Fine-tuning an ECG Foundation Model to Predict Coronary CT Angiography Outcomes

Deyun Zhang; Gongzheng Tang; Guangkun Nie; Haoyu Wang; Hao Zhang; Jian Liu; Jun Li; Kangyin Chen; Qinghao Zhao; Shenda Hong

arxiv: 2512.05136 · v3 · pith:Z6S5AAPHnew · submitted 2025-11-29 · 💻 cs.CV · cs.AI

Fine-tuning an ECG Foundation Model to Predict Coronary CT Angiography Outcomes

Yujie Xiao , Qinghao Zhao , Gongzheng Tang , Hao Zhang , Zhuoran Kan , Deyun Zhang , Jun Li , Guangkun Nie

show 7 more authors

Xiaocheng Fang Haoyu Wang Shun Huang Tong Liu Jian Liu Kangyin Chen Shenda Hong

This is my paper

Pith reviewed 2026-05-21 19:10 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords AI-ECGCoronary artery diseaseCCTARisk stratificationFoundation modelStenosis predictionClinical triagePre-test probability

0 comments

The pith

ECG AI model predicts vessel-specific coronary stenosis on CT angiography and improves risk stratification over pre-test probability alone

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

This multicenter study fine-tunes an ECG foundation model to predict vessel-specific coronary artery stenosis using CCTA as the anatomical reference standard. The model achieves AUCs of 0.683-0.744 internally with consistent external results, and its output probabilities rise steadily with increasing stenosis severity. When these probabilities are turned into low-, intermediate-, and high-risk strata, combining them with guideline pre-test probability categories raises rule-out rates, shrinks the gray zone, and produces positive net reclassification improvement. Longitudinal follow-up links the strata to separated major adverse cardiovascular event risks, while waveform analysis identifies physiologically relevant ECG features.

Core claim

Fine-tuning an ECG foundation model on paired ECG-CCTA data yields per-vessel stenosis probability estimates that discriminate moderate-to-severe lesions, calibrate to observed risk, deliver net clinical benefit on decision-curve analysis, and add incremental value to pre-test probability assessment through improved rule-out performance and reduced gray-zone cases.

What carries the argument

Vessel-specific stenosis probability outputs from the fine-tuned ECG foundation model, thresholded into low-, intermediate-, and high-risk strata using sensitivity- and specificity-based cutoffs.

If this is right

Integration with PTP improves rule-out performance and reduces the gray-zone proportion compared with PTP alone.
Model-defined risk groups show clear separation of major adverse cardiovascular event risk in longitudinal follow-up.
Decision-curve analysis indicates net clinical benefit over treat-all and treat-none strategies.
Waveform- and attribution-based analyses identify structured ECG morphology differences tied to high-risk predictions.

Where Pith is reading between the lines

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

If confirmed in prospective trials, the approach could reduce the number of low-yield CCTA scans by providing an initial ECG-based filter.
The same fine-tuning strategy might be applied to predict other CCTA-derived outcomes such as plaque burden or functional ischemia.
Feature attribution on the ECG waveforms could surface new candidate electrical biomarkers for coronary disease that merit targeted physiologic study.

Load-bearing premise

CCTA serves as an accurate and unbiased reference standard for vessel-specific stenosis with no systematic differences in image quality or reader interpretation across centers.

What would settle it

An external validation set in which the model's predicted probabilities show no monotonic relationship with CCTA-defined stenosis severity or produce zero or negative net reclassification improvement when added to pre-test probability.

Figures

Figures reproduced from arXiv: 2512.05136 by Deyun Zhang, Gongzheng Tang, Guangkun Nie, Haoyu Wang, Hao Zhang, Jian Liu, Jun Li, Kangyin Chen, Qinghao Zhao, Shenda Hong, Shun Huang, Tong Liu, Xiaocheng Fang, Yujie Xiao, Zhuoran Kan.

**Figure 2.** Figure 2: ROC analyses at the CCTA examination and patient levels. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Subgroup analyses of model performance. Forest plots show the AUC with 95% CI across clinically relevant subgroups, including age, sex, time from symptom onset, and medical history. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Distribution of model-predicted probabilities across different degrees of coronary stenosis. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Calibration curve of model prediction results. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 6.** Figure 6: Kaplan–Meier curves stratified by model-predicted risk across major coronary arteries. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Decision curve analysis for model-guided coronary risk stratification. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Explainable analysis results. The results reveal the key ECG regions that the model focuses on during prediction and stratification. Red lines indicate individuals classified as the high-risk group, whereas blue lines indicate individuals classified as the low-risk group. Discussion 235 This study constructed and validated an ECG-based artificial intelligence model for predicting 236 severe stenosis and to… view at source ↗

