Prognostication and clinical opportunities with AI for coronary artery calcium: a scoping review

May combine two claims. The quote only partly states this claim.

Risk of cardiovascular death increased with CAC burden: CAC 101-400, OR 1.72; CAC >400, OR 2.62.

“Risk increased with CAC burden, for example, CVD for CAC 101–400, OR 1.72; CAC >400, OR 2.62.32”

• cites [32] Stemmer, 2020, Using machine learning algorithms to review computed tomography scans and assess risk for cardiovascular disease: Retrospective analysis from the National Lung Screening Trial (NLST), PLoS ONE sourceSupported

May combine two claims.

AI-CAC models generally relied on traditional categorical CAC (e.g., Agatston score risk categories of 0, 1-99, 100-399, ≥400) along with clinical features as core inputs, although several approaches extended quantification to individual coronary vessels or segments.

“AI-CAC models generally relied on traditional categorical CAC (eg, Agatston score risk categories of 0, 1–99, 100–399, ≥400) along with clinical features as core inputs, although several approaches extended quantification to the level of individual coronary vessels or segments to capture regional burden and distribution.3334”

• cites [33] Dekker, 2020, Automated calcium scores collected during myocardial perfusion imaging improve identification of obstructive coronary artery disease, IJC Heart & Vasculature sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [34] Zhou, 2025, Automated proximal coronary artery calcium identification using artificial intelligence: advancing cardiovascular risk assessment, European Heart Journal - Cardiovascular Imaging sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

AI-CAC was predictive of haemodynamically significant CAD (CT-FFR ≤0.8), postoperative complications, and cancer therapy-related cardiac dysfunction.

“AI-CAC was predictive of haemodynamically significant CAD (CT-Fractional Flow Reserve ≤0.8),28 postoperative complications,21 and cancer therapy-related cardiac dysfunction.24”

• cites [28] Hu, 2024, Deep learning-based coronary artery calcium score to predict coronary artery disease in type 2 diabetes mellitus, Heliyon sourceSupported
• cites [21] Marjanski, 2024, Can Artificial Intelligence Help Us in the Evaluation of Coronary Artery Calcification Scores by Acting as a Prognosticator in Patients That Are Operated on Due to Non-Small Cell Lung Cancer? A Pivotal Study, J Clin Med sourcePartly supported

  Citation 21 is attached to the postoperative-complications finding. The Results (page unavailable) report only a trend for automated cardiac age predicting cardiovascular and pulmonary complications in patients undergoing NSCLC surgery (p = 0.066); the Discussion calls it a potential prognosticator. Saying AI-CAC was predictive overstates that tentative finding. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [24] Shen, 2023, Cardiovascular Risk Stratification by Automatic Coronary Artery Calcium Scoring on Pretreatment Chest Computed Tomography in Diffuse Large B-Cell Lymphoma Receiving Anthracycline-Based Chemotherapy: A Multicenter Study, Circ Cardiovasc Imaging sourceSupported

Classified as probably not a substantial claim. May combine two claims.

AI implementation for real-world impact depends not only on algorithm accuracy but also on designing, testing, monitoring and adapting systems to ensure proper reporting, follow-up and outcomes.

“AI implementation for real-world impact depends not only on algorithm accuracy but also on designing, testing, monitoring and adapting systems to ensure proper reporting, follow-up and outcomes.17”

• cites [17] Jain, 2025, Moving Beyond the Model: Our Perspective on Meaningful AI Research in Cardiovascular Care, J Am Coll Cardiol sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The Agatston score is a validated composite metric of CAC density and volume derived from non-contrast CT and is the most widely used method for CAC quantification.

“The Agatston score, a validated composite metric of CAC density and volume derived from non-contrast CT, is the most widely used method for CAC quantification.2”

• cites [2] sourcePartly supported

  The Abstract describes a CT calcium score based on lesion area and peak Hounsfield number, and reports interobserver agreement and associations with clinical coronary artery disease. It does not specify that the scans were non-contrast or establish that this is the most widely used CAC quantification method. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

Using existing imaging avoids barriers such as additional radiation, costs or patient visits and can identify subclinical atherosclerosis across more diverse populations.

