pith. sign in

arxiv: 2605.21745 · v1 · pith:J7WOBYNUnew · submitted 2026-05-20 · 💻 cs.LG

Quantitative coronary calcification analysis for prediction of myocardial ischemia using non-contrast CT calcium scoring

Juhwan Lee , Sadeer Al-Kindi , Ammar Hoori , Tao Hu , Hao Wu , Justin N. Kim , Robert Gilkeson , Sanjay Rajagopalan
show 1 more author
David L. Wilson
This is my paper

Pith reviewed 2026-05-22 09:42 UTC · model grok-4.3

classification 💻 cs.LG
keywords myocardial ischemianon-contrast CTcalcium scoringmachine learningAgatston scorecalcium-omicsprediction modelcardiovascular risk
0
0 comments X

The pith

Machine learning model predicts myocardial ischemia from non-contrast CT calcium scans using eight calcium-omics features.

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

This paper develops a machine learning framework to predict myocardial ischemia from routine non-contrast CT calcium scoring scans in 987 patients. The model combines the Agatston score, age, and eight calcium-omics features to reach 98.9 percent precision, 79.2 percent sensitivity, and 87.7 percent F1 score. These features add measurable value beyond clinical variables or the Agatston score alone. The approach aims to support cardiovascular risk assessment with scans that are already widely performed without contrast or extra tests.

Core claim

The authors claim that a machine learning model using the Agatston score, eight calcium-omics features, and age predicts myocardial ischemia from non-contrast CTCS scans with precision of 98.9 percent, sensitivity of 79.2 percent, and F1 score of 87.7 percent, while showing that calcium-omics features deliver significant incremental predictive value over models limited to clinical variables or the Agatston score.

What carries the argument

XGBoost classifier guided by SHAP values, applied to calcium-omics features extracted from CTCS images along with the Agatston score.

If this is right

  • Calcium-omics features significantly improve prediction compared with clinical variables alone or with the Agatston score.
  • The number of calcified arteries shows a strong link to myocardial ischemia even if ranked lower by SHAP.
  • This framework enables cardiovascular risk assessment from routinely acquired non-contrast scans.
  • Machine learning integration of multiple calcium measures can refine identification of ischemia risk.

Where Pith is reading between the lines

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

  • Widespread adoption could reduce reliance on contrast-enhanced or stress imaging for initial risk screening in some patients.
  • The same features might be tested for predicting other outcomes like future cardiac events in larger registries.
  • Embedding the model in existing CT software could alter how calcium scores are reported in clinical practice.

Load-bearing premise

The 987 patients analyzed represent an unbiased sample of those who receive both CTCS and stress PET, and the calcium-omics features can be extracted consistently from standard non-contrast scans without special tuning.

What would settle it

An independent study on a separate group of patients with both non-contrast CTCS and regadenoson stress PET that finds no statistically significant gain in precision or F1 score when calcium-omics features are added to the Agatston score and age would refute the claim of incremental value.

read the original abstract

Non-contrast computed tomography calcium scoring (CTCS) is widely recognized as an effective tool for cardiovascular risk stratification. This study aimed to develop a novel machine learning framework for predicting myocardial ischemia from routine non-contrast CTCS scans using quantitative coronary calcium assessment. This study analyzed 1,375 patients who underwent both non-contrast CTCS and regadenoson stress cardiac positron emission tomography myocardial perfusion imaging within one year at University Hospitals Cleveland Medical Center. A total of 74 variables, including clinical variables, Agatston score, and calcium-omics features, were evaluated. Relevant features were identified using XGBoost with Shapley Additive exPlanations (SHAP). Predictive models were trained and evaluated using 5-fold cross-validation. Among 987 patients, 89 (9%) were positive for myocardial ischemia. The final model incorporated the Agatston score, eight calcium-omics features, and age. The proposed model achieved a precision of 98.9+/-3.0%, sensitivity of 79.2+/-8.4, and F1 score of 87.7+/-5.3%. The addition of calcium-omics features significantly improved predictive performance compared with models using clinical variables alone or clinical variables with the Agatston score (p<0.05). Interestingly, the number of calcified arteries, despite being the lowest-ranked feature based on SHAP analysis, showed the strongest association with myocardial ischemia in logistic regression analysis (odds ratio: 3.63, 95% confidence interval: 2.80-4.77, p<0.00001). We developed a machine learning approach for predicting myocardial ischemia using routinely acquired non-contrast CTCS scans. Calcium-omics features provided incremental predictive value beyond conventional risk factors and Agatston scoring and may support more accessible cardiovascular risk stratification.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a machine learning framework to predict myocardial ischemia from routine non-contrast CT calcium scoring (CTCS) scans by combining clinical variables, the Agatston score, and novel 'calcium-omics' features. After analyzing 987 patients (89 with ischemia) from a single center who underwent both CTCS and stress PET within one year, the authors use XGBoost with SHAP to select eight calcium-omics features plus Agatston score and age, then train and evaluate models via 5-fold cross-validation. They report precision of 98.9±3.0%, sensitivity of 79.2±8.4, F1 of 87.7±5.3, and statistically significant improvement (p<0.05) over clinical-only and clinical+Agatston baselines. An additional logistic regression finding notes that the number of calcified arteries has a strong association (OR 3.63) despite low SHAP rank.