read the original abstract

CAD remains a major global public health burden, yet scalable screening tools are limited. Although CCTA is a first-line non-invasive diagnostic modality, its use is constrained by resource requirements and radiation exposure. AI-ECG may offer a complementary approach for CAD risk stratification. In this multicenter study, we developed and validated an AI-ECG model using CCTA as the anatomical reference standard to predict vessel-specific coronary stenosis. In internal validation, the model achieved AUC values of 0.683-0.744 across vessels and showed consistent external performance. Discrimination was maintained in clinically normal ECGs and remained broadly stable across subgroups. Model-predicted probabilities increased monotonically with CCTA-defined stenosis severity. Model probabilities were converted into vessel-specific low-, intermediate-, and high-risk strata using predefined sensitivity- and specificity-based thresholds. Calibration analysis showed agreement between predicted and observed risk, while DCA indicated net clinical benefit over treat-all and treat-none strategies. Integrating AI-derived risk strata with guideline-based PTP categories improved rule-out performance, reduced the gray-zone proportion, and achieved positive NRI compared with PTP alone. In a longitudinal follow-up cohort, Kaplan-Meier analysis showed clear separation of major adverse cardiovascular event risk across model-defined risk groups. Waveform- and attribution-based analyses further identified structured ECG morphology differences and physiologically meaningful signal regions associated with high-risk predictions. These findings support AI-ECG as a feasible tool for complementary CAD screening, anatomical risk estimation, and clinical triage, while prospective studies are needed to confirm its clinical impact.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Fine-tuning an ECG foundation model yields moderate vessel-level AUCs against CCTA and modest NRI gains over PTP, but the gains rest on unexamined consistency of the CCTA labels across centers.

read the letter

Colleague, the main thing to know is that this work takes an existing ECG foundation model, fine-tunes it on multicenter data, and produces vessel-specific stenosis probabilities that rise with CCTA severity and give a small lift when added to pre-test probability scores. The AUC range of 0.68-0.74, external validation, monotonicity check, DCA, and NRI are the concrete outputs, plus a KM split on events in follow-up. That combination is new enough to be worth a look even if it is an application rather than a new method. The paper does a few things cleanly. It moves from patient-level to vessel-level targets, keeps the evaluation on clinically normal ECGs, and reports calibration and net benefit metrics instead of stopping at AUC. The attribution maps pointing to plausible waveform regions are a small plus for interpretability. These elements make the results more usable than a pure black-box claim. The soft spots are mostly in the methods and reference-standard handling. The abstract gives no numbers on training samples per vessel, no description of the fine-tuning schedule or regularization, and no account of missing data or how thresholds for the low-intermediate-high strata were locked before seeing test performance. More critically, the CCTA labels come from routine clinical reads across centers with no mention of standardized protocols, blinded re-review, or adjustment for scanner or reader differences. If image quality or reporting habits vary systematically by site, that variation could leak into the ECG features and inflate the apparent added value over PTP. The stress-test concern about center-specific biases therefore lands on the data that are shown; it is not a minor quibble. This paper is for cardiologists and imaging groups who already work with ECG AI and want numbers on vessel-level prediction plus PTP integration. A reader who needs to decide whether to run a similar fine-tuning experiment will get usable benchmarks and a clear picture of where the current limits sit. It is coherent on its own terms and has enough external checks that it should go to peer review rather than a desk reject, though any referee will rightly press for fuller methods and sensitivity checks on the CCTA reference.

Referee Report

2 major / 2 minor

Summary. The paper develops and externally validates a fine-tuned ECG foundation model to predict vessel-specific coronary stenosis severity using CCTA as the anatomical reference standard. It reports AUCs of 0.683–0.744 internally, maintained performance on normal ECGs and subgroups, monotonic probability-stenosis relationships, calibration, decision-curve net benefit, and improved rule-out plus positive NRI when AI risk strata are integrated with guideline PTP categories; Kaplan-Meier separation for MACE is also shown.

Significance. If the central claims hold, the work offers a scalable, radiation-free complement to CCTA for CAD triage that could meaningfully shrink the gray zone and improve rule-out when combined with pre-test probability. The multicenter design, external validation, DCA, NRI, and longitudinal KM analysis are concrete strengths that would support clinical translation if the reference-standard and threshold issues are resolved.

major comments (2)

[Abstract / Results] Abstract and Results: the claim that AI-derived risk strata improve rule-out, reduce gray-zone proportion, and yield positive NRI rests on vessel-specific low/intermediate/high thresholds that are described as 'predefined' but whose exact derivation (training-set only vs. any use of validation or test data) is not stated; without this, the reported integration metrics cannot be interpreted as fully independent of the evaluation set.
[Methods / Discussion] Methods / Discussion: the manuscript provides no description of CCTA protocol standardization, blinded re-reading, or adjustment for scanner/reader/site variability despite being a multicenter study with external validation. Because every reported metric (AUC, monotonicity, NRI, DCA) is computed against these labels, systematic center-specific differences in image quality or stenosis interpretation constitute a load-bearing threat to the claim that the ECG model has learned genuine vessel-specific stenosis signals rather than site artifacts.