“Using existing imaging avoids barriers such as additional radiation, costs or patient visits and can identify subclinical atherosclerosis across more diverse populations.8”

• cites [8] Peng, 2024, Sociodemographic Differences Among Patients Receiving Coronary Artery Calcium Imaging vs Nongated Chest Computed Tomography Imaging, JACC Adv sourcePartly supported

  The paper’s conclusion (page not provided) says that quantifying coronary artery calcium on routine nongated chest CTs could identify more diverse populations at higher cardiovascular risk. Its introduction notes that dedicated calcium scans are not usually covered by insurance. The paper does not address whether reusing scans avoids additional radiation, costs, or patient visits, so those asserted benefits are not fully supported. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

AI algorithms can automate CAC detection and accurately quantify calcium burden from non-gated CT scans.

“AI algorithms can automate CAC detection and accurately quantify calcium burden,1213 potentially accelerating interpretation and improving reproducibility from non-gated CT scans.”

• cites [12] Wang, 2024, Meta-analysis of deep learning approaches for automated coronary artery calcium scoring: Performance and clinical utility AI in CAC scoring: A meta-analysis, Comput Biol Med sourceNot checked — no full text of the source
• cites [13] Gennari, 2024, Artificial intelligence in coronary artery calcium score: rationale, different approaches, and outcomes, Int J Cardiovasc Imaging sourceSupported

May combine two claims.

Clinical use of CAC scoring remains low due to limited clinician awareness, inconsistent insurance coverage, and concerns regarding radiation exposure and incidental findings.

“Despite strong evidence and clear recommendations, clinical use of CAC scoring remains low due to limited clinician awareness, inconsistent insurance coverage, and concerns regarding radiation exposure and incidental findings.5”

• cites [5] Foraker, 2025, 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 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

AI models often process input data in ways that no human interpreter can easily explain, but interpretability is crucial for physicians and patients to trust and use the outputs in clinical decision-making.

“In terms of transparency, AI models often process input data in ways that no human interpreter can easily explain, but interpretability is crucial for physicians and patients to trust and use the outputs in clinical decision-making.62”

• cites [62] Marey, 2024, Applications and implementation of generative artificial intelligence in cardiovascular imaging with a focus on ethical and legal considerations: what cardiovascular imagers need to know!, BJR|Artificial Intelligence sourcePartly supported

  The ‘Considerations around transparency and explainability’ section says deep-learning models can be difficult to explain and that interpretability is crucial for physicians to trust and use generative AI in clinical practice. The ‘Evaluation metrics’ section says transparency helps build patients’ trust, but does not say patients need interpretability to use AI outputs in clinical decision-making. No page numbers are available in our copy. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Borrowing from industry, target product profiles may also guide AI-CAC development toward clinically useful, guideline-aligned and financially viable implementations.

“Borrowing from industry, target product profiles may also guide AI-CAC development toward clinically useful, guideline-aligned and financially viable implementations.67”

• cites [67] Ibnidris, 2024, Facilitating the use of the target product profile in academic research: a systematic review, J Transl Med sourcePartly supported

  The Introduction describes TPPs as an industry planning tool that guides product development, and the Discussion links them to clinical utility, regulatory considerations, marketability and reimbursement. The paper does not specifically address AI-CAC or alignment with clinical guidelines, so applying those points to AI-CAC is an extrapolation. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

May combine two claims.

Using established frameworks for AI maturity in cardiovascular care, the majority of studies remain at stage 1 (Initial Development and Validation), with only limited progress to stage 2 (Implementation in Real World Settings).

“Using established frameworks for AI maturity in cardiovascular care,17 the majority of studies remain at stage 1 (Initial Development and Validation), with only limited progress to stage 2 (Implementation in Real World Settings).”

• cites [17] Jain, 2025, Moving Beyond the Model: Our Perspective on Meaningful AI Research in Cardiovascular Care, J Am Coll Cardiol sourceSupported

May combine two claims.

AI-CAC models that generate Agatston scores and guideline-aligned CAC categories integrate naturally with existing ASCVD prevention frameworks, supporting threshold-based decisions regarding statin prescription or intensification.

“AI-CAC models that generate Agatston scores and guideline-aligned CAC categories integrate naturally with existing ASCVD prevention frameworks, supporting threshold-based decisions regarding statin prescription or intensification.53”

• cites [53] Maron, 2024, Coronary Artery Calcium Staging to Guide Preventive Interventions: A Proposal and Call to Action, JACC Adv sourcePartly supported

  In “Coronary Artery Calcium Staging to Guide Preventive Interventions” (page not provided), the proposed staging table links CAC score ranges to statin recommendations based on ACC/AHA guidance. The paper does not address AI-CAC models or establish that their outputs integrate with these frameworks; that part of the claim is an inference beyond the source. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

Notification of patients and clinicians based on AI-CAC results increases cardiovascular preventive discussions, statin prescription rates and statin persistence.

