Unlocking the hidden potential of coronary artery calcium scans: AI-derived chamber volume ratios as novel predictors of heart failure

Twenty million noncardiac chest computed tomography (CT) scan are performed annually in the United States including 1 million dedicated coronary artery calcium (CAC) scans (1). CAC scan requires low radiation (1 mS) and is low cost, however incidental findings and the cascade they trigger can cause 5–10 times more radiation and cost (2). Moreover, while a normal CAC score of 0 is useful, the overall effects of risk reclassification by CAC is modest among low-risk individuals who are reclassified upward by CACS or subjects at high risk who are down classified by CACS during a 5–10-year follow-up (3).

In this issue, Naghavi et al. (4) used cardiac segmentation module in their proprietary artificial intelligence (AI) tool to obtain additional information of cardiac four-chamber volumes and their ratios from CT CAC scans performed in two large population-based studies. This AI-derived chamber volumetry represents what might be termed the ideal opportunistic finding. Importantly, this information is generated without the downstream imaging cascades, and cumulative radiation exposure that characterize traditional incidental findings. What distinguishes the Naghavi approach is that the hazard ratios (HRs) for left ventricular/right ventricular (LV/RV) volumes ≥95th percentile substantially exceed those typically seen with CAC scoring alone. If validated prospectively, this approach could fundamentally redefine the value of the so-called “CAC” scan—to a “comprehensive cardiovascular (CV) scan” that provides a comprehensive cardiac risk profile derived from a cost-effective test with minimal radiation, and which is already being performed in millions of individuals annually (Figure 1).

Figure 1 Beyond the Agatston score: AI-enabled cardiac phenotyping from routine coronary artery calcium scans. Routine non-contrast electrocardiographically gated CAC scans contain substantially more anatomical information than the Agatston score alone. (A) Conventional CAC assessment generates a single calcium score that has been the primary clinical output for cardiovascular risk stratification. (B) AI-based deep learning algorithms can automatically segment the four cardiac chambers from the same non-contrast CAC scan, enabling reproducible quantification of cardiac chamber morphology without additional imaging, radiation exposure, contrast administration, or patient burden. (C) AI-derived chamber volume measurements permit calculation of physiologically relevant chamber volume ratios, such as the LV/RV volume ratio, which may detect ventricular interdependence and subtle chamber remodeling that are not apparent from absolute chamber volumes alone. (D) Elevated AI-derived chamber volume ratios provide incremental prognostic information for predicting incident heart failure, independent of traditional cardiovascular risk factors, illustrating how routine CAC scans may evolve from a single-purpose calcium scoring examination into a comprehensive cardiovascular phenotyping tool. AI, artificial intelligence; CAC, coronary artery calcium; CI, confidence interval; CT, computed tomography; HF, heart failure; HR, hazard ratio; LA, left atrium; LV, left ventricle; RA, right atrium; RV, right ventricle.

The study findings

Using AI-based automated segmentation, Naghavi and colleagues analyzed CAC scans from 5,732 participants in the Multi-Ethnic Study of Atherosclerosis (MESA) over a median follow-up of 17.7 years and validated their findings in the Framingham Heart Study Offspring cohort (FOS) of 1,052 subjects followed for a median of 14.4 years. Heart failure incidence was 6.3% in MESA and 5.3% in FOS. They have previously validated this AI segmentation approach against contrast-enhanced coronary CT angiography (5) and have also demonstrated that the prognostic value of AI-derived LV and RV) chamber volumes is comparable to cardiac magnetic resonance imaging (MRI) in an earlier study (6).

The central finding is that LV/RV volume ratios above the 75th percentile predicted heart failure, with the strongest signal at the highest extreme: participants ≥95th percentile had a four-fold increased risk of heart failure in MESA (Figure 1), while ratios above the 75th percentile were significantly predictive in FOS. All results were adjusted for age, sex, race, and Predicting Risk of CVD EVENTs (PREVENT) risk score components including smoking status, systolic blood pressure, total and high-density lipoprotein (HDL) cholesterol, diabetes status, estimated glomerular filtration rate, body mass index, and use of antihypertensive or lipid-lowering medications.