minor comments (2)

[Abstract] Abstract: exact per-vessel sample sizes, handling of missing ECG or CCTA data, and whether thresholds were locked before any test-set evaluation should be stated explicitly.
[Results] Results: the statement that discrimination 'remained broadly stable across subgroups' would benefit from a table or supplementary figure showing AUCs stratified by age, sex, and ECG normality.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these constructive comments, which help clarify key methodological details. We address each point below and have revised the manuscript to improve transparency and acknowledge limitations where appropriate.

read point-by-point responses

Referee: [Abstract / Results] Abstract and Results: the claim that AI-derived risk strata improve rule-out, reduce gray-zone proportion, and yield positive NRI rests on vessel-specific low/intermediate/high thresholds that are described as 'predefined' but whose exact derivation (training-set only vs. any use of validation or test data) is not stated; without this, the reported integration metrics cannot be interpreted as fully independent of the evaluation set.

Authors: We appreciate the referee highlighting the need for explicit detail on threshold derivation. The vessel-specific low-, intermediate-, and high-risk thresholds were determined exclusively from the training set by selecting cutoffs that achieved prespecified sensitivity and specificity targets. No information from the validation or external test sets was used in threshold selection. We have revised the Methods section to state this explicitly and updated the Abstract and Results to ensure the NRI, rule-out, and gray-zone metrics are clearly understood as independent of the evaluation data. revision: yes
Referee: [Methods / Discussion] Methods / Discussion: the manuscript provides no description of CCTA protocol standardization, blinded re-reading, or adjustment for scanner/reader/site variability despite being a multicenter study with external validation. Because every reported metric (AUC, monotonicity, NRI, DCA) is computed against these labels, systematic center-specific differences in image quality or stenosis interpretation constitute a load-bearing threat to the claim that the ECG model has learned genuine vessel-specific stenosis signals rather than site artifacts.

Authors: We acknowledge that the original manuscript lacked sufficient detail on CCTA acquisition and interpretation procedures. All participating sites followed local clinical CCTA protocols consistent with contemporary guidelines, with stenosis severity reported by board-certified cardiologists during routine care. We have now added a Methods subsection describing the CCTA scanners, contrast protocols, and reconstruction parameters used across centers. We have also added a limitation paragraph in the Discussion noting the absence of centralized blinded re-reading and the potential for inter-site variability in stenosis grading. To partially mitigate this concern, we performed post-hoc site-stratified performance checks (now reported in the supplement) and note that external validation on an independent cohort provides some protection against site-specific artifacts. We agree this remains a limitation and that future prospective studies would benefit from standardized core-lab reading. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes a standard supervised fine-tuning of an ECG foundation model to predict CCTA-defined vessel-specific stenosis, with performance evaluated on internal and external validation cohorts. Risk strata are formed from model probabilities using predefined sensitivity- and specificity-based thresholds, and downstream integration with PTP categories is assessed via NRI, DCA, and Kaplan-Meier analysis on held-out data. No equations or steps reduce by construction to the inputs; the central claims rest on empirical discrimination, calibration, and clinical utility metrics against independent labels rather than self-definition or fitted parameters renamed as predictions. External validation and monotonicity checks supply independent grounding.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The work rests on standard supervised learning assumptions plus clinical reference-standard assumptions. No new physical entities are postulated. Free parameters include the risk-strata thresholds and fine-tuning hyperparameters; these are not enumerated in the abstract.

free parameters (2)

risk-strata thresholds
Predefined sensitivity- and specificity-based cutoffs used to convert model probabilities into low/intermediate/high risk groups; chosen to achieve target operating points but details of selection process not given in abstract.
fine-tuning hyperparameters
Learning rate, epochs, batch size, and any regularization used when adapting the foundation model; not reported in abstract.

axioms (2)

domain assumption CCTA provides an accurate and unbiased reference standard for vessel-specific stenosis
The entire training and evaluation pipeline treats CCTA stenosis severity as ground truth; any systematic error in CCTA would propagate directly into model performance claims.
domain assumption ECG recordings and CCTA exams are independent conditional on the underlying anatomy
Standard assumption for supervised learning on paired diagnostic tests; violation would produce over-optimistic AUCs.

pith-pipeline@v0.9.0 · 5846 in / 1629 out tokens · 57571 ms · 2026-05-21T19:10:32.197893+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We employed a transfer learning framework to fine-tune ECGFounder... multi-task deep learning framework... adaptive weighting strategy based on task uncertainty... Projected Conflicting Gradient (PCGrad)
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Model-predicted probabilities increased monotonically with CCTA-defined stenosis severity... vessel-specific low-, intermediate-, and high-risk strata using predefined sensitivity- and specificity-based thresholds