“Notification of patients and clinicians based on these results has been shown to increase cardiovascular preventive discussions, statin prescription rates and statin persistence.40–43”

• cites [40] Ladak, 2026, Large Language Model Agentic Workflows for Automated Opportunistic Screening of Incidental Coronary Artery Calcium on Chest CT Reports, Can Assoc Radiol J sourceNot checked — no full text of the source
• cites [41] sourcePartly supported

  The Results (Primary Outcome; page not provided) report more documented statin discussions or new prescriptions and a higher six-month statin prescription rate in the notification arm. The Discussion (Limitations; page not provided) says persistence of statin therapy beyond six months still needs to be studied, so this paper does not establish increased statin persistence. (LLM, conf=0.91)

  Experimental — accuracy not yet measured
• cites [42] Dudum, 2026, Effects of Real-Time Notification of AI-Detected Incidental Coronary Artery Calcium on Statin Prescription: The NOTIFY-PICTURE Trial, Circulation sourcePartly supported

  The NOTIFY-PICTURE results report higher statin prescription rates after notification (44% versus 9.8% with usual care), with the increase seen among patients without ASCVD but not those with ASCVD. The paper does not report an increase in cardiovascular preventive discussions, and its limitations state that statin adherence was not assessed; it therefore does not establish increased statin persistence. The supplied copy has no page numbers. (LLM, conf=0.90)

  Experimental — accuracy not yet measured
• cites [43] Bouladian, 2025, Statin Therapy Persistence Following Opportunistic Screening for Coronary Artery Calcium on Nongated Chest CTs: NOTIFY-EXTEND, JACC Cardiovasc Imaging sourcePartly supported

  The NOTIFY-EXTEND introduction reports that statin initiation was 51% after notification versus 7% with usual care, supporting an increase in statin prescribing. Its results report documented statin discussions and high persistence at 18 months, but do not establish that notification increased preventive discussions or statin persistence relative to a comparator. No page numbers are provided in our copy. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Ethical considerations surrounding AI-CAC include bias, privacy and consent, transparency and explainability.

“Ethical considerations surrounding AI-CAC include bias, privacy and consent, transparency and explainability.59”

• cites [59] Lewin, 2024, Ethical Challenges and Opportunities in Applying Artificial Intelligence to Cardiovascular Medicine, Can J Cardiol sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Approaches such as federated machine learning may improve inclusivity while preserving data privacy across institutions.

“Approaches such as federated machine learning may improve inclusivity while preserving data privacy across institutions.60”

• cites [60] Wolff, 2022, Federated machine learning for a facilitated implementation of Artificial Intelligence in healthcare – a proof of concept study for the prediction of coronary artery calcification scores, J Integr Bioinform sourcePartly supported

  The Introduction says federated learning lets institutions train a joint model without sharing raw patient data, and the Conclusion and outlook says it could enable a more comprehensive patient collective through privacy-preserving access across institutions (page numbers unavailable). This supports broader participation in model training, but the paper does not establish improved inclusivity of patient populations or healthcare access. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

One framework referenced the PRIME checklist for cardiovascular imaging AI as an emerging reporting standard, reflecting increasing awareness of the need for standardised evaluation and transparent reporting.

“One framework referenced the Proposed Requirements for Cardiovascular Imaging–Related Machine Learning Evaluation (PRIME) checklist for cardiovascular imaging AI as an emerging reporting standard, reflecting increasing awareness of the need for standardised evaluation and transparent reporting.47”

• cites [47] Miller, 2024, AI-Defined Cardiac Anatomy Improves Risk Stratification of Hybrid Perfusion Imaging, JACC Cardiovasc Imaging sourceSupported

May combine two claims.

Patient and clinician notifications based on AI-CAC results improved statin prescription and downstream lipid testing, statin persistence, LDL-C reduction and CAD evaluation.