Beyond the LV/RV ratio, atrial ratios also carried prognostic significance. Left atrial/right atrial (LA/RA) and LA/RV ratios ≥95th percentile were associated with approximately three-fold and two-and-a-half-fold increased heart failure risk, respectively (Figure 1). Notably, the predictive value was substantially stronger in younger participants: among those under 65 years, the HR for LV/RV was 5.60 compared with 3.62 in those over 65 years, and LA/RA ratios above the 95th percentile conferred approximately twice the risk in younger versus older participants in MESA.

Individual absolute chamber volume indices ≥95th percentile also predicted outcomes, though with important differences between cohorts. In MESA, elevated LV and RV volume indices carried four-fold and two-fold increased risk, respectively, and LA volume index above the 95th percentile was more predictive than the LA/RA ratio. In FOS, LA and RA volume indices were more strongly predictive than LV or RV volume indices—highlighting the need for validation across diverse populations.

Perhaps the most clinically compelling finding was that chamber volume ratios captured risk that absolute volumes missed. Beyond the increased heart failure risk conferred by LV volume index above the 95th percentile, participants with normal LV volume index (25th–75th percentile) but an elevated LV/RV ratio ≥95th percentile still had more than a two-fold increased risk (Figure 1). This observation suggests that relative chamber imbalances—reflecting ventricular interdependence and early remodeling—identify a distinct at-risk population invisible to conventional volumetric thresholds.

The physiological foundation: ventricular interdependence

The predictive power of chamber volume ratios is rooted in ventricular interdependence—the mechanical and hemodynamic coupling between ventricles that share the interventricular septum and are constrained by the pericardium (7). Ventricular interdependence manifests through parallel interactions (rightward septal bowing impairing RV filling and vice versa) and series interactions (reduced RV stroke volume decreasing LV preload) (8-10). Recent computational modeling suggests that some patients diagnosed with heart failure with preserved ejection fraction (HFpEF) based on impaired LV diastolic function may have primary RV failure, with apparent LV stiffening representing ventricular interdependence rather than intrinsic LV pathology (11). This biological plausibility provides a framework for understanding why ratios outperform absolute volumes.

Why ratios outperform absolute volumes

Traditional indexed volumes normalize for body surface area but fail to account for individual variations in heart size and the dynamic relationship between chambers (12). The MESA cardiac MRI study by Hoballah et al. demonstrated that high LV/RV ratio (>1.3) predicted heart failure (HR: 2.54), atrial fibrillation (HR: 1.57), and death (HR: 1.62)—and these associations persisted even when both RV and LV volumes were within normal ranges (12). Similarly, in tricuspid regurgitation, over one-third of patients with normal RV end-diastolic volume index had abnormal RV/LV ratios (13).

AI as the critical enabler

The value of cardiac chamber volume ratios is dependent on accurate measurement of cardiac chamber volumes in non-contrast chest CT images and the clinical feasibility of chamber volumetry from CAC scans depends entirely on automated AI-based segmentation. Validation studies confirm that automated chamber volumes from non-contrast CT correlate well with contrast-enhanced studies (Pearson r=0.95; mean relative volume errors <7%) (14,15).

CT-derived versus echocardiographic volumetry

Echocardiography remains the first-line modality for assessing cardiac chamber volumes due to portability, lower cost, and lack of radiation (16). However, LV foreshortening in apical views on two-dimensional (2D) imaging remains a limitation. In addition, the right ventricle presents unique challenges for 2D echocardiographic assessment due to its complex crescent-shaped geometry—with the RV free wall wrapping around the interventricular septum—cannot be captured in a single tomographic plane (16,17). The RV consists of three anatomically distinct components (inlet, apical trabecular, and outlet) that are difficult to image simultaneously. Consequently, 2D echocardiographic methods which rely on geometric assumptions systematically underestimate MRI-derived volumes (18).