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

[1]

Martin, S.S., Aday, A.W., Almarzooq, Z.I., Anderson, C.A., Arora, P ., Avery, C.L., Baker-Smith,527 C.M., Barone Gibbs, B., Beaton, A.Z., Boehme, A.K. et al. (2024). 2024 heart disease 528 and stroke statistics: a report of us and global data from the american heart association. 529 Circulation149, e347–e913. 530

work page 2024
[2]

Mensah, G.A., Fuster, V., Murray, C.J., Roth, G.A., of Cardiovascular Diseases, G.B., and 531 Collaborators, R. (2023). Global burden of cardiovascular diseases and risks, 1990-2022. 532 Journal of the American College of Cardiology82, 2350–2473.533

work page 2023
[3]

Luengo-Fernandez, R., Walli-Attaei, M., Gray, A., Torbica, A., Maggioni, A.P ., Huculeci, 534 R., Bairami, F ., Aboyans, V., Timmis, A.D., Vardas, P . et al. (2023). Economic burden of 535 cardiovascular diseases in the european union: a population-based cost study. European 536 heart journal44, 4752–4767. 537

work page 2023
[4]

Vrints, C., Andreotti, F ., Koskinas, K.C., Rossello, X., Adamo, M., Ainslie, J., Banning, A.P .,538 Budaj, A., Buechel, R.R., Chiariello, G.A. et al. (2024). 2024 esc guidelines for the man- 539 agement of chronic coronary syndromes: developed by the task force for the management 540 of chronic coronary syndromes of the european society of cardiology (es...

work page 2024
[5]

Kim, C., Park, C.H., Lee, B.Y ., Park, C.H., Kang, E.J., Koo, H.J., Kitagawa, K., Cha, M.J., 544 Krittayaphong, R., Choi, S.I. et al. (2024). 2024 consensus statement on coronary stenosis545 and plaque evaluation in ct angiography from the asian society of cardiovascular imaging- 546 practical tutorial (asci-pt). Korean journal of radiology25, 331.547 19

work page 2024
[6]

Members, W.C., Gulati, M., Levy, P .D., Mukherjee, D., Amsterdam, E., Bhatt, D.L., Birtcher,548 K.K., Blankstein, R., Boyd, J., Bullock-Palmer, R.P . et al. (2021). 2021 aha/acc/ase/ch- 549 est/saem/scct/scmr guideline for the evaluation and diagnosis of chest pain: a report of 550 the american college of cardiology/american heart association joint commit...

work page 2021
[7]

Narula, J., Chandrashekhar, Y ., Ahmadi, A., Abbara, S., Berman, D.S., Blankstein, R., Leipsic,553 J., Newby, D., Nicol, E.D., Nieman, K. et al. (2020). Scct 2021 expert consensus document 554 on coronary computed tomographic angiography: a report of the society of cardiovascular 555 computed tomography. Journal of cardiovascular computed tomography15, 192.556

work page 2020
[8]

Yu, H., Song, H., Sun, X., Song, T., Xie, A., Xu, J., Qin, R., Jing, L., Zuo, T., Zhao, J. et al. 557 (2024). Reduced radiation dose and volume of contrast medium in heart rate-based, one-stop 558 computed tomography angiography of coronary, carotid and cerebrovascular arteries. Acta559 Radiologica65, 84–90. 560

work page 2024
[9]

Jain, S., and Majidi, F . (2025). Ct coronary angiography vs cardiac catheterization and risk of 561 acute kidney injury–comparative study. Journal of Clinical Cardiology6, 84–90.562

work page 2025
[10]

Byrne, R.A., Rossello, X., Coughlan, J., Barbato, E., Berry, C., Chieffo, A., Claeys, M.J., Dan, 563 G.A., Dweck, M.R., Galbraith, M. et al. (2024). 2023 esc guidelines for the management 564 of acute coronary syndromes: developed by the task force on the management of acute 565 coronary syndromes of the european society of cardiology (esc). European Hear...

work page 2024
[11]

Bhatt, D.L., Lopes, R.D., and Harrington, R.A. (2022). Diagnosis and treatment of acute 568 coronary syndromes: a review. Jama327, 662–675. 569

work page 2022
[12]

Thygesen, K., Alpert, J.S., Jaffe, A.S., Chaitman, B.R., Bax, J.J., Morrow, D.A., White, 570 H.D., and on behalf of the Joint European Society of Cardiology (ESC)/American College of 571 Cardiology (ACC)/American Heart Association (AHA)/World Heart Federation (WHF) Task 572 Force for the Universal Definition of Myocardial Infarction, E.G. (2018). Fourth u...

work page 2018
[13]