“Patient and clinician notifications based on AI-CAC results improved statin prescription and downstream lipid testing, statin persistence, low-density lipoprotein cholesterol (LDL-C) reduction and CAD evaluation.41–43”

• cites [41] sourcePartly supported

  The Results (Primary Outcome and Secondary Clinical Outcomes; page unavailable) report more statin prescriptions and lipid testing with notification, and lower LDL-C among patients who had repeat lipid testing. The Abstract Results report more CAD testing. The Limitations state that statin persistence beyond six months was not yet known, so this paper does not support the persistence part of the claim. (LLM, conf=0.91)

  Experimental — accuracy not yet measured
• cites [42] Dudum, 2026, Effects of Real-Time Notification of AI-Detected Incidental Coronary Artery Calcium on Statin Prescription: The NOTIFY-PICTURE Trial, Circulation sourcePartly supported

  The NOTIFY-PICTURE results (page unavailable) report more statin prescriptions with notification than with usual care, particularly among patients without baseline ASCVD. The limitations state that statin adherence was not assessed and the trial lacked power to detect LDL-C changes; the paper does not report improved downstream lipid testing or CAD evaluation and says healthcare utilization did not increase. (LLM, conf=0.91)

  Experimental — accuracy not yet measured
• cites [43] Bouladian, 2025, Statin Therapy Persistence Following Opportunistic Screening for Coronary Artery Calcium on Nongated Chest CTs: NOTIFY-EXTEND, JACC Cardiovasc Imaging sourcePartly supported

  The NOTIFY-EXTEND text (page unavailable) reports higher statin initiation with notification in NOTIFY-1, high statin persistence among those who started therapy, lower LDL-C among patients retested, and more CAC scans and stress tests after notification. It says 97 patients underwent repeat lipid testing, but does not show that notification increased lipid testing or improved statin persistence relative to a comparison group. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

The literature suggests that AI-CAC has reached advanced stages of technology readiness, reflecting substantial technical development and validation, but is still in relatively early stages of translational maturity with limited implementation and downstream clinical impact.

“Overall, the literature suggests that AI-CAC has reached advanced stages of technology readiness,18 reflecting substantial technical development and validation, but is still in relatively early stages of translational maturity with limited implementation and downstream clinical impact.”

• cites [18] Mastrodicasa, 2025, Use of AI in Cardiac CT and MRI: A Scientific Statement from the ESCR, Radiology sourceSupported

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

AI-CAC could support evolution toward a more informative, physiologically grounded CAC score, preserving clinical familiarity and, combined with radiomics features or additional image-derived markers such as epicardial adipose tissue, cardiac chamber volumetry, as well as complementary biomarkers such as thoracic aortic calcification and body composition markers if these are shown to be predictive

“AI-CAC could support evolution toward a more informative, physiologically grounded CAC score, preserving clinical familiarity and, combined with radiomics features or additional image-derived markers such as epicardial adipose tissue, cardiac chamber volumetry, as well as complementary biomarkers such as thoracic aortic calcification and body composition markers if these are shown to be predictive”

• cites [55] Naghavi, 2024, Artificial intelligence applied to coronary artery calcium scans (AI-CAC) significantly improves cardiovascular events prediction, Npj Digit Med sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [56] sourcePartly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [57] sourcePartly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

AI-CAC enables scalable assessment of vessel-specific calcium burden and plaque density and composition.

“In this context, AI-CAC enables scalable assessment of vessel-specific calcium burden and plaque density and composition.55”

• cites [55] Naghavi, 2024, Artificial intelligence applied to coronary artery calcium scans (AI-CAC) significantly improves cardiovascular events prediction, Npj Digit Med sourcePartly supported

  The Methods section (page unavailable) reports plaque location, density, number of plaques, and number of vessels affected. It does not report vessel-specific calcium burden or plaque composition: the study used calcified-plaque characteristics only, and its MESA-1 plaque reports were prepared manually, so scalable assessment of those features is not established. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

AI methods may detect subthreshold coronary calcification not captured by conventional Agatston scoring, as in the Agatston 2.0 framework, which uses a relative-density approach with internal calibration to extend assessment below the conventional 130-HU threshold, potentially refining risk stratification within the zero CAC category.

“AI methods may also detect subthreshold coronary calcification not captured by conventional Agatston scoring, as in the Agatston 2.0 framework, which uses a relative-density approach with internal calibration to extend assessment below the conventional 130-HU threshold, potentially refining risk stratification within the zero CAC category.58”

• cites [58] Naghavi M , Azimi A , Atlas K , et al . Agatston-2.0: A Next-Generation AI-Based Coronary Calcium Quantification Approach to Improve Risk Stratification Among Individuals with Zero Agatston Scores – Part I. Am J Prev Cardiol 2026;101698. doi:10.1016/j.ajpc.2026.101698 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

The review identified 695 total records across three databases and three expert recommendations.

“We identified 695 total records across three databases and three expert recommendations.”

May combine two claims.

38 studies met the inclusion criteria and were included in the final review; all were published between 2020 and 2025.