Three-dimensional (3D) echocardiography substantially overcomes these limitations. By using volumetric analysis based on pixel counts within the 3D endocardial surface, 3D echocardiography eliminates foreshortening and geometric assumptions (19). 3D echocardiographic RV volumes correlate well with cardiac MRI (r>0.90), with only small negative bias (20). However, 3D echocardiographic volumes, particularly RV 3D volumes remain underutilized.

CT offers distinct advantages compared to echocardiography. Head-to-head comparisons using MRI as the reference standard demonstrate that 64-row CT shows no significant bias in LV volumes, whereas both 2D and 3D echocardiography significantly underestimate these parameters (21). CT demonstrates smaller limits of agreement with MRI and superior interobserver reliability (22). The critical advantage of CT-derived volumetry is simultaneous, standardized assessment of all four chambers from a single acquisition without operator-dependent positioning or acoustic window limitations—particularly important for calculating ratios where systematic biases could distort prognostic value. Given the volume of noncardiac chest CT scans performed CT remains an opportunistic risk-stratification tool without additional imaging burden

The missing distinction: HFpEF versus heart failure with reduced ejection fraction (HFrEF) and the implications for chamber volume ratios

A notable limitation of the Naghavi study is the treatment of heart failure as a single endpoint without distinguishing between HFpEF and HFrEF—two conditions with fundamentally different remodeling signatures that would be expected to produce divergent chamber volume ratio profiles. In HFrEF, the hallmark is eccentric LV remodeling with progressive chamber dilation, increased LV end-diastolic volumes, and a reduced mass-to-volume ratio (23). This LV dilation, often accompanied by secondary RV enlargement due to pulmonary hypertension and ventricular interdependence, would be expected to increase the LV/RV ratio early in disease but potentially normalize it as biventricular failure develops—a pattern seen in approximately two-thirds of patients with severe LV dysfunction who develop concomitant right heart failure (24). In contrast, HFpEF is characterized by concentric LV remodeling with normal or reduced LV cavity volumes, increased mass-to-volume ratio, and preserved or mildly reduced left ventricular ejection fraction (LVEF) (25). LV volumes in HFpEF are typically smaller than in HFrEF, which would be expected to produce a lower absolute LV/RV ratio. However, the pathophysiology of HFpEF is increasingly recognized as a four-chamber disease in which LA remodeling and dysfunction are often more prominent than LV changes, and RV dysfunction—present in one-third to one-half of HFpEF patients—develops from both afterload mismatch and intrinsic myocardial disease (25).

The atrial ratios reported by Naghavi and colleagues may be particularly relevant to HFpEF. LA enlargement is a hallmark of HFpEF, driven by chronically elevated LV filling pressures, and LA dysfunction is more predictive of adverse outcomes than LV abnormalities in this population (26). In HFpEF patients with permanent atrial fibrillation, cardiomegaly is driven exclusively by atrial dilation with biventricular volumes remaining equivalent, producing four-fold greater atrial volumes than controls (27). This suggests that elevated LA and RA ratios in the Naghavi study may preferentially identify patients on the trajectory toward HFpEF with LA myopathy, whereas elevated LV/RV ratios driven by LV dilation may capture those progressing toward HFrEF. In their current study, LA volume index remained as important as LV/RV ratio even after adjustment. The distinction matters because ventricular interdependence—the mechanical coupling of the ventricles through the shared septum, myocardial fibers, and pericardium—operates differently in each phenotype. In HFpEF, particularly in obese patients, enhanced pericardial restraint causes the ventricles to compete for space within the pericardial sac, producing a D-shaped LV cavity and reducing effective LV filling even as measured filling pressures rise (27).