Zhao, Q., Geng, S., Wang, B., Sun, Y ., Nie, W., Bai, B., Yu, C., Zhang, F ., Tang, G., Zhang, D.576 et al. (2024). Deep learning in heart sound analysis: from techniques to clinical applications. 577 Health Data Science4, 0182. 578

work page 2024
[14]

Luo, Y ., Liu, X.Y ., Y ang, K., Huang, K., Hong, M., Zhang, J., Wu, Y ., and Nie, Z. (2024). 579 Toward unified ai drug discovery with multimodal knowledge. Health Data Science4, 0113.580

work page 2024
[15]

Zhang, S., Mu, W., Dong, D., Wei, J., Fang, M., Shao, L., Zhou, Y ., He, B., Zhang, S., Liu,581 Z. et al. (2023). The applications of artificial intelligence in digestive system neoplasms: a 582 review. Health Data Science3, 0005. 583

work page 2023
[16]

Tham, Y .C., Goh, J.H.L., Anees, A., Lei, X., Rim, T.H., Chee, M.L., Wang, Y .X., Jonas, 584 J.B., Thakur, S., Teo, Z.L. et al. (2022). Detecting visually significant cataract using retinal 585 photograph-based deep learning. Nature aging2, 264–271.586

work page 2022
[17]

et al.587 (2021)

Liu, X., Gao, K., Liu, B., Pan, C., Liang, K., Y an, L., Ma, J., He, F ., Zhang, S., Pan, S. et al.587 (2021). Advances in deep learning-based medical image analysis. Health Data Science 588 2021, 8786793. 589 20

work page 2021
[18]

590 (2025)

Xiao, Y ., Tang, G., Liu, W., Li, J., Nie, G., Kan, Z., Zhang, D., Zhao, Q., and Hong, S. 590 (2025). Anyecg-lab: An exploration study of fine-tuning an ecg foundation model to estimate 591 laboratory values from single-lead ecg signals. arXiv preprint arXiv:2510.22301.592

work page arXiv 2025
[19]

Croon, P .M., Dhingra, L.S., Biswas, D., Oikonomou, E.K., and Khera, R. (2025). Phenotypic 593 selectivity of artificial intelligence-enhanced electrocardiography in cardiovascular diagnosis 594 and risk prediction. Circulation152. 595

work page 2025
[20]

Poterucha, T.J., Jing, L., Ricart, R.P ., Adjei-Mosi, M., Finer, J., Hartzel, D., Kelsey, C., Long, A.,596 Rocha, D., Ruhl, J.A. et al. (2025). Detecting structural heart disease from electrocardiograms 597 using ai. Nature644, 221–230. 598

work page 2025
[21]

Y ang, K., Hong, M., Zhang, J., Luo, Y ., Zhao, S., Zhang, O., Yu, X., Zhou, J., Y ang, L., Zhang,599 P . et al. (2025). Ecg-lm: Understanding electrocardiogram with a large language model. 600 Health Data Science5, 0221. 601

work page 2025
[22]

Al-Zaiti, S.S., Martin-Gill, C., Zègre-Hemsey, J.K., Bouzid, Z., Faramand, Z., Alrawashdeh, 602 M.O., Gregg, R.E., Helman, S., Riek, N.T., Kraevsky-Phillips, K. et al. (2023). Machine 603 learning for ecg diagnosis and risk stratification of occlusion myocardial infarction. Nature 604 Medicine29, 1804–1813. 605

work page 2023
[23]

Lin, C.S., Liu, W.T., Tsai, D.J., Lou, Y .S., Chang, C.H., Lee, C.C., Fang, W.H., Wang, C.C.,606 Chen, Y .Y ., Lin, W.S. et al. (2024). Ai-enabled electrocardiography alert intervention and 607 all-cause mortality: a pragmatic randomized clinical trial. Nature Medicine30, 1461–1470.608

work page 2024
[24]

609 (2023)

Choi, S.H., Lee, H.G., Park, S.D., Bae, J.W., Lee, W., Kim, M.S., Kim, T.H., and Lee, W.K. 609 (2023). Electrocardiogram-based deep learning algorithm for the screening of obstructive 610 coronary artery disease. BMC cardiovascular disorders23, 287.611

work page 2023
[25]

612 (2024)

Park, J., Kim, J., Kang, S.H., Lee, J., Hong, Y ., Chang, H.J., Cho, Y ., and Y oon, Y .E. 612 (2024). Artificial intelligence–enhanced electrocardiography analysis as a promising tool 613 for predicting obstructive coronary artery disease in patients with stable angina. European 614 Heart Journal-Digital Health5, 444–453. 615

work page 2024
[26]

Wang, H., Gao, Z., Zhang, H., Zhu, Y ., Lian, S., Bo, K., Li, S., Gao, Y ., Zhuang, B., Zhou, 616 Z. et al. (2025). Artificial intelligence-based accurate myocardial infarction mapping using 617 12-lead electrocardiography. European Heart Journal-Digital Health pp. ztaf077.618

work page 2025
[27]

