–643

    2025

  31. [31]

    Ecgflowcmr: Pretraining with ecg-generated cine cmr improves cardiac disease classification and phenotype predic- tion,

    X. Fang, Z. Ding, J. Cai, Y . Xiao, B. Liu, J. Jin, H. Wang, G. Nie, S. Huang, T. Chenet al., “Ecgflowcmr: Pretraining with ecg-generated cine cmr improves cardiac disease classification and phenotype predic- tion,”arXiv preprint arXiv:2601.20904, 2026

    Pith/arXiv arXiv 2026

  32. [32]

    Opportunities and challenges of deep learning methods for electrocardiogram data: A systematic review,

    S. Hong, Y . Zhou, J. Shang, C. Xiao, and J. Sun, “Opportunities and challenges of deep learning methods for electrocardiogram data: A systematic review,”Computers in biology and medicine, vol. 122, p. 103801, 2020

    2020

  33. [33]

    Prospective validation of a deep learning electrocardiogram algorithm for the detection of left ventricular systolic dysfunction,

    Z. I. Attia, S. Kapa, X. Yao, F. Lopez-Jimenez, T. L. Mohan, P. A. Pellikka, R. E. Carter, N. D. Shah, P. A. Friedman, and P. A. Noseworthy, “Prospective validation of a deep learning electrocardiogram algorithm for the detection of left ventricular systolic dysfunction,”Journal of cardiovascular electrophysiology, vol. 30, no. 5, pp. 668–674, 2019

    2019

  34. [34]

    Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network,

    A. Y . Hannun, P. Rajpurkar, M. Haghpanahi, G. H. Tison, C. Bourn, M. P. Turakhia, and A. Y . Ng, “Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network,”Nature medicine, vol. 25, no. 1, pp. 65–69, 2019

    2019

  35. [35]

    Deep learning for ecg arrhythmia detection and classification: an overview of progress for period 2017–2023,

    Y . Ansari, O. Mourad, K. Qaraqe, and E. Serpedin, “Deep learning for ecg arrhythmia detection and classification: an overview of progress for period 2017–2023,”Frontiers in Physiology, vol. 14, p. 1246746, 2023

    2017

  36. [36]

    Value of the electrocardiogram in localizing the occlusion site in the left anterior descending coronary artery in acute anterior myocardial infarction,

    D. J. Engelen, A. P. Gorgels, E. C. Cheriex, E. D. De Muinck, A. J. Oude Ophuis, W. R. Dassen, J. Vainer, V . G. van Ommen, and H. J. Wellens, “Value of the electrocardiogram in localizing the occlusion site in the left anterior descending coronary artery in acute anterior myocardial infarction,”Journal of the American College of Cardiology, vol. 34, no. ...

    1999

  37. [37]

    Acute st-segment elevation myocardial infarction (stemi),

    H. Akbar and S. Mountfort, “Acute st-segment elevation myocardial infarction (stemi),”StatPearls, 2024

    2024

  38. [38]

    The selvester qrs scoring system for estimating myocardial infarct size: the development and application of the system,

    R. H. Selvester, G. S. Wagner, and N. B. Hindman, “The selvester qrs scoring system for estimating myocardial infarct size: the development and application of the system,”Archives of Internal Medicine, vol. 145, no. 10, pp. 1877–1881, 1985

    1985

  39. [39]

    Electrocardiographic imag- ing for cardiac arrhythmias and resynchronization therapy,

    H. Pereira, S. Niederer, and C. A. Rinaldi, “Electrocardiographic imag- ing for cardiac arrhythmias and resynchronization therapy,” 2020

    2020

  40. [40]

    Nonin- vasive electrocardiographic imaging for cardiac electrophysiology and arrhythmia,

    C. Ramanathan, R. N. Ghanem, P. Jia, K. Ryu, and Y . Rudy, “Nonin- vasive electrocardiographic imaging for cardiac electrophysiology and arrhythmia,”Nature medicine, vol. 10, no. 4, pp. 422–428, 2004

    2004

  41. [41]

    Arrhythmia risk stratification of patients after myocardial infarction using personalized heart models,

    H. J. Arevalo, F. Vadakkumpadan, E. Guallar, A. Jebb, P. Malamas, K. C. Wu, and N. A. Trayanova, “Arrhythmia risk stratification of patients after myocardial infarction using personalized heart models,”Nature communications, vol. 7, no. 1, p. 11437, 2016

    2016

  42. [42]

    nnu-net: a self-configuring method for deep learning-based biomedical image segmentation,

    F. Isensee, P. F. Jaeger, S. A. Kohl, J. Petersen, and K. H. Maier-Hein, “nnu-net: a self-configuring method for deep learning-based biomedical image segmentation,”Nature methods, vol. 18, no. 2, pp. 203–211, 2021

    2021

  43. [43]

    Uk biobank’s cardiovascular magnetic resonance protocol,

    S. E. Petersen, P. M. Matthews, J. M. Francis, M. D. Robson, F. Zemrak, R. Boubertakh, A. A. Young, S. Hudson, P. Weale, S. Garrattet al., “Uk biobank’s cardiovascular magnetic resonance protocol,”Journal of cardiovascular magnetic resonance, vol. 18, no. 1, p. 8, 2016

    2016

  44. [44]

    Ti2v-zero: Zero-shot image conditioning for text-to-video diffusion models,

    H. Ni, B. Egger, S. Lohit, A. Cherian, Y . Wang, T. Koike-Akino, S. X. Huang, and T. K. Marks, “Ti2v-zero: Zero-shot image conditioning for text-to-video diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 9015– 9025

    2024

  45. [45]

    Feature-conditioned cascaded video diffusion models for precise echocardiogram synthesis,

    H. Reynaud, M. Qiao, M. Dombrowski, T. Day, R. Razavi, A. Gomez, P. Leeson, and B. Kainz, “Feature-conditioned cascaded video diffusion models for precise echocardiogram synthesis,” inInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention. Springer, 2023, pp. 142–152

    2023