“A total of 38 studies met the inclusion criteria and were included in the final review; all were published between 2020 and 2025.”

May combine two claims.

Across heterogeneous imaging modalities and clinical contexts, included studies generally reported associations between AI-CAC and outcomes including obstructive CAD, MACE and mortality, although effect sizes and study populations varied.

“Across heterogeneous imaging modalities and clinical contexts, studies generally reported associations between AI-CAC and outcomes including obstructive coronary artery disease (CAD), MACE and mortality, although effect sizes and study populations varied.”

May combine two claims.

Most evidence on AI-CAC remains retrospective, with limited evaluation of implementation, cost-effectiveness, equity and long-term patient and health system impact.

“However, most evidence remains retrospective with limited evaluation of cost-effectiveness, equity or long-term patient and health system impact.”

• Also stated in: § WHAT THIS STUDY ADDS · plain

  “However, most evidence is retrospective, with limited data on implementation, cost-effectiveness, equity and downstream clinical impact.”

May combine two claims.

Integration of AI-CAC with CTAC scans associated with SPECT or PET CT myocardial perfusion imaging improved the detection of obstructive or haemodynamically significant CAD, reducing false-negative results and enhancing diagnostic accuracy.

“Integration of AI-CAC with CTAC scans associated with single-photon emission CT (SPECT) or PET CT myocardial perfusion imaging improved the detection of obstructive or haemodynamically significant CAD, reducing false-negative results and enhancing diagnostic accuracy.”

May combine two claims.

AI-CAC studies demonstrated rapid processing time, scalability for retrospective application of models to previously acquired CT scans, and ease of integration into radiology reporting systems.

“These studies demonstrated rapid processing time, scalability for retrospective application of models to previously acquired CT scans, and ease of integration into radiology reporting systems.”

May combine two claims.

All included studies reported institutional review board or ethics committee approval, with some discussion of bias, fairness, and model transparency.

“All studies reported institutional review board or ethics committee approval, with some discussion of bias, fairness, and model transparency.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI-CAC can be integrated with existing imaging protocols (non-ECG-gated chest CT, attenuation correction CT, low-dose lung cancer screening CT) to enable CAC quantification without new imaging and avoid additional radiation.

“Integration with existing imaging protocols | Non-ECG-gated chest CTAttenuation correction CT (eg, with PET or SPECT)Low-dose lung cancer screening CTs | Enable CAC quantification without new imagingAvoid additional radiation”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Auto-generated patient notification by large language model and increased statin prescription rates for incidental CAC enable automated identification of opportunities for preventive therapy.

“Auto-generated patient notification by large language modelIncreased statin prescription rates for incidental CAC | Automated identification of opportunities for preventive therapy”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Auto-generated CAC reports can be transferred to PACS; commercial and FDA-approved software options exist.

“Auto-generated CAC reports transferred to PACSCommercial and FDA-approved software options”

May combine two claims.

AI-CAC provides significant reductions in time/labour compared with manual scoring, maintains excellent reproducibility, and integrates seamlessly into PACS or structured reporting systems.

“Significant reductions in time/labour compared with manual scoringMaintaining excellent reproducibilitySeamless integration into PACS or structured reporting systems”

May combine two claims.

In the NOTIFY-1 and NOTIFY-PICTURE trials, AI-CAC was integrated using a hybrid workflow combining automated detection and EHR screening with partial manual validation, with potential for further automation at scale.

“In the NOTIFY-1 and NOTIFY-PICTURE trials, AI-CAC was integrated using a hybrid workflow combining automated detection and EHR screening with partial manual validation, with potential for further automation at scale.4142”

• cites [4142]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Across studies, AI-CAC demonstrated strong and consistent associations with all-cause mortality, cardiovascular mortality, MACE and obstructive CAD, including in surgical and oncology cohorts.

“Across studies, AI-CAC demonstrated strong and consistent associations with all-cause mortality, cardiovascular mortality, MACE and obstructive CAD, including in surgical and oncology cohorts.”

May combine two claims.

Although most studies cited time and labour savings compared with manual CAC scoring, none performed formal cost-benefit analyses.

“Although most studies cited time and labour savings compared with manual CAC scoring, none performed formal cost–benefit analyses.”

May combine two claims.

AI-CAC has potential for large-scale or retrospective screening, can be deployed on intrahospital servers, and incorporated into routine workflows.

“Large scale or retrospective screening potentialDeployed on intrahospital serversIncorporated into routine workflows”

May combine two claims.

No included studies were prospective, limiting clinical translation.