Holmes: health online model ensemble serving for deep learning models in intensive 620 care units

Hong, S., Xu, Y ., Khare, A., Priambada, S., Maher, K., Aljiffry, A., Sun, J., and Tumanov, A.619 (2020). Holmes: health online model ensemble serving for deep learning models in intensive 620 care units. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge 621 Discovery & Data Mining. pp. 1614–1624. 622

work page 2020
[28]

Li, J., Aguirre, A.D., Junior, V.M., Jin, J., Liu, C., Zhong, L., Sun, C., Clifford, G., Bran- 623 don Westover, M., and Hong, S. (2025). An electrocardiogram foundation model built on over 624 10 million recordings. NEJM AI2, AIoa2401033. 625

work page 2025
[29]

Kendall, A., Gal, Y ., and Cipolla, R. (2018). Multi-task learning using uncertainty to weigh 626 losses for scene geometry and semantics. In Proceedings of the IEEE conference on 627 computer vision and pattern recognition. pp. 7482–7491.628

work page 2018
[30]

Yu, T., Kumar, S., Gupta, A., Levine, S., Hausman, K., and Finn, C. (2020). Gradient surgery 629 for multi-task learning. Advances in neural information processing systems33, 5824–5836.630 21

work page 2020
[31]

Bergström, G., Persson, M., Adiels, M., Björnson, E., Bonander, C., Ahlström, H., Alfredsson, 631 J., Angerås, O., Berglund, G., Blomberg, A. et al. (2021). Prevalence of subclinical coronary 632 artery atherosclerosis in the general population. Circulation144, 916–929.633

work page 2021
[32]

Knuuti, J., Wijns, W., Saraste, A., Capodanno, D., Barbato, E., Funck-Brentano, C., Prescott, 634 E., Storey, R.F ., Deaton, C., Cuisset, T. et al. (2020). 2019 esc guidelines for the diagnosis635 and management of chronic coronary syndromes: The task force for the diagnosis and 636 management of chronic coronary syndromes of the european society of cardi...

work page 2020
[33]

Temple, R. (2010). Enrichment of clinical study populations. Clinical Pharmacology & 639 Therapeutics88, 774–778. 640 22

work page 2010

[1] [1]

Martin, S.S., Aday, A.W., Almarzooq, Z.I., Anderson, C.A., Arora, P ., Avery, C.L., Baker-Smith,527 C.M., Barone Gibbs, B., Beaton, A.Z., Boehme, A.K. et al. (2024). 2024 heart disease 528 and stroke statistics: a report of us and global data from the american heart association. 529 Circulation149, e347–e913. 530

work page 2024

[2] [2]

Mensah, G.A., Fuster, V., Murray, C.J., Roth, G.A., of Cardiovascular Diseases, G.B., and 531 Collaborators, R. (2023). Global burden of cardiovascular diseases and risks, 1990-2022. 532 Journal of the American College of Cardiology82, 2350–2473.533

work page 2023

[3] [3]

Luengo-Fernandez, R., Walli-Attaei, M., Gray, A., Torbica, A., Maggioni, A.P ., Huculeci, 534 R., Bairami, F ., Aboyans, V., Timmis, A.D., Vardas, P . et al. (2023). Economic burden of 535 cardiovascular diseases in the european union: a population-based cost study. European 536 heart journal44, 4752–4767. 537

work page 2023

[4] [4]

Vrints, C., Andreotti, F ., Koskinas, K.C., Rossello, X., Adamo, M., Ainslie, J., Banning, A.P .,538 Budaj, A., Buechel, R.R., Chiariello, G.A. et al. (2024). 2024 esc guidelines for the man- 539 agement of chronic coronary syndromes: developed by the task force for the management 540 of chronic coronary syndromes of the european society of cardiology (es...

work page 2024

[5] [5]

Kim, C., Park, C.H., Lee, B.Y ., Park, C.H., Kang, E.J., Koo, H.J., Kitagawa, K., Cha, M.J., 544 Krittayaphong, R., Choi, S.I. et al. (2024). 2024 consensus statement on coronary stenosis545 and plaque evaluation in ct angiography from the asian society of cardiovascular imaging- 546 practical tutorial (asci-pt). Korean journal of radiology25, 331.547 19

work page 2024

[6] [6]

Members, W.C., Gulati, M., Levy, P .D., Mukherjee, D., Amsterdam, E., Bhatt, D.L., Birtcher,548 K.K., Blankstein, R., Boyd, J., Bullock-Palmer, R.P . et al. (2021). 2021 aha/acc/ase/ch- 549 est/saem/scct/scmr guideline for the evaluation and diagnosis of chest pain: a report of 550 the american college of cardiology/american heart association joint commit...