The authors do have the follow up echocardiographic data which was included in the adjudication of HF. Echocardiographic chamber volumes would have been useful to distinguish the two phenotypes of heart failure. The LV/RV ratio might have shown opposite directional associations: elevated ratios predicting HFrEF (reflecting preferential LV dilation) and reduced or normal ratios and elevated LA/RA ratios predicting HFpEF (reflecting concentric LV remodeling with predominant atrial disease). This phenotypic granularity would substantially enhance the clinical utility of AI-derived chamber ratios, potentially enabling not only heart failure risk prediction but also early identification of the likely heart failure subtype—information that carries direct therapeutic implications given the divergent treatment paradigms for HFpEF and HFrEF. This category of LV volumes may fall well within the 25th to 75th centile volume ranges.

While MESA and FOS excluded patients with CV disease and valve surgery, development of valvular heart disease as well as isolated RV dysfunction from pulmonary hypertension, pulmonary embolism as the etiology of heart failure are difficult to exclude and may confound the results.

Study limitations

Several limitations warrant consideration. First, the study relies on a single time-point assessment of cardiac chamber volumes, which captures a snapshot of cardiac geometry but cannot distinguish whether observed chamber ratios reflect disease progression, therapeutic remodeling, or both. While Antihypertensive and lipid-lowering medication use was adjusted as a covariate in the PREVENT risk score adjustment, these medications directly influence cardiac remodeling and these medications might confound or interact with the study’s findings on cardiac chamber volumes—the very substrate the Naghavi study measures—creating a potential confounding pathway not fully addressed by statistical adjustment alone. Binary adjustment for medication use does not capture drug class, dose, duration or degree of remodeling reversal.

Second, both MESA and Framingham Heart Study (FHS) have demographic constraints—the Framingham cohort is approximately 99% White with European ancestry (28). While MESA’s multi-ethnic design (38% White, 28% African American, 22% Hispanic, 12% Chinese) addresses this limitation, both cohorts represent volunteer participants who may differ from the general population (29). The inconsistent predictive performance of individual chamber volume indices between cohorts (LV and RV indices predictive in MESA but not FOS) raises questions about generalizability and suggests that population-specific thresholds may be needed.

Third, AI-derived chamber volumes from non-contrast CT tend to be slightly lower than those from contrast-enhanced studies, with discrepancies increasing at higher volumes (30). The lack of contrast makes precise endocardial border delineation challenging, particularly for the right ventricle.

Fourth, the phenotypic heterogeneity of heart failure—particularly the distinction between HFrEF and HFpEF—was not fully characterized.

Fifth, potential confounding by non-coronary causes of heart failure such as valvular heart disease, pulmonary hypertension, pulmonary embolism, and isolated RV dysfunction may confound interpretation of chamber geometry and ratios and require exclusion of such pathologies by other imaging modalities, most commonly echocardiography as was done in this study.

Finally, AI model performance can degrade substantially causing “domain or data set shift” when applied to populations or imaging protocols different from those used in training (32,33). (I) Imaging equipment, acquisition and reconstruction technique, slice thickness, and contrast attenuation; (II) patients’ age, sex, ethnicity, body habitus, disease prevalence, and comorbidity; and (III) institutional clinical practice and referral patterns, may cause this mismatch (34).

Future research directions

The American Heart Association (AHA) 2024 scientific statement on AI in CV imaging outlines applications across the imaging workflow (33). These volumetric ratios derived from echocardiography, MRI or CT may allow similar if not higher risk prediction than non-contrast CT derived volumes (35) suggesting that opportunistic cardiac assessment may be feasible across multiple imaging platforms. However low radiation dose with non-contrast chest CT, lower cost and its ready availability in resource poor countries make it a more useful tool to assess cardiac volumes.

Integration of chamber ratios with other imaging biomarkers may enhance risk stratification. AI tools may assist in development of standardized reporting frameworks and clinical decision pathways by drawing information from electronic medical records, converting it to International Classification of Diseases codes and combine these with laboratory data and clinical notes, to assist physicians in ordering the most appropriate, cost-effective cardiac imaging test for a specific patient (36).