“although no included studies were prospective, limiting clinical translation.”

Classified as probably not a substantial claim. May combine two claims.

Key implementation challenges include technological infrastructure and interoperability between image archiving and electronic health record systems; decisions regarding embedded versus cloud-based processing; and automated versus clinician-triggered workflows.

“Key implementation challenges include technological infrastructure and interoperability between image archiving and electronic health record systems; decisions regarding embedded versus cloud-based processing; and automated versus clinician-triggered workflows.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

These studies established that deep learning models can reliably quantify CAC and stratify cardiovascular risk across heterogeneous imaging protocols and patient populations.

“These studies established that deep learning models can reliably quantify CAC and stratify cardiovascular risk across heterogeneous imaging protocols and patient populations.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Equally important are questions of responsibility and actionability (ie, who owns AI-CAC findings, how they are communicated and what clinical actions they prompt), particularly when applied opportunistically across broad imaging populations.

“Equally important are questions of responsibility and actionability (ie, who owns AI-CAC findings, how they are communicated and what clinical actions they prompt), particularly when applied opportunistically across broad imaging populations.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

AI-CAC processing takes 1-2 min per patient on standard CPU/GPU.

“1–2 min per patient on standard CPU/GPU”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

More recently, prospective applications have emerged, enabling opportunistic screening from chest CTs performed for various indications.

“More recently, prospective applications have emerged, enabling opportunistic screening from chest CTs performed for various indications.”

May combine two claims.

Four studies reported that fully automated pipelines could reduce radiologist labour time and improve throughput compared with manual scoring, supporting scalability for opportunistic screening.

“Four reported that fully automated pipelines could reduce radiologist labour time and improve throughput compared with manual scoring, supporting scalability for opportunistic screening.20252851”

• cites 20252851Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

AI-CAC models are being used to identify and engage patients and clinicians in preventive therapy opportunities.

“AI-CAC models are being used to identify and engage patients and clinicians in preventive therapy opportunities.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

AI-CAC enables full automation and batch processing, automatic vessel labelling and Agatston categorisation.

“Full automation and batch processingAutomatic vessel labelling and Agatston categorisation”

Classified as probably not a substantial claim.

Broader implementation gaps including infrastructure requirements, human-in-the-loop versus automated workflows, stakeholder engagement and capacity for follow-up care at scale remain unaddressed.

“Broader implementation gaps including infrastructure requirements, human-in-the-loop versus automated workflows, stakeholder engagement and capacity for follow-up care at scale remain unaddressed and represent important directions for future investigation.”

May combine two claims.

The studies reviewed here are early in the AI translational pathway, with only a few moving beyond AI model output to actual clinical reporting and care pathways, and none addressing longer-term model performance monitoring and outcomes.

“The studies reviewed here are early in the AI translational pathway, with only a few moving beyond AI model output to actual clinical reporting and care pathways, and none addressing longer-term model performance monitoring and outcomes (figure 3).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Nested Knowledge's AI-assisted relevance ranking was used to prioritise screening order, but all retrieved records were manually reviewed and no study was excluded solely on the basis of algorithmic ranking.

“Nested Knowledge’s AI-assisted relevance ranking was used to prioritise the order of title and abstract screening (https://nested-knowledge.com/), but all retrieved records were manually reviewed and no study was excluded solely on the basis of algorithmic ranking.”

Probably not a substantial claim: may not pass S3 (load-bearing).

This scoping review followed the PRISMA-ScR framework.

“This scoping review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) framework.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A single reviewer screened all titles and abstracts and made all full-text decisions.

“A single reviewer screened all titles and abstracts and full-text decisions were made by the same reviewer (VJFM).”

May combine two claims.

CAC scoring is an established biomarker of atherosclerotic plaque burden and a powerful predictor of future cardiovascular events.

“Coronary artery calcium (CAC) scoring is an established biomarker of atherosclerotic plaque burden and a powerful predictor of future cardiovascular events.1”

• cites [1] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Data were charted using a structured 25-variable template with an AI-assisted strategy and independent review of a 10% sample of the data.

“Data were charted using a structured 25-variable template using an AI-assisted strategy with independent review of a 10% sample of the data.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Single-reviewer screening is an additional limitation; a dual-reviewer design would have strengthened methodological rigour.

“Single-reviewer screening is an additional limitation; a dual-reviewer design would have strengthened methodological rigour.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Tracking these changes across imaging encounters may allow development of individualised cardiovascular risk trajectories, shifting opportunistic CAC assessment from a static risk marker toward a dynamic tool for monitoring broader cardiovascular disease risk rather than ASCVD alone.