work page 2021

[7] [7]

Narula, J., Chandrashekhar, Y ., Ahmadi, A., Abbara, S., Berman, D.S., Blankstein, R., Leipsic,553 J., Newby, D., Nicol, E.D., Nieman, K. et al. (2020). Scct 2021 expert consensus document 554 on coronary computed tomographic angiography: a report of the society of cardiovascular 555 computed tomography. Journal of cardiovascular computed tomography15, 192.556

work page 2020

[8] [8]

Yu, H., Song, H., Sun, X., Song, T., Xie, A., Xu, J., Qin, R., Jing, L., Zuo, T., Zhao, J. et al. 557 (2024). Reduced radiation dose and volume of contrast medium in heart rate-based, one-stop 558 computed tomography angiography of coronary, carotid and cerebrovascular arteries. Acta559 Radiologica65, 84–90. 560

work page 2024

[9] [9]

Jain, S., and Majidi, F . (2025). Ct coronary angiography vs cardiac catheterization and risk of 561 acute kidney injury–comparative study. Journal of Clinical Cardiology6, 84–90.562

work page 2025

[10] [10]

Byrne, R.A., Rossello, X., Coughlan, J., Barbato, E., Berry, C., Chieffo, A., Claeys, M.J., Dan, 563 G.A., Dweck, M.R., Galbraith, M. et al. (2024). 2023 esc guidelines for the management 564 of acute coronary syndromes: developed by the task force on the management of acute 565 coronary syndromes of the european society of cardiology (esc). European Hear...

work page 2024

[11] [11]

Bhatt, D.L., Lopes, R.D., and Harrington, R.A. (2022). Diagnosis and treatment of acute 568 coronary syndromes: a review. Jama327, 662–675. 569

work page 2022

[12] [12]

Thygesen, K., Alpert, J.S., Jaffe, A.S., Chaitman, B.R., Bax, J.J., Morrow, D.A., White, 570 H.D., and on behalf of the Joint European Society of Cardiology (ESC)/American College of 571 Cardiology (ACC)/American Heart Association (AHA)/World Heart Federation (WHF) Task 572 Force for the Universal Definition of Myocardial Infarction, E.G. (2018). Fourth u...

work page 2018

[13] [13]

Zhao, Q., Geng, S., Wang, B., Sun, Y ., Nie, W., Bai, B., Yu, C., Zhang, F ., Tang, G., Zhang, D.576 et al. (2024). Deep learning in heart sound analysis: from techniques to clinical applications. 577 Health Data Science4, 0182. 578

work page 2024

[14] [14]

Luo, Y ., Liu, X.Y ., Y ang, K., Huang, K., Hong, M., Zhang, J., Wu, Y ., and Nie, Z. (2024). 579 Toward unified ai drug discovery with multimodal knowledge. Health Data Science4, 0113.580

work page 2024

[15] [15]

Zhang, S., Mu, W., Dong, D., Wei, J., Fang, M., Shao, L., Zhou, Y ., He, B., Zhang, S., Liu,581 Z. et al. (2023). The applications of artificial intelligence in digestive system neoplasms: a 582 review. Health Data Science3, 0005. 583

work page 2023

[16] [16]

Tham, Y .C., Goh, J.H.L., Anees, A., Lei, X., Rim, T.H., Chee, M.L., Wang, Y .X., Jonas, 584 J.B., Thakur, S., Teo, Z.L. et al. (2022). Detecting visually significant cataract using retinal 585 photograph-based deep learning. Nature aging2, 264–271.586

work page 2022

[17] [17]

et al.587 (2021)

Liu, X., Gao, K., Liu, B., Pan, C., Liang, K., Y an, L., Ma, J., He, F ., Zhang, S., Pan, S. et al.587 (2021). Advances in deep learning-based medical image analysis. Health Data Science 588 2021, 8786793. 589 20

work page 2021

[18] [18]

590 (2025)

Xiao, Y ., Tang, G., Liu, W., Li, J., Nie, G., Kan, Z., Zhang, D., Zhao, Q., and Hong, S. 590 (2025). Anyecg-lab: An exploration study of fine-tuning an ecg foundation model to estimate 591 laboratory values from single-lead ecg signals. arXiv preprint arXiv:2510.22301.592

work page arXiv 2025

[19] [19]

Croon, P .M., Dhingra, L.S., Biswas, D., Oikonomou, E.K., and Khera, R. (2025). Phenotypic 593 selectivity of artificial intelligence-enhanced electrocardiography in cardiovascular diagnosis 594 and risk prediction. Circulation152. 595

work page 2025

[20] [20]

Poterucha, T.J., Jing, L., Ricart, R.P ., Adjei-Mosi, M., Finer, J., Hartzel, D., Kelsey, C., Long, A.,596 Rocha, D., Ruhl, J.A. et al. (2025). Detecting structural heart disease from electrocardiograms 597 using ai. Nature644, 221–230. 598

work page 2025

[21] [21]