Significance. If the performance gains hold under proper validation, the work could support more accessible ischemia risk stratification using widely available non-contrast CTCS without requiring contrast or stress testing. The explicit comparison to Agatston and clinical baselines plus SHAP-based interpretability are strengths. The small positive class (9%) and single-center retrospective design, however, limit immediate generalizability even if methodological concerns are resolved.

major comments (2)
  1. [Methods] Methods (feature selection and model evaluation): The abstract states that relevant features were identified using XGBoost with SHAP and that predictive models were then trained and evaluated using 5-fold cross-validation. If SHAP-based selection of the eight calcium-omics features occurred on the full 987-patient cohort rather than being nested inside each training fold, the reported precision, F1 score, and p<0.05 improvement over baselines are optimistically biased due to leakage. With only 89 positive cases this is especially consequential; the manuscript must clarify the exact procedure and, if needed, repeat the analysis with nested selection to substantiate the central claim of incremental value from calcium-omics features.
  2. [Methods] Methods (cohort construction): The reduction from the initial 1,375 patients to the final 987 analyzed is described without explicit exclusion criteria or a flow diagram. This information is load-bearing for assessing selection bias and whether the analyzed sample supports the generalizable prediction claims made in the abstract and discussion.
minor comments (2)
  1. [Introduction/Methods] The term 'calcium-omics features' is introduced without a concise definition or reference to their exact computation from CTCS images; a short methods paragraph or supplementary table listing the 74 initial variables and how the eight retained features are derived would improve reproducibility.
  2. [Methods] The abstract and results do not state how class imbalance (9% positive) was handled during model training or whether stratified folds were used; this detail should be added for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive review of our manuscript. The comments on feature selection procedures and cohort reporting are well-taken and have prompted revisions that improve methodological transparency. We address each major comment below.

read point-by-point responses

  1. Referee: [Methods] Methods (feature selection and model evaluation): The abstract states that relevant features were identified using XGBoost with SHAP and that predictive models were then trained and evaluated using 5-fold cross-validation. If SHAP-based selection of the eight calcium-omics features occurred on the full 987-patient cohort rather than being nested inside each training fold, the reported precision, F1 score, and p<0.05 improvement over baselines are optimistically biased due to leakage. With only 89 positive cases this is especially consequential; the manuscript must clarify the exact procedure and, if needed, repeat the analysis with nested selection to substantiate the central claim of incremental value from calcium-omics features.

    Authors: We acknowledge that the original manuscript did not explicitly state whether SHAP-based feature selection was performed inside or outside the cross-validation loop. The selection was conducted on the full cohort prior to partitioning, which raises the valid concern of optimistic bias noted by the referee. To correct this, we have revised the Methods to implement fully nested feature selection: within each training fold, XGBoost+SHAP is used to identify the eight calcium-omics features plus Agatston score and age; the model is then trained only on those features within the same fold and evaluated on the held-out fold. The revised analysis preserves the central finding that calcium-omics features yield statistically significant improvement over clinical-only and clinical+Agatston baselines (p<0.05). Updated performance metrics and a clear description of the nested procedure appear in the revised manuscript. revision: yes

  2. Referee: [Methods] Methods (cohort construction): The reduction from the initial 1,375 patients to the final 987 analyzed is described without explicit exclusion criteria or a flow diagram. This information is load-bearing for assessing selection bias and whether the analyzed sample supports the generalizable prediction claims made in the abstract and discussion.