“Tracking these changes across imaging encounters may allow development of individualised cardiovascular risk trajectories, shifting opportunistic CAC assessment from a static risk marker toward a dynamic tool for monitoring broader cardiovascular disease risk rather than ASCVD alone.”

May combine two claims.

Most included studies were retrospective and heterogeneous in their study design, demographics, imaging protocols and outcomes, which limits direct comparison or pooled synthesis of results.

“Most included studies were retrospective and heterogeneous in their study design, demographics, imaging protocols and outcomes, which limits direct comparison or pooled synthesis of results.”

Probably not a substantial claim: may not pass S4 (based).

Because many patients undergo repeated thoracic imaging over time, AI-derived CAC measurements could enable serial assessment of CAC progression alongside related imaging features such as plaque distribution, myocardial structure or cardiometabolic markers.

“Because many patients undergo repeated thoracic imaging over time, AI-derived CAC measurements could enable serial assessment of CAC progression alongside related imaging features such as plaque distribution, myocardial structure or cardiometabolic markers.”

• AssumptionRests on an assumption: Many patients undergo repeated thoracic imaging over time.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI-CAC enables high-throughput, reproducible analyses with potential application to millions of chest CTs performed annually in USA.

“High-throughput, reproducible analysesPotential application to millions of chest CTs performed annually in USA”

• AssumptionRests on an assumption: Millions of patients undergo chest CT in the USA annually.

Probably not a substantial claim: may not pass S2 (checkable).

Applying these principles to AI-CAC shifts the focus from algorithmic accuracy alone toward long-term clinical impact, accountability and integration within learning health systems.

“Applying these principles to AI-CAC shifts the focus from algorithmic accuracy alone toward long-term clinical impact, accountability and integration within learning health systems.”

Classified as probably not a substantial claim.

AI-CAC creates the potential for longitudinal cardiovascular phenotyping.

“Beyond single-time point assessment, AI-CAC also creates the potential for longitudinal cardiovascular phenotyping.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

By integrating multidimensional features with longitudinal assessment, AI-CAC could support more personalised preventive strategies, including individualised LDL-C targets, tailored pharmacologic intensity and precision medicine approaches such as polypill deployment.

“By integrating multidimensional features with longitudinal assessment, AI-CAC could support more personalised preventive strategies, including individualised LDL-C targets, tailored pharmacologic intensity and precision medicine approaches such as polypill deployment.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

This progression reflects an ongoing translational arc from technical performance into prospective real-world validation and implementation science.

“This progression reflects an ongoing translational arc from technical performance into prospective real-world validation and implementation science.”

May combine two claims.

Some included studies were based on overlapping source populations (eg, the National Lung Screening Trial), and thus the observed consistency of findings across studies should not be interpreted as reflecting entirely independent datasets.

“Some included studies were based on overlapping source populations (eg, the National Lung Screening Trial3251), and thus the observed consistency of findings across studies should not be interpreted as reflecting entirely independent datasets.”

• cites [3251]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable).

Developing and validating such enhanced CAC constructs will likely require evaluation approaches that extend beyond traditional randomised clinical trials, incorporating real-world data, adaptive learning frameworks and longitudinal monitoring to assess clinical utility and downstream outcomes.

“Developing and validating such enhanced CAC constructs, however, will likely require evaluation approaches that extend beyond traditional randomised clinical trials, incorporating real-world data, adaptive learning frameworks and longitudinal monitoring to assess clinical utility and downstream outcomes.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Given the inherent opacity of many AI models and the risk of automation bias, interpretability methods such as Shapley additive explanations may help build clinician and patient trust in AI-CAC outputs.

“Given the inherent opacity of many AI models and the risk of automation bias, interpretability methods such as Shapley additive explanations may help build clinician and patient trust in AI-CAC outputs.6364”

• cites [6364]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

Such approaches remain hypothesis-generating pending prospective validation.

“Such approaches remain hypothesis-generating pending prospective validation.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

In oncology care, automatic CAC detection may inform cardioprotective therapy planning and radiotherapy optimisation.

“In oncology care, automatic CAC detection may inform cardioprotective therapy planning and radiotherapy optimisation.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Frameworks such as realised net benefit and the APLUS usefulness assessment could be applied to quantify preventive benefit relative to workflow and resource constraints.