Prospective clinical trials are needed to determine whether acting on AI-derived chamber ratios improves patient outcomes (37). However, implementation faces challenges including model degradation with out-of-distribution data, automation bias, and need for continuous performance monitoring (38). A joint scientific statement from multiple imaging societies emphasizes that successful deployment requires regulatory compliance, workflow integration, provider trust, and reimbursement mechanisms (38).

Conclusions

The AHA 2025 scientific statement on opportunistic CAC detection emphasizes that approximately 19 million non-electrocardiogram (ECG)-gated chest CT scans are performed annually in the United States, representing a massive, untapped opportunity for CV risk stratification without additional radiation, cost, or patient burden (1). The study by Naghavi et al. represents an important step toward extracting greater clinical value from the millions of CAC scans performed annually. By demonstrating that AI-derived chamber volume ratios strongly predict heart failure—even in patients with normal absolute chamber volumes—the authors have identified a potentially powerful imaging biomarker rooted in the fundamental physiology of ventricular interdependence. The challenge will be translating this prognostic insight into improved patient outcomes through thoughtful clinical implementation, prospective validation, and integration into evidence-based care pathways.

Used Open Evidence with prompts to do literature search on non-contrast CT and AI and for improved readability of parts of the editorial. Used ChatGPT for ideas on Figure 1, which was subsequently hand-drawn by the author. Used ChatGPT to help with figure legend.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Journal of Medical Artificial Intelligence. The article has undergone external peer review.

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-0088/prf

Funding: This study was partly supported by the National Institutes of Health Grant (No. HL 156855).

Conflicts of Interest: The author has completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-0088/coif). The author has no conflicts of interest to declare.