    Authors: We agree that explicit documentation of cohort construction is required for evaluating selection bias. The revised Methods section now provides the complete exclusion criteria applied to the initial 1,375 patients who underwent both CTCS and stress PET within one year. A CONSORT-style flow diagram has also been added to illustrate the step-wise reduction to the final analytic cohort of 987 patients (89 with ischemia). These additions allow readers to assess the generalizability of the reported results. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation or claims

full rationale

The paper describes an empirical ML pipeline: feature identification via XGBoost+SHAP followed by 5-fold cross-validation for training and evaluation on the 987-patient cohort, with reported metrics (precision 98.9±3.0%, F1 87.7±5.3%) and incremental value claims (p<0.05) over baselines. No mathematical derivation chain, equations, or first-principles results exist that reduce by construction to inputs; the central performance claims rest on standard cross-validation rather than self-definitional fits, fitted inputs renamed as predictions, or load-bearing self-citations. While the abstract leaves the exact nesting of SHAP selection unspecified (raising potential leakage risk if performed outside folds), this is a methodological detail, not a reduction of the reported result to its own inputs by definition. The study is self-contained against external benchmarks with no circular steps matching the enumerated patterns.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on data-driven selection of eight calcium-omics features via SHAP and training on a single-center cohort with paired CTCS and stress PET data; no machine-checked proofs or external benchmarks are referenced.

free parameters (1)
  • Number of calcium-omics features retained
    Selected via SHAP from 74 total variables to form the final model
axioms (1)
  • domain assumption Patients undergoing both non-contrast CTCS and regadenoson stress PET within one year form a suitable training population for ischemia prediction
    This defines the analyzed cohort of 987 patients
invented entities (1)
  • calcium-omics features no independent evidence
    purpose: Provide quantitative coronary calcium measures beyond the Agatston score for improved ischemia prediction
    Newly introduced term and set of features in this work

pith-pipeline@v0.9.0 · 5898 in / 1522 out tokens · 77577 ms · 2026-05-22T09:42:59.245486+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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

27 extracted references · 27 canonical work pages

  1. [1]

    The Lancet 392:1736–1788

    Roth GA, Abate D, Abate KH, et al (2018) Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017. The Lancet 392:1736–1788. https://doi.org/10.1016/S0140-6736(18)32203-7

    work page doi:10.1016/s0140-6736(18)32203-7 2018

  2. [2]

    European Heart Journal 35:2950–2959

    Nichols M, Townsend N, Scarborough P, Rayner M (2014) Cardiovascular disease in Europe 2014: epidemiological update. European Heart Journal 35:2950–2959. https://doi.org/10.1093/eurheartj/ehu299

    work page doi:10.1093/eurheartj/ehu299 2014

  3. [3]

    https://doi.org/10.1152/ajpheart.00139.2019

    Heusch G (2019) Myocardial ischemia: lack of coronary blood flow, myocardial oxygen supply-demand imbalance, or what? American Journal of Physiology-Heart and Circulatory Physiology 316:H1439–H1446. https://doi.org/10.1152/ajpheart.00139.2019

    work page doi:10.1152/ajpheart.00139.2019 2019

  4. [4]

    Journal of Clinical Lipidology 15:33–60

    Orringer CE, Blaha MJ, Blankstein R, et al (2021) The National Lipi d Association scientific statement on coronary artery calcium scoring to guide preventive strategies for ASCVD risk reduction. Journal of Clinical Lipidology 15:33–60. https://doi.org/10.1016/j.jacl.2020.12.005

    work page doi:10.1016/j.jacl.2020.12.005 2021

  5. [5]