“Frameworks such as realised net benefit and the APLUS usefulness assessment could be applied to quantify preventive benefit relative to workflow and resource constraints.6566”

• cites [6566]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

CMS reimbursement pathways now exist for AI-detected CAC, which may further accelerate the economic case for adoption.

“Notably, CMS reimbursement pathways now exist for AI-detected CAC,69 which may further accelerate the economic case for adoption.”

• cites [69] Network ML . MLN matters: articles on medicare coverage, billing & payment. April 2026 update of the hospital outpatient prospective payment system (OPPS). 2026. Available: https://www.cms.gov/files/document/mm14380-hospital-outpatient-prospective-payment-system-april-2026-update.pdf [Accessed 1 Apr 2026].Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Guidance from the Stanford HEAL AI framework suggests that patient disclosure and consent to AI should depend on both the risk of physical harm and the degree of patient agency afforded by the tool.

“Guidance from the Stanford HEAL AI framework suggests that patient disclosure and consent to AI should depend on both the risk of physical harm and the degree of patient agency afforded by the tool.61”

• cites [61] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

A practical AI-CAC integration pathway may involve sequential steps: automated triggering of the AI-CAC model by the imaging or records system; structured result reporting; clinician and patient notification with contextual risk framing; linkage to guideline-based preventive care recommendations and follow-up tracking.

“A practical AI-CAC integration pathway may involve sequential steps: automated triggering of the AI-CAC model by the imaging or records system; structured result reporting; clinician and patient notification with contextual risk framing; linkage to guideline-based preventive care recommendations and follow-up tracking.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

AI-CAC could enable scalable, opportunistic cardiovascular risk stratification and expand access to preventive care without additional imaging.

“AI-CAC could enable scalable, opportunistic cardiovascular risk stratification and expand access to preventive care without additional imaging.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The development of AI-CAC mirrors that of many clinical AI tools, with early development dominated by retrospective validation focused on accuracy and reproducibility.

“The development of AI-CAC mirrors that of many clinical AI tools, with early development dominated by retrospective validation focused on accuracy and reproducibility.”

Probably not a substantial claim: may not pass S2 (checkable).

Rather than categorising patients into discrete risk bins, composite AI-derived risk trajectories may better reflect cardiovascular risk as a continuum.

“Rather than categorising patients into discrete risk bins, composite AI-derived risk trajectories may better reflect cardiovascular risk as a continuum.”

May combine two claims.

Three included studies were identified through coauthor expert recommendation because they were not yet indexed at the time of search; although all met prespecified eligibility criteria, this approach may introduce selection bias.

“Three included studies were identified through coauthor expert recommendation because they were not yet indexed at the time of search; although all met prespecified eligibility criteria, this approach may introduce selection bias.”

Probably not a substantial claim: may not pass S2 (checkable).

Sustained implementation will require capacity for follow-up care, including preventive cardiology services and coordinated pathways to translate AI-identified risk into meaningful intervention.

“Finally, sustained implementation will require capacity for follow-up care, including preventive cardiology services and coordinated pathways to translate AI-identified risk into meaningful intervention.”

Classified as probably not a substantial claim. May combine two claims.

This review was a scoping rather than a systematic review, mapping existing literature on AI-CAC models for prognostication and clinical value rather than formally assessing study quality.

“This review was a scoping rather than a systematic review, mapping existing literature on AI-CAC models for prognostication and clinical value rather than formally assessing study quality.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Recent AHA recommendations emphasise four guiding principles: alignment with organisational strategic priorities, ethical design and use, demonstrable usefulness and effectiveness and sustainable financial performance.

“Recent AHA recommendations emphasise four guiding principles: alignment with organisational strategic priorities, ethical design and use, demonstrable usefulness and effectiveness and sustainable financial performance.68”

• cites [68] Jain, 2025, Pragmatic Approaches to the Evaluation and Monitoring of Artificial Intelligence in Health Care: A Science Advisory From the American Heart Association, Circulation sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Advancing toward these stages will require prospective deployment within clinical environments, with attention to usability, interpretability and integration into decision-support pathways.

“Advancing toward these stages will require prospective deployment within clinical environments, with attention to usability, interpretability and integration into decision-support pathways.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Advancing AI-CAC towards routine clinical use will require greater standardisation, integration into real-world workflows and ongoing monitoring to demonstrate not only predictive accuracy but meaningful improvements in preventive care delivery and patient outcomes.

“Advancing AI-CAC towards routine clinical use will require greater standardisation, integration into real-world workflows and ongoing monitoring to demonstrate not only predictive accuracy but meaningful improvements in preventive care delivery and patient outcomes.”