Ethical Statement: The author is accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Foraker R, Sperling L, Bratzke L, et al. Opportunistic Detection of Coronary Artery Calcium on Noncardiac Chest Computed Tomography: An Emerging Tool for Cardiovascular Disease Prevention: A Scientific Statement From the American Heart Association. Circulation 2025;152:e391-e401. [Crossref] [PubMed]
2. Dabrowska M, Przybylo Z, Zukowska M, et al. SHOULD WE BE CONCERNED ABOUT THE DOSES OF IONIZING RADIATION RELATED TO DIAGNOSTIC AND FOLLOW-UP IMAGING IN PATIENTS WITH SOLITARY PULMONARY NODULES? Radiat Prot Dosimetry 2018;178:201-7. [Crossref] [PubMed]
3. Bell KJL, White S, Hassan O, et al. Evaluation of the Incremental Value of a Coronary Artery Calcium Score Beyond Traditional Cardiovascular Risk Assessment: A Systematic Review and Meta-analysis. JAMA Intern Med 2022;182:634-42. [Crossref] [PubMed]
4. Naghavi M, Mirjalili SR, Atlas K, et al. AI-derived left-to-right cardiac chamber volume ratios in coronary artery calcium scans strongly predict heart failure. Eur Heart J Cardiovasc Imaging 2026;27:791-802. [Crossref] [PubMed]
5. Naghavi M, Atlas K, Zhang C, et al. AI-enabled cardiac chambers volumetry in non-contrast coronary artery calcium CT scans vs. contrast-enhanced coronary CT angiography scans in the same patients. Eur Heart J 2024;45:ehae666.
6. Naghavi M, Reeves A, Budoff M, et al. AI-enabled cardiac chambers volumetry in coronary artery calcium scans (AI-CAC^TM) predicts heart failure and outperforms NT-proBNP: The multi-ethnic study of Atherosclerosis. J Cardiovasc Comput Tomogr 2024;18:S16-S17. [Crossref] [PubMed]
7. Konstam MA, Kiernan MS, Bernstein D, et al. Evaluation and Management of Right-Sided Heart Failure: A Scientific Statement From the American Heart Association. Circulation 2018;137:e578-622. [Crossref] [PubMed]
8. Santamore WP, Dell'Italia LJ. Ventricular interdependence: significant left ventricular contributions to right ventricular systolic function. Prog Cardiovasc Dis 1998;40:289-308. [Crossref] [PubMed]
9. Vonk Noordegraaf A, Westerhof BE, Westerhof N. The Relationship Between the Right Ventricle and its Load in Pulmonary Hypertension. J Am Coll Cardiol 2017;69:236-43. [Crossref] [PubMed]
10. Friedberg MK. Imaging Right-Left Ventricular Interactions. JACC Cardiovasc Imaging 2018;11:755-71. [Crossref] [PubMed]
11. Kim SM, Randall EB, Jezek F, et al. Computational modeling of ventricular-ventricular interactions suggest a role in clinical conditions involving heart failure. Front Physiol 2023;14:1231688. [Crossref] [PubMed]
12. Hoballah M, Chehab O, Abdollahi A, et al. Left-to-Right Chamber Volume Ratios at Cardiac MRI Predict Cardiac Events: A Multi-Ethnic Study of Atherosclerosis Secondary Analysis. Radiology 2025;316:e243462. [Crossref] [PubMed]
13. Zhang RS, Villar-Calle P, Jin L, et al. Prognostic Value of the Right Ventricular-to-Left Ventricular Volume Ratio in Tricuspid Regurgitation. JACC Adv 2025;4:101922. [Crossref] [PubMed]
14. Shahzad R, Bos D, Budde RPJ, et al. Automatic segmentation and quantification of the cardiac structures from non-contrast-enhanced cardiac CT scans. Phys Med Biol 2017;62:3798-813. [Crossref] [PubMed]
15. Jacob AJ, Abdelkarim O, Zook S, et al. AI-based, automated chamber volumetry from gated, non-contrast CT. J Cardiovasc Comput Tomogr 2023;17:336-40. [Crossref] [PubMed]
16. Guo FQ, Wu BL, Liu XW, et al. Three-Tesla magnetic resonance imaging of left ventricular volume and function in comparison with computed tomography and echocardiography. Medicine (Baltimore) 2023;102:e33549. [Crossref] [PubMed]
17. Hahn RT, Lerakis S, Delgado V, et al. Multimodality Imaging of Right Heart Function: JACC Scientific Statement. J Am Coll Cardiol 2023;81:1954-73. [Crossref] [PubMed]
18. Medvedofsky D, Mor-Avi V, Kruse E, et al. Quantification of Right Ventricular Size and Function from Contrast-Enhanced Three-Dimensional Echocardiographic Images. J Am Soc Echocardiogr 2017;30:1193-202. [Crossref] [PubMed]
19. Addetia K, Muraru D, Badano LP, et al. New Directions in Right Ventricular Assessment Using 3-Dimensional Echocardiography. JAMA Cardiol 2019;4:936-44. [Crossref] [PubMed]