    JACC: Cardiovascular Imaging 16:98–117

    Golub IS, Termeie OG, Kristo S, et al (2023) Major Global Coronary Artery Calcium Guidelines. JACC: Cardiovascular Imaging 16:98–117. https://doi.org/10.1016/j.jcmg.2022.06.018

    work page doi:10.1016/j.jcmg.2022.06.018 2023

  6. [6]

    Journal of the Amer ican College of Cardiology 15:827–832

    Agatston AS, Janowitz WR, Hildner FJ, et al (1990) Quantification of coronary artery calcium using ultrafast computed tomography. Journal of the Amer ican College of Cardiology 15:827–832. https://doi.org/10.1016/0735-1097(90)90282-T

    work page doi:10.1016/0735-1097(90)90282-t 1990

  7. [7]

    Sci Rep 14:11134

    Hoori A, Al-Kindi S, Hu T, et al (2024) Enhancing cardiovascular risk prediction through AI-enabled calcium-omics. Sci Rep 14:11134. https://doi.org/10.1038/s41598-024-60584-8

    work page doi:10.1038/s41598-024-60584-8 2024

  8. [8]

    ArXiv arXiv:2401.16190v1

    Hu T, Freeze J, Singh P, et al (2 024) AI prediction of cardiovascular events using opportunistic epicardial adipose tissue assessments from CT calcium score. ArXiv arXiv:2401.16190v1

    work page arXiv

  9. [9]

    Journal of Cardiovascular Computed Tomography 19:224–231

    Lee J, Hu T, Williams MC, et al (2025) Prediction of obstructive cor onary artery disease using coronary calcification and epicardial adipose tissue assessments from CT calcium scoring scans. Journal of Cardiovascular Computed Tomography 19:224–231. https://doi.org/10.1016/j.jcct.2025.01.007

    work page doi:10.1016/j.jcct.2025.01.007 2025

  10. [10]

    In: Medical Imaging 2025: Computer-Aided Diagnosis

    Al-Rawi A, Kalra D, Bricker N, et al (2025) AI prediction of obstructive coronary artery disease using calcium-omics from non-contrast CT calcium scorin g scans. In: Medical Imaging 2025: Computer-Aided Diagnosis. SPIE, pp 595–602

    work page 2025

  11. [11]

    Front Cardiovasc Med 12:1543816

    Lee J, Hu T, Williams MC, et al (2025) Detection of arterial remodeling using epicardial adipose tissue assessment from CT calcium scoring scan. Front Cardiovasc Med 12:1543816. https://doi.org/10.3389/fcvm.2025.1543816

    work page doi:10.3389/fcvm.2025.1543816 2025

  12. [12]

    JACC: Advances 3:101188

    Hu T, Freeze J, Singh P, et al (2024) Artificial In telligence Prediction of Ca rdiovascular Events Using Opportunistic Epicardial Adipose Tissue Assessments From Computed Tomography Calcium Score. JACC: Advances 3:101188. https://doi.org/10.1016/j.jacadv.2024.101188

    work page doi:10.1016/j.jacadv.2024.101188 2024

  13. [13]

    Sci Rep 14:26898

    Singh P, Hoori A, Freeze J, et al (2024) Leveraging calcium score CT radiomics for heart failure risk prediction. Sci Rep 14:26898. https://doi.org/10.1038/s41598-024-77269-x 8

    work page doi:10.1038/s41598-024-77269-x 2024

  14. [14]

    In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

    Chen T, Guestrin C (2016) XGB oost: A Scalable Tree Boosting Syst em. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. pp 785–794

    work page 2016

  15. [15]

    In: Proceedings of the 31st International Conference on Neural Information Pr ocessing Systems

    Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Pr ocessing Systems. Curran Associates Inc., Red Hook, NY, USA, pp 4768–4777

    work page 2017

  16. [16]

    Sci Rep 12:2276

    Hoori A, Hu T, Lee J, et al (2022) Deep learni ng segmentation and quantification method for assessing epicardial adipose tissue in CT calcium score scans. Sci Rep 12:2276. https://doi.org/10.1038/s41598-022- 06351-z

    work page doi:10.1038/s41598-022- 2022

  17. [17]