Y ang, K., Hong, M., Zhang, J., Luo, Y ., Zhao, S., Zhang, O., Yu, X., Zhou, J., Y ang, L., Zhang,599 P . et al. (2025). Ecg-lm: Understanding electrocardiogram with a large language model. 600 Health Data Science5, 0221. 601

work page 2025

[22] [22]

Al-Zaiti, S.S., Martin-Gill, C., Zègre-Hemsey, J.K., Bouzid, Z., Faramand, Z., Alrawashdeh, 602 M.O., Gregg, R.E., Helman, S., Riek, N.T., Kraevsky-Phillips, K. et al. (2023). Machine 603 learning for ecg diagnosis and risk stratification of occlusion myocardial infarction. Nature 604 Medicine29, 1804–1813. 605

work page 2023

[23] [23]

Lin, C.S., Liu, W.T., Tsai, D.J., Lou, Y .S., Chang, C.H., Lee, C.C., Fang, W.H., Wang, C.C.,606 Chen, Y .Y ., Lin, W.S. et al. (2024). Ai-enabled electrocardiography alert intervention and 607 all-cause mortality: a pragmatic randomized clinical trial. Nature Medicine30, 1461–1470.608

work page 2024

[24] [24]

609 (2023)

Choi, S.H., Lee, H.G., Park, S.D., Bae, J.W., Lee, W., Kim, M.S., Kim, T.H., and Lee, W.K. 609 (2023). Electrocardiogram-based deep learning algorithm for the screening of obstructive 610 coronary artery disease. BMC cardiovascular disorders23, 287.611

work page 2023

[25] [25]

612 (2024)

Park, J., Kim, J., Kang, S.H., Lee, J., Hong, Y ., Chang, H.J., Cho, Y ., and Y oon, Y .E. 612 (2024). Artificial intelligence–enhanced electrocardiography analysis as a promising tool 613 for predicting obstructive coronary artery disease in patients with stable angina. European 614 Heart Journal-Digital Health5, 444–453. 615

work page 2024

[26] [26]

Wang, H., Gao, Z., Zhang, H., Zhu, Y ., Lian, S., Bo, K., Li, S., Gao, Y ., Zhuang, B., Zhou, 616 Z. et al. (2025). Artificial intelligence-based accurate myocardial infarction mapping using 617 12-lead electrocardiography. European Heart Journal-Digital Health pp. ztaf077.618

work page 2025

[27] [27]

Holmes: health online model ensemble serving for deep learning models in intensive 620 care units

Hong, S., Xu, Y ., Khare, A., Priambada, S., Maher, K., Aljiffry, A., Sun, J., and Tumanov, A.619 (2020). Holmes: health online model ensemble serving for deep learning models in intensive 620 care units. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge 621 Discovery & Data Mining. pp. 1614–1624. 622

work page 2020

[28] [28]

Li, J., Aguirre, A.D., Junior, V.M., Jin, J., Liu, C., Zhong, L., Sun, C., Clifford, G., Bran- 623 don Westover, M., and Hong, S. (2025). An electrocardiogram foundation model built on over 624 10 million recordings. NEJM AI2, AIoa2401033. 625

work page 2025

[29] [29]

Kendall, A., Gal, Y ., and Cipolla, R. (2018). Multi-task learning using uncertainty to weigh 626 losses for scene geometry and semantics. In Proceedings of the IEEE conference on 627 computer vision and pattern recognition. pp. 7482–7491.628

work page 2018

[30] [30]

Yu, T., Kumar, S., Gupta, A., Levine, S., Hausman, K., and Finn, C. (2020). Gradient surgery 629 for multi-task learning. Advances in neural information processing systems33, 5824–5836.630 21

work page 2020

[31] [31]

Bergström, G., Persson, M., Adiels, M., Björnson, E., Bonander, C., Ahlström, H., Alfredsson, 631 J., Angerås, O., Berglund, G., Blomberg, A. et al. (2021). Prevalence of subclinical coronary 632 artery atherosclerosis in the general population. Circulation144, 916–929.633

work page 2021

[32] [32]

Knuuti, J., Wijns, W., Saraste, A., Capodanno, D., Barbato, E., Funck-Brentano, C., Prescott, 634 E., Storey, R.F ., Deaton, C., Cuisset, T. et al. (2020). 2019 esc guidelines for the diagnosis635 and management of chronic coronary syndromes: The task force for the diagnosis and 636 management of chronic coronary syndromes of the european society of cardi...

work page 2020

[33] [33]

Temple, R. (2010). Enrichment of clinical study populations. Clinical Pharmacology & 639 Therapeutics88, 774–778. 640 22

work page 2010