20. Li Y, Zhang L, Gao Y, et al. Comprehensive Assessment of Right Ventricular Function by Three-Dimensional Speckle-Tracking Echocardiography: Comparisons with Cardiac Magnetic Resonance Imaging. J Am Soc Echocardiogr 2021;34:472-82. [Crossref] [PubMed]
21. Greupner J, Zimmermann E, Grohmann A, et al. Head-to-head comparison of left ventricular function assessment with 64-row computed tomography, biplane left cineventriculography, and both 2- and 3-dimensional transthoracic echocardiography: comparison with magnetic resonance imaging as the reference standard. J Am Coll Cardiol 2012;59:1897-907. [Crossref] [PubMed]
22. Stolzmann P, Scheffel H, Trindade PT, et al. Left ventricular and left atrial dimensions and volumes: comparison between dual-source CT and echocardiography. Invest Radiol 2008;43:284-9. [Crossref] [PubMed]
23. Kanagala P, Arnold JR, Singh A, et al. Characterizing heart failure with preserved and reduced ejection fraction: An imaging and plasma biomarker approach. PLoS One 2020;15:e0232280. [Crossref] [PubMed]
24. Konstam MA, Kiernan MS, Bernstein D, et al. Evaluation and Management of Right-Sided Heart Failure: A Scientific Statement From the American Heart Association. Circulation 2018;137:e578-622. [Crossref] [PubMed]
25. Campbell P, Rutten FH, Lee MM, et al. Heart failure with preserved ejection fraction: everything the clinician needs to know. Lancet 2024;403:1083-92. [Crossref] [PubMed]
26. Reddy YNV, Obokata M, Verbrugge FH, et al. Atrial Dysfunction in Patients With Heart Failure With Preserved Ejection Fraction and Atrial Fibrillation. J Am Coll Cardiol 2020;76:1051-64. [Crossref] [PubMed]
27. Obokata M, Reddy YNV, Borlaug BA. Diastolic Dysfunction and Heart Failure With Preserved Ejection Fraction: Understanding Mechanisms by Using Noninvasive Methods. JACC Cardiovasc Imaging 2020;13:245-57. [Crossref] [PubMed]
28. Andersson C, Nayor M, Tsao CW, et al. Framingham Heart Study: JACC Focus Seminar, 1/8. J Am Coll Cardiol 2021;77:2680-92. [Crossref] [PubMed]
29. Blaha MJ, DeFilippis AP. Multi-Ethnic Study of Atherosclerosis (MESA): JACC Focus Seminar 5/8. J Am Coll Cardiol 2021;77:3195-216. [Crossref] [PubMed]
30. Han D, Shanbhag A, Miller RJ, et al. AI-Derived Left Ventricular Mass From Noncontrast Cardiac CT: Correlation With Contrast CT Angiography and CMR. JACC Adv 2024;3:101249. [Crossref] [PubMed]
31. Rajpurkar P, Lungren MP. The Current and Future State of AI Interpretation of Medical Images. N Engl J Med 2023;388:1981-90. [Crossref] [PubMed]
32. Larson DB, Doo FX, Allen B Jr, et al. Proceedings From the 2022 ACR-RSNA Workshop on Safety, Effectiveness, Reliability, and Transparency in AI. J Am Coll Radiol 2024;21:1119-1129. [Crossref] [PubMed]
33. Hanneman K, Playford D, Dey D, et al. Value Creation Through Artificial Intelligence and Cardiovascular Imaging: A Scientific Statement From the American Heart Association. Circulation 2024;149:e296-311. [Crossref] [PubMed]
34. Williams MC, Weir-McCall JR, Baldassarre LA, et al. Artificial Intelligence and Machine Learning for Cardiovascular Computed Tomography (CCT): A White Paper of the Society of Cardiovascular Computed Tomography (SCCT) J Cardiovasc Comput Tomogr 2024;18:519-32. [Crossref] [PubMed]
35. Hijazi W, Shanbhag A, Miller RJH, et al. Deep Learning-Derived Cardiac Chamber Volumes and Mass From PET/CT Attenuation Scans: Associations With Myocardial Flow Reserve and Heart Failure. Circ Cardiovasc Imaging 2025;18:e018188. [Crossref] [PubMed]
36. Chaudhari GR, Chillakuru YR, Chen TL, et al. Clinical language search algorithm from free-text: facilitating appropriate imaging. BMC Med Imaging 2022;22:18. [Crossref] [PubMed]
37. Armoundas AA, Narayan SM, Arnett DK, et al. Use of Artificial Intelligence in Improving Outcomes in Heart Disease: A Scientific Statement From the American Heart Association. Circulation 2024;149:e1028-50. [Crossref] [PubMed]
38. Mastrodicasa D, van Assen M, Huisman M, et al. Use of AI in Cardiac CT and MRI: A Scientific Statement from the ESCR, EuSoMII, NASCI, SCCT, SCMR, SIIM, and RSNA. Radiology 2025;314:e240516. [Crossref] [PubMed]