    JMI 10:014002

    Song Y, Hoori A, Wu H, et al (2023) Improved bi as and reproducibility of coronary artery calcification features using deconvolution. JMI 10:014002. https://doi.org/10.1117/1.JMI.10.1.014002

    work page doi:10.1117/1.jmi.10.1.014002 2023

  18. [18]

    In: Gimi BS, Krol A (eds) Medical Imaging 2024: Clinical and Biomedical Imaging

    Hoori A, Freeze J, Singh P, et al (2024) Prediction of major adverse cardi ovascular events using comprehensive AI analysis of calcifications and fat de pots in CT calcium score images. In: Gimi BS, Krol A (eds) Medical Imaging 2024: Clinical and Biomedical Imaging. SPIE, San Diego, United States, p 10

    work page 2024

  19. [19]

    In: Gimi BS, Krol A (eds) Medical Imaging 2023: Biomedical Applications in Molecular, Structural, and Functional Imaging

    Hoori A, Hu T, Lee J, et al ( 2023) An enriched survival study of ep icardial adipose tissues risk on major adverse cardiovascular event in CT calcium score imag es. In: Gimi BS, Krol A (eds) Medical Imaging 2023: Biomedical Applications in Molecular, Structural, and Functional Imaging. SPIE, San Diego, United States, p 27

    work page 2023

  20. [20]

    In: Medical Imaging 2024: Clinical and Biomedical Imaging

    Wu H, Song Y, Hoori A, et al (2024) Translating no n-contrast CT calcium score images to virtual CCTA to aid segmentation of coronary arteries and myocardium . In: Medical Imaging 2024: Clinical and Biomedical Imaging. SPIE, pp 26–31

    work page 2024

  21. [21]

    In: Gimi BS, Krol A (eds) Medical Imaging 2024: Clinical and Biomedical Imaging

    Hu T, Hoori A, Lee J, et al (2 024) AI predictions of major adverse card iovascular event using epicardial and paracardial adipose tissue assessments in CT calcium score images. In: Gimi BS, Krol A (eds) Medical Imaging 2024: Clinical and Biomedical Imaging. SPIE, San Diego, United States, p 11

    work page 2024

  22. [22]

    Song Y, Wu H, Lee J, et al (202 4) Pericoronary adipose tissue feature analysis in CT calcium score images with comparison to coronary CTA

    work page

  23. [23]

    Front Cardiovasc Med 12:

    Wu H, Song Y, Hoori A, et al (2025) Quantitative cardiac CT perfusion: physiologically-inspired model and identifying microvascular disease from discordant CTA CAD-RADS. Front Cardiovasc Med 12:. https://doi.org/10.3389/fcvm.2025.1621443

    work page doi:10.3389/fcvm.2025.1621443 2025

  24. [24]

    Journal of Clinical Medicine 14:769

    Wu H, Song Y, Hoori A, et al (2025) Cardiac CT Perfusion Imaging of Pericoronary Adipose Tissue (PCAT) Highlighting Potential Confounds in CTA Analysis. Journal of Clinical Medicine 14:769. https://doi.org/10.3390/jcm14030769

    work page doi:10.3390/jcm14030769 2025

  25. [25]

    JMI 12:014503

    Song Y, Wu H, Lee J, et al (2025) Pericoronary adipose tissue feature analysis in computed tomography calcium score images in comparison to coronary computed tomography angiography. JMI 12:014503. https://doi.org/10.1117/1.JMI.12.1.014503

    work page doi:10.1117/1.jmi.12.1.014503 2025

  26. [26]

    JMI 12:016002

    Kim JN, Song Y, Wu H, et al (2025) Improving coronary artery segmentation with self-supervised learning and automated pericoronary adipose tissue segmentation: a multi-institutional study on coronary computed tomography angiography images. JMI 12:016002. https://doi.org/10.1117/1.JMI.12.1.016002

    work page doi:10.1117/1.jmi.12.1.016002 2025

  27. [27]

    In: 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI)

    Kim JN, Hu T, Wu H, et al (2025) Improved C-In dex and Interpretability with KAN as Compared to COX: Application to Risk Prediction of Major Adverse Cardiovascular Events from CT Calcium Score. In: 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI). pp 1–5 9 Tables Table 1 Baseline clinical characteristics of patients with myocardial isch...

    work page 2025