Highlight
- Epicardial adipose tissue (EAT) undergoes adverse remodeling that reflects myocardial biology and can signal future heart failure (HF) risk.
- Automated radiomic phenotyping of EAT from routine coronary CT angiography (CCTA) enables accurate, noninvasive prediction and risk stratification of incident HF before clinical onset.
- The derived fat radiomic profile for heart failure (FRPHF) demonstrated strong discrimination and improved HF risk models beyond traditional risk factors, including coronary artery disease severity and EAT volume.
- This imaging-based biomarker has potential clinical utility for scalable screening, guiding early intervention and precision prevention of heart failure.
Study Background
Heart failure remains a major global health challenge, characterized by progressive myocardial dysfunction and considerable morbidity and mortality. Early identification of individuals at risk facilitates timely intervention to prevent disease onset or progression. Traditional risk factors and imaging markers, such as coronary artery disease severity and ventricular function, partially capture risk but lack sensitivity to the subtle myocardial-environmental changes that precede clinical heart failure.
Epicardial adipose tissue (EAT) is a metabolically active visceral fat depot contiguous with the myocardium, sharing microcirculation, and is increasingly recognized as both a sensor and modulator of myocardial biology. Changes in EAT composition mirror paracrine signaling from the heart and may reflect adverse myocardial remodeling not readily detected by standard imaging measures. Radiomics, by quantifying high-dimensional texture, shape, and volumetric features from medical images, enables extraction of biologically informative phenotypes undetectable by routine visual assessment.
This multicenter investigation tested whether radiomic characterization of EAT using widely available coronary computed tomographic angiography (CCTA) could noninvasively capture early myocardial pathology and predict future heart failure risk in a large, diverse patient population without known heart failure or prior myocardial infarction.
Study Design
This was a multicenter prospective cohort study involving 72,751 adults undergoing CCTA across nine UK centers from 2007 to 2022. Participants had no prior history of heart failure or myocardial infarction at enrollment.
An automated image analysis pipeline was developed to segment EAT from CCTA data and extract 1,655 radiomic features comprising volumetric, shape-based, and higher-order texture metrics. The study deployed a harmonized survival autoencoder model architecture to integrate these features and derive a composite fat radiomic profile predictive of incident heart failure (FRPHF).
Model development occurred in an internal cohort of 59,327 individuals across seven centers, with external validation performed in an independent cohort of 13,424 patients from two geographically distinct centers. Baseline variables including age, sex, conventional cardiovascular risk factors, coronary artery disease severity, and EAT volume were recorded.
The primary endpoint was incident heart failure during follow-up, adjudicated with a median duration of 5.1 years internally and 4.0 years externally. Survival and discrimination analyses adjusted for confounders.
Key Findings
Incident heart failure occurred in 2.9% (1,737/59,327) of the internal cohort and 2.7% (363/13,424) in the external validation cohort.
The FRPHF demonstrated robust discriminatory power for predicting future heart failure with C-statistics of 0.869 (95% CI: 0.850–0.889) internally and 0.850 (95% CI: 0.831–0.870) externally, indicating excellent predictive accuracy.
Each 25-percentile increase in the FRPHF score corresponded to a nearly fourfold higher adjusted hazard of developing heart failure (HR 3.90 internally; HR 3.79 externally, both P<0.001). Notably, patients in the highest decile of FRPHF faced an approximately 20-fold elevated risk of heart failure relative to those in the lowest decile.
In external cohorts, integrating the FRPHF into established risk models including CAD severity and EAT volume significantly improved 5-year risk discrimination (area under the curve increase of 0.047; 95% CI: 0.029–0.065) and net reclassification improvement (NRI 0.39; 95% CI: 0.29–0.48). Decision curve analysis suggested meaningful net benefit for clinical decision-making.
The predictive associations held consistently across demographic subgroups (age, sex) and across heart failure phenotypes characterized by preserved or reduced ejection fraction, supporting broad applicability.
Expert Commentary
This landmark study elegantly leverages advanced machine learning radiomic approaches applied to routinely obtained cardiac CT data to capture critical, previously unquantifiable EAT remodeling signals that presage heart failure onset. The automated pipeline’s scalability and robustness across geographically distinct cohorts highlight its potential for population-level screening.
The nearly 20-fold risk stratification gradient observed underscores the biological and clinical relevance of the EAT radiomic signature beyond traditional volume measures. This supports the emerging paradigm that visceral fat depots act as endocrine modulators of cardiac pathology.
Limitations include potential differences in CT imaging protocols across sites, though harmonization techniques were employed, and residual confounding cannot be entirely excluded. Further studies to refine the mechanistic correlates of radiomic features and prospective intervention trials guided by FRPHF risk stratification are warranted.
Conclusion
Automated radiomic phenotyping of epicardial adipose tissue using routine coronary CT angiography offers a noninvasive, precise tool for early detection and risk stratification of future heart failure before clinical symptoms manifest. Integrating FRPHF enhances predictive models beyond traditional clinical and imaging parameters.
This approach exemplifies the potential of opportunistic, imaging-based visceral fat profiling for precision cardiovascular prevention, opening avenues for earlier targeted interventions to modify disease trajectories and improve outcomes. Future research should focus on clinical implementation pathways and exploring biological mechanisms underpinning EAT radiomic remodeling in heart failure pathogenesis.
Funding and Trial Registration
Details about funding sources and trial registration were not specified in the available abstract.
References
Oikonomou EK, Chan K, Patel P, et al. Early Prediction of Heart Failure From Routine Cardiac CT Using Radiomic Phenotyping of Epicardial Fat. J Am Coll Cardiol. 2026 Apr 8;87(25):3539-3555. PMID: 41949519.
Additional relevant literature:
1. Sacks HS, Fain JN. Human epicardial adipose tissue: a review. Am Heart J. 2007;153(6):907-917.
2. Antonopoulos AS, Sanna F, Sabharwal N, et al. Detecting human coronary inflammation by imaging perivascular fat. Sci Transl Med. 2017;9(398):eaal2658.
3. Goeller M, Achenbach S, Cadet S, et al. Perivascular adipose tissue computed tomography attenuation and high-risk plaque characteristics in acute coronary syndrome. J Am Coll Cardiol. 2018;71(14):1548-1556.
